158 KiB

Raw Blame History

Contract — P13-S16: the projection gets maintained

Status: RATIFIED 2026-08-09, on the authority of the repository owner, on the evidence of the review records below — the last of which returned zero findings. DISPATCHABLE.

THE PINS ARE FROZEN. They may be executed, not edited. A defect found during execution is reported, not patched in place — if it needs a pin change, that is its own amendment with its own review round. (S27's discipline, and the reason its post-execution findings became amendments 1–7 rather than silent edits to frozen text.)

NOT YET DISPATCHED. Ratification and dispatch are separate acts; no execution instruction has been given, and no implementation work has begun.

P13-S27 landed and was accepted at 4df8e25, so pin 0's blocker is discharged: an authority now defines the implementation's current reduction semantics (epiphany_ops::CURRENT_REDUCTION_ALGORITHM_VERSION, currently 0), and core_spec.tex:11614's requirement is met — a stale canonical base is refused with CanonicalBaseRequiresRebuild on both the read and write paths.

Which independent rounds closed, and what each found, are the review records below. This block does not restate them, and states no count and no ordinal — a status line that describes the review goes stale every time the review advances. The records are the review state.

(This block read "has not been through adversarial review rounds" — true when written and false from the first independent pass onward, while sitting in the summary a top-down reader meets first. Corrected in ratification round 2, which found it as a live false signal rather than as stale history. It then promised to say RATIFIED when the decision was taken; that promise is discharged above.)

Unblocked, ratified, dispatchable and resolved are four different states. This contract is now the first three and not the fourth — nothing is implemented. Conflating the first two is how the S27 contract came to be ratified after a single round and then have that ratification withdrawn; this ratification rests on a different footing: every round was independent, every round but the last returned a blocking finding, and the last returned none.

What ratification does NOT settle, stated at the top so it is not missed

Two cells of §3's expected-outcome table are PREDICTIONS, not observations — t8d under M2 and t9 under M1 depend on how pins 4 and 1a are executed. They are stated so they can be falsified; a mismatch is a finding, in the contract or the implementation.
Every gate, test and mutation is specified and none has been run. Twelve rounds went into the claim that they can be run and that their results would be evidential. Execution is what tests that claim.
§4a's landing obligation is outstanding by construction — CLAUDE.md and the handoff carry statements pin 12's bump falsifies, and they must not be staged during execution.
The execution report is subject to independent review before completion is accepted, as S27's was. That is what turned S27's twelve-round paper record into seven post-execution amendments, and there is no reason to expect this rung to differ in kind.

This rung's first act is bumping the authority to 1 (pin 12), because it changes CreateStaffGroup's reduction verdict. That bump is the discipline S27 installed: no mechanism can detect a semantics change, so the bump is the entire guarantee.

Draft amendment 1 — 2026-08-09, on ratification reconnaissance. NOT a review round.

Fourteen findings, thirteen blocking, against the draft before its first ratification round. Nine came from the reconnaissance pass, five from a second sweep of the same defect classes. The contract is a DRAFT, so these are edits, not amendments to frozen pins — the distinction S27's status block draws, applied here.

#	Finding	Disposition
1	§2 omitted `epiphany-ops/src/lib.rs`, though the rung's defining act is bumping the authority there. The bump had no pin, no touch row, no gate and no report item — it lived only in pin 0's discharge note	Pin 12 added, with touch row 7, gate 10, report item 2b
2	Invariant 21 is a TWO-crate change, not one. `violating_score` (`generators.rs:498`) matches `GraphInvariant` exhaustively — invariant 21 does not compile without an arm — and four `all()`-driven tests then require a real generator	Touch row 8 added, with the consumer surface enumerated. This is the amendment's root cause; everything else is bookkeeping by comparison
3	S27's two tripwires fail at gate 1 with no touch row. `roundtrip.rs` test 10b `panic!`s by design once the authority moves; `serialize.rs` test 10a asserts the literal `0` and its own doc says it is expected to fail when S16 bumps	Touch rows 9 and 10, plus gate 11 requiring they be updated and not silenced
4	`requirement_labels.rs` — the escapee `CLAUDE.md` names by name — was absent, and the contract never decided whether pin 6 or pin 10 mints a label. It touches both counted documents	Pin 10a forces the decision; touch row 11 conditional; report item 2c. All three counters named, per S27's finding
5	Gates 2–3 pinned no toolchain, in a repo whose CI gates on 1.95.0 while this machine's default is 1.97.1	Both pinned
6	Gate 3 formatted two crates of four — and rows 9–10 added the other two, so it would have reported clean over unformatted code	Expanded to ops, core, textproj, testkit
7	Gate 1 had no baseline and no ignored requirement	Baseline 1577 / 0 / 0, three delta buckets, ignored MUST be 0
8	Gate 4 said "exactly §2" — the formulation S27's round 17 found unsatisfiable with a conditional row, and row 11 is now conditional	Subset both ways, with the unused row named in the report
9	M1, M2, M4 accepted "the test failed" as their whole signature	Each now names the behaviour the mutation produces. M3, M5, M6, M9 already met the standard
10	M7 covered two independently guarded doc blocks but reverted "one doc comment" — it could pass while the other guard stayed weak	Split into M7a and M7b, each quoting its own needle's non-match
11	Pin 8 identified four tests by INTERIOR line numbers (`:16838` etc.), roughly a dozen lines into each body — anchored to nothing searchable	All four named, with `fn` lines given second
12	`t6`/`t7`/`t9` locators stale	Re-derived to `:16158`, `:16231`, `:16461`, with the two-`t6`-families trap recorded
13	The bump falsifies live statements in `CLAUDE.md` and the handoff	§4a added: explicitly not staged during execution — staging them would assert the rung had landed while it awaited acceptance — and required as a post-acceptance reconciliation the report must list as outstanding
14	`shrink` also takes `GraphInvariant`; whether it matches exhaustively was not established	Report item 2d requires execution to determine it. Deliberately not guessed — guessing about an exhaustive match is what produced finding 2

Finding 2 is the one that changes the rung's shape. The others make an existing plan checkable; this one says the plan was scoped to the wrong number of files. An enum extension in a crate with a generated exhaustive consumer is never local, and nothing in the contract's own §0 inspection would have surfaced it — it appears at compile time, after execution begins.

Findings 5–8 are all defects S27 had already found and fixed in its own gate set. This contract predates those corrections and inherited none of them. A ratified contract's gate section is reusable knowledge, and the next contract drafted here should start from S27's §5 rather than from an older template.

Draft amendment 1, revision A — independent review of `d06e2f7`

Six findings, five blocking. All six are in draft amendment 1's own text, and four of them are the same mistake: importing an S27 conclusion instead of re-deriving it against this rung's facts.

#	Finding	Disposition
1	Report item 2d asked execution to decide a static fact the draft could read. `shrink` (`generators.rs:932`) does not match `GraphInvariant` — it calls `check_invariant(score, inv)`. The claim was made twice, in item 2d and touch row 8	Both corrected to state `shrink` is generic. Item 2d replaced with the real obligation: invariant 21's fixture must survive shrinking (`:1025`), since `shrink` asserts its input still violates the target
2	Pin 10a's "all three counters move if either document mints a label" is FALSE. `CORE_REQUIREMENT_COUNT` is asserted only against `core_spec.tex` (`:259`); a label in `operation_catalog.tex` moves the two suite counters only	Replaced with a per-document table. S27's "name all three" was derived for a rung touching one document; importing it here made a correction into a new error
3	Gate 11 permitted the exact tautology it exists to prevent. "Updated, not silenced" does not forbid replacing the literals with `CURRENT_REDUCTION_ALGORITHM_VERSION` — the tidiest-looking update, after which both operands move together and M5a/M5b are vacuous	Rewritten as 11a–e: independent literal `1`, never the constant, each quoted. S27's round 3 caught this substitution and `roundtrip.rs:882` forbids it by name
4	Gate 11 omitted `roundtrip.rs:947`, test 10b's mutation-only `Err` arm. Left at `0`, M5b aborts on the base comparison before reaching its two-field `panic!` — failing at the wrong assertion and observing nothing	11d added, plus 11e for the literal-preservation comments, whose reasoning is what stops the next rung making substitution 3
5	§6 demanded "the nine mutations (M1–M9)" while M7's split makes ten executions — a report could not both enumerate them and obey the tally	Count removed; §3 is the single origin
6	Pin 12 said no gate catches a missed bump except the tripwires — written in the same amendment that added gate 10, which compares the value against `HEAD` directly	Split: gate 10 guards this bump, 11a–e guard the wiring, and only the general future case stays undetectable

The tightening that came with finding 1 is the sharpest of the set. M6's two fixtures must each violate one direction only. A fixture disagreeing in both is still reported after either arm is deleted, so the mutation appears to fail correctly while signing nothing — and the same trap applies to touch row 8's generator, whose all()-driven consumers only ask whether 21 is reported.

Findings 1, 2, 3 and 6 share one root cause: an S27 conclusion applied without re-derivation. S27's "name all three counters", its "no mechanism can detect a semantics change", and its literal-independence rule are all true statements about S27. Two of them are false or incomplete here, and one was dropped exactly where it was needed. A ratified contract is reusable as a source of questions, not as a source of answers — which sharpens the note above about starting from S27's §5.

Draft amendment 1, revision B — independent review of `3096c54`

Three findings, all blocking, all in revision A's text — and each is a rule revision A had just corrected, surviving one step downstream of where it was fixed.

#	Finding	Disposition
1	§6 item 2c still said "all three counters and their new values" — the rule pin 10a corrected in the same revision. Third site of one false claim: the pin, then touch row 11, then the report item that reads the pin	Points at pin 10a's table; names which document minted the label and which counters moved
2	§6 still demanded "the nine gate results" while §4 carries eleven gates. Revision A removed the identical tally from item 1 for mutations and left its neighbour standing	Count removed, §4 named as origin, with 11a–e identified as subchecks of one gate rather than five results
3	Touch row 8 still required the generator's fixture to violate "both directions" while M6 requires direction-isolated fixtures — incompatible evidence models in one contract. A both-direction generator stays reported after either M6 arm is deleted	Row 8 now specifies one named direction plus shrink survival; pin 6/M6 own two separate isolated fixtures. `violating_score` returns one `Score` per variant and could not have carried both anyway

A third tally was found by sweeping and fixed with them: gate 7 and §6 item 4 both said "the four pin-8 tests". Correct today, and the same construction — a count restated away from its origin. Removed, pin 8's table named instead.

Draft amendment 1, revision C — independent review of `0a5b936`

Two blocking findings and one factual error. The first two are one issue: invariant 21 mandates two directions, and only one — unspecified — had durable evidence.

#	Finding	Disposition
1	Neither direction had permanent named coverage. Pin 6 asked for "a score violating only invariant 21" (singular), gate 6 asked for one, the generator carries one, and M6 observed both only while mutated. A mutation is reverted, so the restored suite could ship with one branch untested	Pin 6a added: two permanent direction-isolated tests, `m41_..._staff_names_absent_group` (S→G) and `m41b_..._group_lists_unowned_staff` (G→S), each required to satisfy the direction it does not break. Gate 6 requires both; M6 now breaks those exact tests, one each, and requires the sibling to still pass
2	"One named direction" delegated a design decision to execution. Either choice changes the generated witness and the shrink evidence, so reporting it afterward is not specifying it	Pinned to S→G in touch row 8, with the reason: smallest corruption of `valid_score`, matching every other arm's doctrine, and the exact shape pin 2's append failing produces — what M2 observes
3	§6's revision-B history said §4 has "twelve entries — 1–11 plus 4a", a false identity: `4a.` was a gate-numbered scope note with no command and no output, colliding with §4a, the landing-obligation section	The note is demoted out of the gate numbering into gate 4's body, so no gate carries an `a` suffix and `§4a` is unambiguous. (This cell stated a gate count until revision J; §4 is the origin and no count is restated anywhere.)

Revision C's finding 1 is the strongest evidence for a rule this contract already states and did not apply to itself. Pin 3a says "a mutation demonstrates the hazard once; only a test keeps it demonstrated." M6 was carrying both directions on mutation alone, three sections below that sentence. Every branch a contract mandates needs a permanent test, and a mutation is that test's signature — never its substitute.

Draft amendment 1, revision D — independent review of `25473a1`

Two blocking findings, plus one the sweep escalated. Both reported findings are the same failure in different clothes: a requirement stated with nothing able to fail it.

#	Finding	Disposition
1	Pin 10a still deferred the label decision to execution — twice reworded, never decided. The facts were readable the whole time: `core_spec.tex:6529`–`:6648` is one `requirement` box carrying the single label `req:graph:score-graph-invariants`, with exactly 20 `\item`s; invariant 21 is a 21st `\item` inside it. Pin 10 rewrites prose plus a Revision History row	DECIDED: neither document mints a label. Row 11 is UNUSED and MUST NOT be staged. No counter moves. If execution finds otherwise, that is a finding against this contract, not a decision to take at the keyboard. The counter table is retained for that case
2	Pin 6a required each fixture to violate its direction only, and nothing could observe it. The prescribed shape, `m40`, asserts `check_invariants(&s).iter().any(...)` — `any()` cannot see a second unrelated defect — and gate 6 checked the target verdict and the opposite direction, but never the absence of invariants 1–20	Each `m41`/`m41b` must assert the EXACT violation set: one violation, `StaffGroupMembershipAgreement`, witness naming the direction's staff and group ids, opposite direction asserted satisfied. Gate 6 reports the full return of `check_invariants` for both
3	(sweep) The same blind spot covers touch row 8's generator, and worse. `negative_generators_are_reasonably_targeted` bounds `kinds: BTreeSet<GraphInvariant>` at `<= 3`, but both directions of invariant 21 are the same variant — they collapse to one element, so no existing test can observe direction at all; the other three `all()` loops assert only `!is_empty()`	Row 8 now requires a dedicated permanent test that the generator violates S→G and not G→S

Finding 2 names the failure mode precisely: a requirement no assertion can fail is not a requirement. Pin 6a demanded isolation and, in the same breath, pointed at a model test that cannot check isolation. Borrowing a test's shape imports its blind spots along with its virtue — m40 was cited for its dispatch property, which is real and still applies, and nothing about invariant 20 ever turned on exactness.

Finding 1 closes the last conditional in the contract. "Decide and report" reads like rigour and is its opposite: it makes the staged set and the counter expectations depend on a choice made at the keyboard, so the touch table can be wrong in either direction and the report will agree with whatever happened. Where the facts are readable — and these were, in the .tex source — the contract decides.

Draft amendment 1, revision E — independent review of `cae1d32`

Two blocking findings and one stale rationale. Both blocking findings are in requirements revision D itself wrote, and both are its own closing lesson turned back on it.

#	Finding	Disposition
1	Row 8's new generator test had no name, so nothing consumed it. Gate 6 named only `m41`/`m41b`; §6 item 2d asks for shrink evidence. Omitting the test entirely would still compile, satisfy all four `all()` loops, and pass every named gate	Named `invariant_21_negative_generator_breaks_staff_to_group_only`, with its three required assertions spelled out; added to gate 6 and to §6 item 2e
2	Gate 6 demanded runtime evidence the prescribed tests cannot emit. It said to quote `check_invariants`' full return and witness ids — but these are `assert!` tests in `m40`'s shape, and `cargo test` prints `ok`, not local values. Satisfying it literally would need unpinned `--nocapture` instrumentation or source inference presented as observation	Evidence model chosen explicitly: quote the SOURCE assertions plus the pass verdict. A passing exact-set assertion is the observation. Follows S27's gate 6c — a quoted source construct is read, not inferred — and adds no code to `epiphany-core` written solely for a report
3	Gate 4's rationale still said row 11 is "conditional", which revision D changed to decided-unused. (A fourth site under pin 10a said "carrying it conditionally costs nothing" — found by sweep)	Both updated. The subset rule itself is unaffected; only its rationale needed the current term

Finding 1 is revision D's own lesson, unapplied to revision D. It closed by distinguishing a rule with no consumer from a rule with no observer — and then wrote a requirement with neither. An unnamed artifact cannot be gated, because every gate in this contract names what it checks.

Finding 2 is the more interesting failure: a gate that specified the right thing to know and the wrong way to know it. Exactness is the correct requirement; "quote the runtime return" was a mechanism borrowed from gates that run commands and read stdout, applied to a unit test that emits nothing on success. A gate must name evidence the prescribed artifact actually produces — otherwise execution improvises, and improvised instrumentation is unpinned scope arriving through the report.

Draft amendment 1, revision F — independent review of `abe2c35`

One blocking finding, and it is both prior lessons at once.

#	Finding	Disposition
1	The shrink leg had no observable direction or exactness guarantee. Row 8 requires the S→G fixture to survive shrinking, but the named test asserted only on the raw `violating_score(...)`; the shrunk score was checked solely by `every_invariant_shrinks_to_a_small_witness` (`:1003`), whose `!check_invariant(&small, inv).is_empty()` is membership in one variant — and both directions are the same variant, so a shrunk witness that flipped to G→S-only passes it. Because it calls `check_invariant` (singular), a shrunk witness that gained an unrelated second defect passes too. And §6 item 2d still said "quote the shrunk witness" — the unproduced-runtime-evidence defect revision E fixed in gate 6 and did not carry one hop to item 2d	The named test now asserts the same three properties twice — raw and shrunk. Gate 6 requires both legs; item 2d rewritten to revision E's source-assertion-plus-verdict model; §6 item 2e extended to both legs

This finding is the two prior lessons colliding. Revision E established that a gate must name evidence its artifact produces and fixed gate 6 — stopping one hop short of item 2d, which is the fix-propagation failure revisions A–D kept recording. And the underlying gap is revision D's: a requirement — "survives shrinking" — with nothing able to fail it.

The generalisable rule: shrink is a TRANSFORMATION, and a transformation's output needs the same guarantees asserted of its input. Requiring a fixture to "survive" a transformation establishes only that something survived. Every property the input was pinned for must be re-asserted on the output, or the transformation is free to change what the fixture proves.

Draft amendment 1, revision G — independent review of `2818ced`

One blocking finding: M6 was unexecutable against a conforming implementation.

#	Finding	Disposition
1	M6 assumed two independently removable arms; pin 6 required only behaviour. A single shared comparison — one walk emitting a violation whichever way `Staff.group` and `StaffGroup.members` disagree — satisfies `m41`, `m41b`, the generator test and gate 6, and leaves nothing for M6 to delete one at a time. Deleting the shared check disables both directions, so M6's "one fails, the sibling passes" observation cannot be produced. M6 was executable only against one implementation style	Pin 6b pins the mutation surface: two `GraphIndex` methods, `check_staff_names_absent_group` and `check_group_lists_unowned_staff`, both dispatched from `check_invariants`. M6a/M6b delete them by name, and gate 12 proves both exist and are both called before M6 is attempted

This is the unexecutable-mutation class, which S27 hit twice — its M5 and M6 both had to be rewritten after review found no runnable observation behind them. The tell is identical: a mutation phrased as an edit to a structure the pins never required. Behaviour pins constrain outcomes; a mutation deletes code. Where a mutation is the signature, the structure it deletes must itself be pinned — otherwise the contract is satisfiable in a shape that makes its own evidence unobtainable.

Pin 6b is precedent, not invention: check_invariants (invariants.rs:257–:282) already dispatches 23 check_* methods for 20 invariants, so more than one method per invariant is the crate's existing shape. And a shared helper the two methods both call is explicitly permitted — the deletable call site is what M6 needs, not a duplicated walk.

Draft amendment 1, revision H — independent review of `09d8439`

One blocking finding, and one of the same family found by sweeping.

#	Finding	Disposition
1	Gate 12's four-line aggregate could pass with one surface missing. Both definitions can exist (2 lines) while `check_invariants` calls `check_staff_names_absent_group` twice and `check_group_lists_unowned_staff` never (2 lines) — four lines, gate passes, and M6b has no call site to delete. The count proved a population, never a pairing	Replaced by four independent `grep -c` checks, each required to be exactly `1` — a count above 1 now also fails — plus quoted context: each definition with its enclosing `impl GraphIndex<'_>` header, each dispatch with the `pub fn check_invariants` header
2	(sweep) Gate 8 asserted an absence with no method. "contains the empty-members refusal and no member-liveness/`TargetMissing` path" named no command, and a `TargetMissing` path can be spelled without either literal	Method pinned: quote the production body in full and read it, explicitly not a grep. S27's gate-6a lesson; M8 signs exactly this, so a vacuous gate 8 leaves M8's deletion unobserved

Finding 1 is the count-versus-mapping failure, and it is this contract's oldest defect class wearing new clothes. Revisions A–C removed counts that had gone stale; this one was never right — an aggregate can be satisfied by the wrong distribution of the same total. Where a gate must establish a mapping, it cannot count. It has to check each element on its own, which is the structural sibling of the rule this document already carries: where a claim requires completeness, do not enumerate — derive.

Both findings are gates that report success without observing what they claim. One counted instead of pairing; the other asserted an absence with nothing able to establish it. A structural gate needs a method, and the method must distinguish the passing case from every failing one — not merely from the most obvious failing one.

Draft amendment 1, revision I — independent review of `5d2db93`

One blocking finding: gate 8's new method named a boundary that made the gate impossible to pass or honestly fail.

#	Finding	Disposition
1	Gate 8 said "quote the production body to the `#[cfg(test)]` boundary". `create_staff_group` begins at `reduce.rs:4458`; `create_part_definition` begins at `:4515` with its own `PreconditionFailureReason::TargetMissing` at `:4554`; the next `#[cfg(test)]` is at `:9576`. The literal read spans ~5,000 lines and always contains the path the gate says must be absent — while any shorter read violates the stated boundary	Gate 8 now quotes exactly pin 1's slice: `create_staff_group`'s body, brace-matched from its `fn` line to its closing brace, production source only

Pin 1 had the boundary right the whole time — "slice the create_staff_group body (brace-matched from its fn line to its closing brace, production source only)". Revision H invented a second, looser boundary instead of citing the pin. That is revision A's failure in a new place: importing a plausible-sounding rule rather than re-deriving from the source that owns it — there, an S27 conclusion; here, a boundary from a different kind of check entirely. (The #[cfg(test)] boundary is the right instrument for "is this call site production or test?", which is what §0.4 used it for. It is the wrong instrument for "where does this function end.")

The rule: where a pin already defines the artifact, the gate CITES the pin — it does not redescribe it. A redescription is a second definition, and two definitions of one artifact are a contradiction waiting for someone to read the looser one.

And note what kind of failure this was: not a gate that passes when it should fail — revision H's usual shape — but a gate with no passing state at all. It would have been discovered at execution, by an agent forced to choose between obeying the boundary and obeying the requirement, and whichever it chose would have been reported as a pass.

Draft amendment 1, revision J — independent review of `9e43994`

One blocking finding in three live consumers: adding gate 12 restored the tally defect this contract had already removed twice.

#	Finding	Disposition
1	Three sites still said §4 has "eleven gates, 1–11" after revision G added gate 12: revision C's disposition for its finding 3, gate 4's scope note, and §6 item 2 — immediately after the words "No count is stated here — §4 is the single origin."	All three counts REMOVED, not updated. The revision-C record now says only that the `4a.` scope note was demoted; gate 4's note identifies `§4a` without counting gates; §6 item 2 keeps the pointer to §4 and adds only that lettered subchecks report under their gate

§6 item 2 is the one that indicts the method. It declared §4 the single origin and restated a count in the same breath — the defect naming itself. Revision B removed "the nine gate results" from that very item; revision C's correction then wrote the then-current number into the explanation, and revision G's new gate made it false again.

The rule this makes explicit: a correction that EXPLAINS a removed count must not restate the corrected value. Say what changed, not what the number now is — otherwise the record becomes a new instance of the defect it records, and the next addition to the set falsifies the explanation instead of the original. Every count this contract has removed was re-created by the prose written to remove it.

What the sweep confirms is still sound, so the rule is not "no numbers anywhere": a count at its origin, immediately above the table that enumerates the set — pin 1a's three revised tests, pin 8's four G3a tests — is read off, not restated, and a change to the set edits the table and the adjacent word together. The defect is a count living away from the set it counts.

RATIFICATION ROUND 1 — independent, against `253249e`. One blocking finding.

This is the first round against the draft as a whole rather than against the previous revision's edits, and it found a gap ten revisions of amendment review had not: pin 5's repair had no permanent regression test.

#	Finding	Disposition
1	Pin 5 was signed by M5 alone. No named test performed `CreateStaffGroup(g, [])` → `CreateStaff(s, group: Some(g))` → undo the staff → assert the still-live `g.members` no longer holds `s`. Pin 8's four are group-undo guards — `u2tomb_a` undoes a staff but asserts "the group leaves `Score.staff_groups`", so no live group's members is ever inspected; and gate 6's `m41`/`m41b` build materialized fixtures, never running the reducer's undo path	Pin 5a adds `u5_undoing_a_staff_strips_it_from_the_live_groups_members` with three assertions — group still live, `members` lacks `s`, no invariant-21 violation. Gate 13 runs it; §6 item 2f reports it; M5 now breaks it by name and reports the changed `g.members` state rather than a condition

The finding is revision C's, on the sibling pin. Revision C established that a mutation demonstrates the hazard once; only a test keeps it demonstrated — pin 3a's own words — and applied it to pin 6 and M6. Pin 5 and M5 have the identical shape and were left alone, two paragraphs from the sentence "pin 6 is coupled to pin 5 and they split only together."

So the lesson is not "check mutations for permanent tests" — that was already learned. It is that a coupling stated in prose does not propagate a fix. When two pins are declared to stand or fall together, a correction to one is a correction owed to the other, and nothing in this document made that automatic. Every fix-propagation failure recorded in revisions A–J was a correction reaching a consumer one hop late; this one failed to reach a declared peer.

Why ten revisions missed it: each reviewed the previous revision's edits, so the question asked was always "is this change right?" — never "is anything else the same shape?" A first whole-artifact round asks the second question, which is the argument for running one before ratification rather than after.

RATIFICATION ROUND 2 — independent, whole-artifact, against `5ec2ce0`. One blocking finding.

#	Finding	Disposition
1	The live status block still said this contract "has not been through adversarial review rounds" — false from the first independent pass onward, and sitting in the "what remains before dispatch" summary, so a top-down reader received a false review-state signal before reaching any record that contradicts it	Replaced with a non-rotating pointer: the review records below are the review state, and the block states no count and no ordinal

This is the status-block defect S27 spent four amendments removing, reproduced here in its purest form. S27's amendments 4–7 established that a status line describing the review must not describe the review — it must point at the table that owns it, because every such line goes stale exactly when the review advances, which is the one moment nobody is reading the status block. The invariant S27 landed on is the model this correction follows.

What makes this one worse than the counts: a stale count reads as an error; "has not been through adversarial review rounds" reads as a verdict on the artifact's maturity. A reader deciding whether to trust this draft would have taken it as the answer, with ten revisions of independent review recorded immediately below.

And it is the whole-artifact question again. Rounds against edits ask "is this change right?"; only a whole-artifact read asks "is anything the document says about itself still true?" Round 1 found a missing test that way; round 2 found a false status claim. Both were invisible to every incremental pass, and neither was in the content those passes were correcting.

RATIFICATION ROUND 3 — independent, whole-artifact, against `e43cd34`. One blocking finding.

#	Finding	Disposition
1	Three mutations required runtime state no prescribed artifact emits. M1 wants the applied `OperationEffect` and minted members; M2 the still-empty members and invariant-21 verdict; M5 the post-undo members plus the witness. But pin 7 says only that `t8b` inverts, and gate 13 / §6 2f prescribe source assertions plus a pass verdict — a passing-test model. A mutated test stops at its first failed assertion, so state carried elsewhere is never printed, and M5's witness assertion may not execute at all	Pins 7a and 5a add pinned observation harnesses: bind the values before the assertions and format them into every relevant failure message. Gates 13 and 14 quote those diagnostics from source; M1, M2 and M5 quote the resulting failure output verbatim. A general channel rule now heads §3

The finding is revision E's, one case over. Revision E chose "quote the source assertion plus the pass verdict" — correct for a passing test, which emits nothing else. Mutations need the failing case, and the failing case has exactly one channel: the diagnostic of the assertion that fired. Settling the first did not settle the second, and the contract carried a passing-test evidence model into requirements that only failing tests can satisfy.

The ordering half is the part that would have survived a careless fix. Putting each observation in "its own" assertion reads as tidy and is precisely wrong: M1 trips t8b's spurious-order assertion and M2 the missing-order one, so a message carrying only its own value disarms whichever mutation trips the other. In u5 it is worse — M5 trips assertion 2, so assertion 3's witness is unreachable by construction. The state a mutation owes must be bound before the assertion that mutation trips.

The general rule now heads §3 rather than being pinned three times: where the observation is a single value under assert_eq!, the default diagnostic carries it — M3, M4 and M9 qualify, provided they compare the verdict rather than assert!(matches!(…)), which prints nothing, and the report must say which form each uses. Where it is composite or behind an unreachable assertion, a harness is required.

RATIFICATION ROUND 4 — independent, whole-artifact, against `3e0a4d3`. One blocking finding.

#	Finding	Disposition
1	Pin 7a collapsed two fixtures into one binding. `t8b` runs two authoring orders (§0.5) — two reductions over two different groups — but pin 7a bound "the resulting `StaffGroup.members`" and one filtered invariant result. A single `members` local satisfies the harness while leaving either mutation's observation absent, depending on which order it was taken from; the invariant result filtered from the wrong reduction is the wrong verdict, not a missing one	Pin 7a now binds four: spurious-order effect, spurious-order members, missing-order members, missing-order invariant-21 violations — every assertion formatting all four. Gate 14 must establish which order each binding came from, not count bindings. M1 and M2 name their own order at every mention

Round 3 fixed the channel and left the fixtures conflated. It correctly established that a failing test has one output channel and that state must be bound before the assertion a mutation trips — then described the state as though t8b had one reduction. The ordering rule was right and the inventory was wrong, which is why the harness looked complete: three plausible bindings, none of them wrong on its face, and no way to tell from the pin that two of them are per-order.

"The resulting value" is not a value when there are two reductions. That is the generalisable form, and it is the singular-noun cousin of the count defects: a definite article asserts uniqueness exactly as silently as a number asserts a total. Both read as precise and both hide the question of which one.

And gate 14 inherited gate 12's defect before gate 12's fix could reach it. Gate 12 was rewritten in revision H because a four-line aggregate proved a population and not a pairing; gate 14 was written in round 3 requiring "three bindings" — the same shape, one section over, three rounds later. It now requires the mapping. (A fix reaching a peer is what ratification round 1 recorded; this is a fix failing to reach a gate written after it.)

RATIFICATION ROUND 5 — independent, whole-artifact, against `e48bb8d`. One blocking finding.

#	Finding	Disposition
1	M6 was not covered by round 3's own channel rule. Pin 6a requires an exact violation set without saying how, so a conforming `assert!(violations.len() == 1 && …)` satisfies it and prints nothing on failure — and that assertion is precisely what M6 trips. Gate 6 and §6 2e use the passing-test model. So M6 reverts to "the named test failed", the signature this contract rejects	Pin 6a gains a harness for `m41`, `m41b` and the generator test: bind `check_invariants` before any assertion; make the cardinality check an `assert_eq!` on that local with the violations vector in its diagnostic; make it the first assertion. Gate 6 confirms the macro. M6a/M6b quote the failing output, the sibling's pass verdict, and the generator test's outcome

Round 3 wrote the channel rule and applied it to M1, M2 and M5 — the mutations named in that round's finding — and not to M6, which has the identical need. Round 3's own text even carved M6 out, listing it among the mutations that "already met the standard": true of its behavioural observation, which names the state, and false of its channel, which had none.

The macro is part of the channel. assert_eq! prints left/right; assert! prints only its message. Under M6a the S→G arm is gone, check_invariants returns empty for m41's fixture, and assert_eq! prints 0 against 1 with an empty vector — that is "the violation went unreported", quoted rather than inferred. With assert! the same requirement is met with no evidence. Where a mutation's observation is the compared value, the pin must name the macro — a rule now recorded at the head of §3 rather than left to each mutation.

A second gap closed with it: M6a breaks the generator test too, since that fixture is S→G. Unstated, an executor either reports a defect that is not one or ignores a failure it cannot classify. M6b leaving the generator green is now positive evidence that touch row 8's pinned direction holds. A mutation must state every test it is expected to break — otherwise its blast radius is discovered instead of specified.

RATIFICATION ROUND 6 — independent, whole-artifact, against `85429d6`. One blocking finding.

#	Finding	Disposition
1	M6a's blast radius was understated. `check_invariant` is `check_invariants(…).filter(…)` (`invariants.rs:313`–`:318`), so deleting the S→G dispatcher call silences that violation for every consumer. Beyond `m41` and the generator test, it breaks the four `all()`-driven tests on their invariant-21 iteration — `every_invariant_has_a_negative_generator`, `negative_generators_are_reasonably_targeted`, and, inside `shrink`'s entry assertion, `every_invariant_shrinks_to_a_small_witness` and `shrink_is_idempotent`	M6a now lists six expected failures in a table, each with where it fails and why, flagging that the last two panic in `shrink` (`:933`) rather than in any test body. M6b's single failure is recorded as positive evidence of row 8's pinned direction. And §3 now requires every mutation to derive its complete radius from the recorded consumer surfaces

Round 5 added the blast-radius rule and then enumerated the radius from memory. Touch row 8 already lists those four consumers with these exact line numbers — draft amendment 1 put them there as the two-crate root cause — and M6a's outcome was written without consulting the row that owns them. The document knew; the mutation did not ask.

That is a new variety of the propagation failure. Revisions A–J recorded corrections failing to reach a consumer; round 1, a declared peer; round 4, a gate written later. This one is a fact recorded in one section not reaching a requirement written in another — nothing was stale, nothing was contradicted, and no sweep for restated text would find it. It needed the question "what does this section already know that this one should be using?"

The shrink-entry detail is the part most likely to be mishandled. Two of the six panic with "shrink starting point must violate the target invariant", raised in a function the mutation never edited. Four unannounced failures, two of them pointing at untouched code, is the shape an executor resolves by concluding the contract is wrong — and then improvising. A radius discovered rather than specified invites exactly what the gates exist to prevent.

The rule became an obligation on every mutation rather than a list for M6 — and ratification round 7 found that insufficient: an obligation to derive is still the discover-the-radius model. §3 now carries the contract's own expected-outcome table.

RATIFICATION ROUND 7 — independent, whole-artifact, against `80be47c`. One blocking finding.

#	Finding	Disposition
1	Round 6's blast-radius rule contradicted itself. It required every mutation to state every expected failing test, then left M1, M2 and M5 for execution to derive — the discover-the-radius model it claimed to eliminate, with the expected set depending on the executor's reading. And the cited surfaces were insufficient: M2 removing the append breaks `t8c`, which asserts `g.members == [s]`, and pin 3a was not among the named surfaces	§3 now carries the contract's own per-mutation expected-outcome table — failing tests, notable required survivors, structural-gate outcomes — with the dependency surfaces cited, now including pins 3a, 4a, 5a, 6a and 7. Execution re-derives it against the staged tree and reports any mismatch, in either direction, as a finding

"Execution derives it" was the defect, not the remedy. A mutation's expected outcome is a claim about the contract's own design, so the contract has to make it — and execution cannot author the table and then certify its own result against it. Round 6 correctly diagnosed that a discovered radius invites improvisation, then prescribed discovery for three of the nine mutations. The obligation to derive is the same model with the responsibility moved.

t8c is the proof the surfaces were wrong, not just incomplete. Round 6 listed touch row 8, pin 1a and pin 8 — three surfaces, none of which owns t8c — while pin 3a, four sections away, defines a test asserting the exact value pin 2 maintains. Any derivation from the stated surfaces would have missed it, so the instruction was unfollowable as well as misassigned.

What the table does and does not claim. Cells carrying a reason — t8d under M2, t9 under M1 — are derivations from pins not yet executed, and are stated so they can be falsified. That is the division of labour the finding asks for: the contract predicts, execution measures, and a mismatch is a finding against whichever is wrong. It also avoids the stale-count class entirely — the table has no totals, only named tests.

RATIFICATION ROUND 8 — independent, whole-artifact, against `2a445d1`. One blocking finding.

#	Finding	Disposition
1	The expected-outcome table named required survivors without requiring them to be run. §6 consumed mutation output generally and no cell — `t8c`, `t8d`, `u5`, pin 8's four, the structural gates — and only M6 required its sibling's verdict. "Report any mismatch" cannot detect a survivor that was never run: an unrun test yields no mismatch and no evidence, so every survivor cell was an unobserved claim and the table was advisory while reading as evidential	One uniform rule now governs every mutation: full `cargo test --workspace` with the complete observed failure set reported, the named failing tests' output, each named survivor's pass verdict by name, and each named structural gate's output. An omitted named artifact is itself a finding. §6 item 1 consumes it; M6's per-half list is marked as the general rule applied rather than a special case

This is revision D's class — a requirement nothing can fail — reproduced by the table written to end the discovery model. Round 7 moved authorship of the expected outcome from execution to the contract, which was right, and left the outcome unobserved. A claim the contract owns is not thereby a claim anything checks.

The finding also forced a distinction I had left implicit. The survivors column said "notable", which quietly meant not exhaustive — so a uniform "run every named survivor" rule needed to say what is exhaustive. Now stated: the MUST-fail column is exhaustive, verified by the full-suite run; the survivors column is illustrative, naming the ones a reader would doubt, and everything outside the failing column must survive whether named or not. Without requirement 1's full-suite run, the failing column's completeness was unverified too — so the fix closes a second hole the finding did not name.

The recurring shape, now three rounds running: round 6 wrote a rule and exempted three mutations; round 7 wrote a table and left it unrun; round 8 makes the table evidential. Each fix was correct about what to specify and incomplete about who observes it — which is the same axis revision D identified and the reason the channel rules exist at all.

RATIFICATION ROUND 9 — independent, whole-artifact, against `942f261`. One blocking finding, and one the sweep escalated.

#	Finding	Disposition
1	The expected-outcome table recombined M7a and M7b — "its own grep guard; the other guard". That recreates the exact ambiguity M7's split existed to remove: a report can say "own guard failed" without establishing which guard failed or which survived, and the full-suite output round 8 now requires identifies tests by name	Two rows, with the exact names: M7a fails `t14_staff_group_field_doc_comment_states_sole_authority` (`graph.rs:2136`) while `t14_staff_group_members_…_projection` (`:2160`) must pass; M7b the reverse. Each must quote its needle-miss message and the `doc_block` it dumps
2	(sweep) M7b's required output was unobtainable. Both guards locate their slice by a phrase from the doc text, and `:2163`'s anchor is the disposition-B claim pin 10 rewrites. Once updated to the A wording, M7b's revert makes `.find()` return `None` and `.expect` panics — naming no needle and dumping no block	Pin 10 now requires the slice to be located from the field declaration — `pub members: Vec<StaffId>,`, which no wording changes — extending backwards over the contiguous `///` lines

Finding 1 is a split undone by a summary. M7 was divided in draft amendment 1 precisely because one mutation covering two guards could pass with the other guard still weak. The table then re-merged them for brevity, and round 8's full-suite rule made the cost concrete: the observed failure set names tests, so a row naming none cannot be checked against it. A summary row is a restatement, and a restatement of a distinction can drop it.

Finding 2 is the absence-grep lesson one level down. Pin 10 hardened the needles against wording both dispositions satisfy — and left the anchors matching wording only one disposition has. A guard whose locator depends on the text it inspects fails before it can report, and failing early looks like failing correctly: the test is red, the mutation "worked", and the required evidence never existed. Harden the locator, not only the assertion.

RATIFICATION ROUND 11 — independent, against `63a2dc8`. One blocking finding, fixed by the owner.

Pin 6's m40 locators had drifted: the doc/body span cited invariants.rs:6045, which is the enclosing module's doc comment, not m40's. Corrected at 25b4925 to :6060–:6135 — m40's own doc plus body — and gate 6's citation to :6063, the fn line. Verified against the tree: all three resolve, the span is m40's and not the module's, no :6045 reference remains, and the commit touched one file with no implementation change.

A locator-only finding is a different kind from rounds 1–10, every one of which found a defect in what the contract required. This one found a defect in how it points — the class the execution-time re-derivation obligations exist to absorb.

RATIFICATION ROUND 12 — independent, whole-artifact, against `25b4925`. ZERO FINDINGS.

Rechecked the live mutation-site locators, M7's field anchors, M6's structural mapping, touch-table coverage, the gate and report consumers, and the expected-outcome table. The revised m40 citations resolve and the remaining named mutation sites resolve against the tree.

What a clean round does and does not establish. It is the criterion S27's round 11 named and this contract inherited: treat dispatchable as a claim requiring evidence of convergence, not a status reached by running out of findings. Twelve independent rounds, eleven of them blocking, and the twelfth clean is that evidence — and it is the argument for having run whole-artifact rounds before ratification rather than treating ratification as the first whole-artifact read.

It is not proof of correctness. Round 12 reviewed the artifact as a whole; it did not run anything. Every gate, test and mutation here remains specified and unexecuted, and S27's history is the relevant precedent: a clean paper round, then seven post-execution amendments once the gates were actually run. The value of the clean round is that the document no longer contradicts itself — not that its predictions are right.

The defect record, stated plainly so ratification is not read as vindication: draft amendment 1 and revisions A–J closed 19 findings before the first ratification round; rounds 1–11 closed 13 more. Not one was in the pins' substance — the maintenance rule, the refusal, invariant 21, the undo strip and the authority bump have been stable since draft amendment 1. Every finding was in the evidence apparatus: what observes a requirement, what channel carries an observation, who owns a claim, and whether a locator resolves. That is where this contract was weak, and it is where execution should be read hardest.

The pattern across revisions A–J is sharper than any individual finding: a correction propagates one hop and stops. Rev A fixed pin 10a and left touch row 11; the sweep caught row 11 and stopped before §6's consumer; rev D found pin 10a's decision still deferred after two reworders. Rev A removed item 1's mutation tally and left item 2's gate tally on the next line. The fix-every-site rule is not satisfied by fixing the site and its obvious neighbour — it requires asking who reads the corrected rule, and correcting them too.

And rev D adds its converse: ask what observes each requirement. Findings 2 and 3 were both requirements with no assertion able to fail them — invisible to every sweep that looks for restated text, because nothing was restated. A rule with no consumer goes stale; a rule with no observer was never enforced at all.

The original status, retained: DRAFT — BLOCKED on P13-S27. Not executable as written. Pin 0 exposes that no authority defines the implementation's current reduction semantics, and prose saying old canonical bases "must be rebuilt" does not make them unusable. core_spec.tex:11614's requirement stays unmet until S27 supplies a version authority and a rejection-or-rebuild path.

Rung type: canonical reduction-semantics change. This is stronger than "behaviour change" and the first draft of this contract understated it. The same operation set now reduces to a different canonical Score: a graph that reduces cleanly today is refused, and a field that was never written is written. Disposition A of spec/CONTRACT_GENESIS_G3A_ENTITIES.md §1.1, ratified there as "the later maintenance/enforcement fix" and sequenced after G3b — which has landed.

No schema major or minor moves and no wire byte changes — but the Operation Catalog's version does move (pin 10), and the reduction-semantics question is separate from the schema/wire version question. See pin 0.

No policy ruling is owed. Disposition A is already ratified. What follows is its implementation.

§0. What was verified before drafting

Read out of the working tree at f876836, not recalled. Every line number below was confirmed by reading the line.

0.1 The refusal needs no new machinery — and this is the rung's biggest saving

Disposition A says CreateStaffGroup "MUST carry members: []; a non-empty members is refused or normalized away at construction." An earlier reading of this contract assumed refusal required a new PreconditionFailureReason — which would have made S16 a schema-minor epoch event with a footprint in binary_format.tex, PLAN_GMINOR_SCHEMA_MINOR.md, decode.rs, and the history test. It does not.

reduce.rs:1236 already defines the shared helper, and its own doc comment already covers this case:

The precondition no-op a structural create or delete returns when a container is non-empty where the operation requires it empty (a create carrying children, or a delete of a container with live children).

Three creates already call it for exactly this reason — create_region (:4174), create_staff_instance (:4246), create_voice (:4310) — each refusing a carried value that bears a separately-minted typed child. create_staff_group is the sole outlier: it accepts carried members and merely checks each is live (:4489–:4502).

So the refusal is the fourth instance of an established pattern, not a new one. PreconditionFailureReason::ContainerNotEmpty stays at discriminant 10. No new reason, no epoch, no wire change, no accept-set move.

0.2 Normalizing at construction or decode is excluded, on ratified grounds

Disposition A's "or normalized away" alternative must not be taken. The committed decode vector at ops/src/vectors.rs:829 pins 130 literal bytes of a CreateStaffGroup envelope whose payload carries members = [StaffId(…)], and its contract is decode-then-re-encode injectivity. The text-projection golden at textproj/src/vectors.rs:198–:201 carries the same value. Folding members away at decode would break both, and violates req:binfmt:decode-vectors (binary_format.tex:3404ff): "Canonical decode is injective: distinct byte strings denote distinct values."

Refusal happens at reduction. Decode is untouched, and both artifacts survive unchanged. This is the same fork P13-S8 faces, resolved the same way, for the same ratified reason.

0.3 The authorship cycle makes empty-at-mint the only self-consistent rule

create_staff requires a carried group to be live (:4372–:4383); create_staff_group requires each carried member to be live (:4489–:4502). There is no ModifyStaff, ModifyStaffGroup, DeleteStaff, or DeleteStaffGroup anywhere in OperationKind (payload.rs:170–:307; confirmed normatively at operation_catalog.tex:1288 and :1553). With mints only, no authoring order produces an agreeing pair.

Therefore the only authorable agreeing sequence is: mint the group empty, then mint staves naming it. Requiring members: [] is not an arbitrary restriction — it is the sole rule consistent with the operation surface that exists.

0.4 The re-carry hazard, and the base-ingest hazard hiding behind it

create_staff_group's idempotence check compares op.group against self.staff_group_values (:4466–:4469). If maintenance wrote the appended members into that map, a byte-identical re-carry of CreateStaffGroup(g, []) would compare [] against [s] and return RecreateContentMismatch instead of AlreadyApplied. Disposition A names this sub-pin explicitly.

The non-obvious half: base ingest (:1611–:1620) reseeds staff_group_values from score.staff_groups — the maintained value. So even if reduction keeps carried and derived apart in one session, a snapshot round trip launders the derived value into the carried slot, and the same misverdict returns after a reload. A test that never reloads cannot see it.

The resolution falls out of §0.1: once a non-empty carried members is refused, the carried value is by construction always empty, so the base seed can reconstruct it exactly rather than approximately (pin 4).

0.5 `t8b` is the obstacle, and it inverts rather than dies

t8b_both_permitted_stale_forms_hold (reduce.rs:16339, doc :16317–:16337) asserts both stale forms hold, and its doc block names disposition A's two maintenance rules as mutations that must make those assertions fail. The test is a correctly-built detector for precisely this change.

It is therefore rewritten, not deleted: the same two authoring orders, with the verdicts inverted — the missing order now yields agreement, the spurious order now yields ContainerNotEmpty. Its doc block's mutation notes become the rung's own mutation evidence, pointing the other way.

0.6 Invariant 21 has real work left after maintenance

Maintenance plus refusal does not make disagreement impossible:

Undo. reduce.rs:2967 removes a Staff from score.staves on undo but leaves its id in any live group's members. The reverse direction is guarded (:6736–:6744, which blocks undoing a group still named by a live staff); this direction is not.
Base ingest. A blob authored before this rung, or by another implementation, can carry a disagreeing pair straight in.

There are 20 invariants (invariants.rs:149, count guard :6064). 21 is free.

§1. Pins

PIN 0 IS DISCHARGED — read this before the pin. P13-S27 landed 2026-08-09.

Everything in pin 0 below is a dated record of the pre-S27 tree, and its three numbered requirements are superseded. Its claims that no constant names the current reduction semantics, that a conformingly-propagated stale base is accepted, that ids.rs:288's catalog claim is false, and that this rung therefore cannot execute are now false.

But one of pin 0's claims is STILL TRUE and S27 did not touch it: there is no mechanism to detect a reduction-semantics change. S27 enforces declared-version mismatches; it cannot tell that the semantics behind a version number changed. Do not read the discharge as closing that gap — see Enforcement is not detection at the end of this pin, which is why this rung's bump to 1 is mandatory.

The discharge, with each claim answered individually and the replacement requirements, is at the end of this pin. Read the pin as history; do not execute it.

Pin 0 — declare the reduction-semantics break; do not pretend it is containable.

core_spec.tex:14369–:14372 is normative: "Two replicas with different ReductionAlgorithmVersion may produce different canonical states from the same operation set; the active superblock declares the version under which the bundle's canonical base was materialized." And :11614–:11617: "Snapshots produced under an earlier algorithm version cannot be used as canonical bases under a later one without rebuilding." This rung is exactly such a change.

The obvious disposition — bump the version — is not available, and the reason is subtler than an absence of machinery. The machinery exists; it is self-referential, so it cannot detect what it appears to guard.

ReductionAlgorithmVersion (bundle/src/ids.rs:291) is a wire field of the bundle superblock (bytes 68..72, superblock.rs:20), and there is a production writer path:

reduction_version_for (bundle.rs:989) sets a new superblock's version from the canonical base's own reported version, or default() (zero) when there is no base — its doc: "only a base records a reduction."
open (bundle.rs:396–:399) rejects a bundle whose base's version disagrees with the superblock's.

So a document's declared reduction version is sourced from the base and then checked against itself. Nothing anywhere compares either value against the semantics the running implementation actually implements. A base reduced under old semantics carries its own version forward, agrees with the superblock it seeded, and is accepted. The check is not vacuous — it catches a corrupt or tampered base whose version disagrees with its superblock. What it cannot catch is a valid stale base: one whose version was conformingly propagated. So the check necessarily passes for a conformingly propagated stale base, which is exactly the case core_spec.tex:11614 exists to prevent.

(An earlier draft of this pin claimed "every construction site is an unrelated hardcoded literal." That was false, and the error is instructive: the search behind it looked for ReductionAlgorithmVersion( constructor calls, which by construction cannot find a path that propagates an existing value without constructing one. The instrument could not observe the thing it was used to rule out.)

Two supporting facts stand: there is no constant or accessor naming the implementation's current reduction semantics, and ids.rs:288–:289 states that "the algorithm catalog itself lives in epiphany-ops" while nothing of the kind exists there — a doc comment asserting a false fact about another crate, which is P13-S26's pattern, second instance.

Scope of the claim, deliberately narrowed: this inspection establishes that the current implementation has no mechanism to detect a reduction-semantics change. It does not establish that no such change in the project's history was ever detectable; that would need a history audit this rung has not done, and the stronger sentence must not be written into the ledger.

Therefore this rung does NOT invent the missing machinery, which would be a scope explosion and a separate design — and, because it does not, this rung cannot execute. Pin 0 records the break; it does not discharge it. S16 waits on S27's disposition. What pin 0 still requires of the eventual rung:

State the break in spec/PASS13_CANDIDATES.md's P13-S16 row — which stays blocked on P13-S27, not resolved — and in operation_catalog.tex's Revision History entry: canonical bases materialized before this rung must be rebuilt, not reused, because create_staff now writes StaffGroup.members and create_staff_group now refuses inputs it previously accepted.
File P13-S27 — that the reduction-version machinery is self-referential: reduction_version_for (bundle.rs:989) sources a new superblock's version from the canonical base's own self-report, and open (bundle.rs:396) checks only that the two agree, so the current implementation has no mechanism comparing either against the semantics it actually implements. Supporting: no constant or accessor names the current semantics, and ids.rs:288's catalog claim is false. The stronger historical claim — that no such change has ever been detectable — is NOT established by this inspection and must not be written. This rung is the occasion, not the cause.
Assert nothing it cannot enforce. No test may claim stale bases are rejected; nothing rejects them. The break is recorded, not guarded, and the contract says so plainly rather than implying coverage.

PIN 0 IS DISCHARGED — P13-S27 LANDED 2026-08-09 (`4df8e25`)

Everything above in pin 0 is a dated record of the pre-S27 tree and MUST NOT be executed as written. S27 built the machinery whose absence pin 0 documented, so the pin's premises, its conclusion, and all three of its numbered requirements are superseded. They are retained because the reasoning is what motivated S27, and because operation_catalog.tex's rebuild note still has to be written.

Which of pin 0's factual claims are now FALSE, stated individually so none is left standing by implication:

Pin 0 said	Now
"no constant or accessor naming the implementation's current reduction semantics"	`epiphany_ops::CURRENT_REDUCTION_ALGORITHM_VERSION` (currently `0`), plus `Bundle::capabilities()` as the accessor
"`ids.rs:288`–`:289` … asserts a false fact about another crate"	Made true by S27's pin 8 — the same doc comment now names the real location and mechanism
"a valid stale base is accepted — one whose version was conformingly propagated"	False now. A base whose declared version differs from the authority is refused with `CanonicalBaseRequiresRebuild { base, current }` on both the read path (`open`) and the write path (`commit`/`commit_versioned`)
"the current implementation has no mechanism to detect a reduction-semantics change"	STILL TRUE, and S27 did not change it. See the split immediately below — this row is the one an earlier draft of this supersession got wrong
"this rung cannot execute"	It can. This contract is UNBLOCKED — see the status block. It is still a DRAFT and needs ratification, which is a different bar

ENFORCEMENT IS NOT DETECTION — the split S27 did not close

An earlier draft of this supersession declared pin 0's "no mechanism to detect a reduction-semantics change" false because mismatches are now rejected. That was wrong, and it contradicted requirement 3 thirty lines below it. The two are different claims and only one of them moved:

	Declared-version mismatch	Semantics change
What it is	a base whose recorded `reduction_algorithm_version` differs from the running authority	the implementation's canonical reduction verdicts changing, with or without a bump
After S27	ENFORCED — refused with `CanonicalBaseRequiresRebuild` on read (`open`) and write (`commit`/`commit_versioned`)	STILL UNDETECTABLE. Nothing compares the semantics the code implements against the number it declares
What guarantees it	the check, mechanically	a human remembering to bump. There is no backstop

So S27's enforcement is conditional on the discipline, not a substitute for it. If this rung changes CreateStaffGroup's verdict and the bump is missed, every base it produces declares 0, matches an authority still reading 0, and passes every check S27 installed — the enforcement fires correctly on a number that is itself wrong. epiphany-ops's authority doc states this outright: no mechanism can detect a semantics change; the discipline is the guarantee; there is no backstop.

This is precisely why pin 0's requirement 3 inverts into a mandatory bump rather than dissolving. S16 is the first rung to change a canonical reduction verdict, so S16's bump to 1 is the human-enforced half of the guarantee, and the only half that applies to itself.

Pin 0's narrowing survives and still binds. This inspection does not establish that no reduction-semantics change in the project's history was ever detectable — S27 did not perform that history audit either, and the stronger sentence still must not be written into the ledger.

What replaces the three requirements:

State the break in operation_catalog.tex's Revision History exactly as written above — canonical bases materialized before this rung must be rebuilt, not reused. Unchanged. But the ledger half is superseded: the P13-S16 row is UNBLOCKED, not blocked on P13-S27 — see pin 11, amended the same day.
~~File P13-S27~~ — DISCHARGED. S27 has its own row, its own ratified contract, and its own RESOLVED — IMPLEMENTED 2026-08-09 (pin 10) marker. Nothing here is left to file.
Assert nothing it cannot enforce — the principle stands; its application inverts. Stale bases are now rejected, and S27 owns the tests that prove it. So this rung must not re-assert S27's guarantee, and must not add a second detection path. What it MUST do instead is bump CURRENT_REDUCTION_ALGORITHM_VERSION to 1, because it changes CreateStaffGroup's reduction verdict. No mechanism can detect a missed bump — S27's authority doc is explicit that the discipline is the entire guarantee — so the bump is this rung's obligation and nothing will catch its absence.

Line-number citations in this contract predate S27 and are only PARTLY re-derived. S27 changed bundle.rs by 795 lines, so bundle.rs:989, :396 and ids.rs:288 above — and every other bundle.rs reference in this document — remain stale as locators even where the claim about them is historical.

Draft amendment 1 re-derived the ones that name executable targets: t6 :16154→**:16158, t7 :16229→:16231, t9 :16454→:16461**, and pin 8's four interior line numbers replaced by test names. The rest are not done and finishing them is ratification work.

A trap the re-derivation found, recorded so the next pass does not fall into it: reduce.rs contains two t6/t7 families — t6_undo_restores_the_chain_... (:15577) and t6_g3a_referential_loops_... (:16158). A re-derivation that greps fn t6 and takes the first hit lands on the wrong test, and both hits are plausible in context. Match on the full _g3a_ name, never the prefix.

Pin 1 — refuse a non-empty carried members, using the existing helper. In create_staff_group (reduce.rs:4458), before the liveness loop, refuse a carried non-empty members with container_not_empty(). Match the idiom and comment style of :4174, :4246, :4310. Do not introduce a new PreconditionFailureReason; ContainerNotEmpty stays at 10 and no schema document moves.

The liveness loop (:4489–:4502) becomes unreachable for non-empty members and must be deleted, not left dead — a precondition that cannot fire is the kind of residue this pass keeps finding.

No behavioural test can prove that deletion. A retained dead loop passes M1 and every other assertion here, because refusal short-circuits before it. Pin 1 therefore requires a structural gate: read crates/epiphany-ops/src/reduce.rs, slice the create_staff_group body (brace-matched from its fn line to its closing brace, production source only — never the test module), whitespace-normalize, and assert the slice contains the empty-members refusal and does not contain any TypedObjectId::Staff liveness check or TargetMissing construction.

Pin 1a — three existing tests assert what pin 1 removes; all three must be revised. Each currently passes by asserting the outgoing rule:

Test	Line	Currently asserts	Under pin 1
`t6`	`reduce.rs:16158`	`CreateStaffGroup.members` naming a non-live target is `TargetMissing`, one of three referential loops asserted separately	that arm becomes `ContainerNotEmpty` and stops testing a referential loop at all; the `CreatePartDefinition.staves` and `CreateView.active_layers` arms are untouched and must stay
`t7`	`reduce.rs:16231`	those same preconditions are not enforced base-free	pin 1's refusal is not graph-gated — it is a property of the carried value — so this arm now refuses base-free too, inverting the claim for that arm only
`t9`	`reduce.rs:16461`	a from-empty score passes `check_invariants`, and each skipped reducer check independently makes invariant 10 fire, on a fixture that already attempts a dangling reference in each of the four ops	the dangling-member fixture can no longer enter the graph, so both the fixture and t9's own mutation set change

t7's inversion is the subtle one and must be stated in its doc block: a graph-aware precondition asks about the universe and cannot run base-free; an empty-container precondition asks only about the carried value and therefore runs everywhere. Different classes — conflating them is how a later reader would wrongly "restore" the graph gate.

Each keeps its id and gains a doc note recording what it asserted before this rung and why the assertion inverted.

Pin 2 — maintain the projection in create_staff. In create_staff (:4330), after the graph push at :4386 and only when op.staff.group == Some(g), append the new staff's id to g's members in self.graph's score.staff_groups. Under base-free reduction (self.graph.is_none()) there is no graph to maintain and nothing is written — matching how :4385 already guards the staff push itself.

Appending is set-union and order-independent per staff id; the append must be idempotent (never add an id already present), since convergence replays.

Pin 3 — carried and derived are kept apart, and staff_group_values holds the carried value. self.staff_group_values (:4507) MUST continue to store the value as authored — which pin 1 guarantees is always empty-membered. Pin 2 writes the maintained members only into self.graph. The re-carry comparator at :4466 is not changed and must keep comparing against the carried value.

Pin 3a — a permanent named regression test, not only mutation M3. Add t8c_recarry_compares_against_the_carried_members_not_the_derived: in one session, CreateStaffGroup(g, []) → CreateStaff(s, group: Some(g)) → re-carry CreateStaffGroup(g, []), asserting AlreadyApplied and that the graph's g.members == [s] at that moment. A mutation demonstrates the hazard once; only a test keeps it demonstrated.

Pin 4 — base ingest reseeds the carried value, not the derived one. At :1619, seed staff_group_values with the group's value with members emptied, not group.clone(). This is exact rather than lossy: pin 1 makes empty the only authorable carried value, so the reconstruction is the carried value. The comment at :1614–:1618 must be extended to say why, naming the reload hazard of §0.4 — otherwise a later reader "fixes" it back to group.clone() and reintroduces a defect that only appears after a snapshot.

Pin 4a — a permanent named regression test for the reload path. Add t8d_recarry_after_reduction_onto_a_materialized_base_stays_idempotent: reduce the pin-3a sequence, materialize the score, re-reduce onto that score as a base, then re-carry CreateStaffGroup(g, []) and assert AlreadyApplied. This is the only test that would exercise the base-seed path; pin 4 is unguarded without it.

Pin 5 — close the undo hole in the unguarded direction. The Staff undo retain arm (:2967) MUST also strip the removed staff's id from every group's members, in both self.graph and — if any derived copy is held — wherever else the projection lives. The StaffGroup arm (:2977) needs no change: :6736's reference guard already blocks undoing a group a live staff names.

Pin 5a — a permanent named regression test for the staff-undo strip. ADDED IN RATIFICATION ROUND 1.

Pin 5's repair was signed by M5 alone, and a mutation is reverted. Add to crates/epiphany-ops/src/reduce.rs, named exactly:

u5_undoing_a_staff_strips_it_from_the_live_groups_members

Sequence — the one the pin repairs, end to end: CreateStaffGroup(g, []) → CreateStaff(s, group: Some(g)) → undo the CreateStaff → then assert, on the materialized graph:

g is still live in Score.staff_groups — if the group is gone there is no projection left to be wrong, and the test asserts nothing;
g.members does not contain s, quoted by id;
check_invariants reports no StaffGroupMembershipAgreement violation — the exactness form pin 6a uses, so the test fails on a residue whichever direction it leaves.

OBSERVATION HARNESS — ADDED IN RATIFICATION ROUND 3, same rule as pin 7a. Bind both the post-undo g.members and the check_invariants violations (with their witness ids) to locals before assertion 2, and format both into assertion 2's and assertion 3's failure messages.

Under M5 assertion 2 is the one that fires — members still contains s — so assertion 3 never executes and its witness is never printed. M5 nevertheless owes that witness. Computing the violations only where assertion 3 needs them puts the required observation behind an assertion the mutation guarantees is unreachable. The state a mutation owes must be bound before the assertion that mutation trips.

Why the existing tests do not supply this, checked rather than assumed. Pin 8's four are group-undo guards: u2tomb_a (:17125) undoes a staff, but its assertions are that the undo tombstones the referencer, that T1's undo then proceeds, and that "the group leaves Score.staff_groups" (:17187) — it undoes the group too, so no live group's members is ever inspected. Gate 6's m41/m41b build materialized fixtures and never run the reducer's undo path. So nothing permanent exercised the sequence pin 5 exists for.

This is pin 3a's rule — a mutation demonstrates the hazard once; only a test keeps it demonstrated — and revision C applied it to pin 6/M6 and not to pin 5/M5. The contract states two paragraphs below that "pin 6 is coupled to pin 5 and they split only together"; their evidence models were nonetheless fixed one at a time. A coupling stated in prose is not a coupling applied.

Pin 6 — graph invariant 21, StaffGroupMembershipAgreement. Flags both directions: a staff whose group names a group whose members omit it, and a group listing a staff whose own group is not that group. Witnesses must name both ids and the direction.

This is an append with a documentation footprint — invariants.rs:149, all(), the count guard at :6064, and core_spec.tex's normative enumeration (which currently ends at 20 and which P13-S26 has just established is already under-describing invariant 10 — do not attempt to repair invariant 10's prose here; that is S26's rung).

The count is not the test. all().len() == 21 passes with the dispatch arm deleted — the project already knows this, which is why m40_check_invariants_dispatches_invariant_20 exists (invariants.rs:6060–:6135, whose own doc says "all().len() == 20 passes even with the dispatch arm deleted, so this row must instead show a score violating ONLY invariant 20 is actually flagged by the top-level check_invariants entry point"). Pin 6 requires the same behavioural shape: a score violating only invariant 21, flagged through check_invariants, with the dispatch-arm deletion as its signing mutation. A count assertion may accompany it and may not replace it.

Pin 6a — TWO permanent, direction-isolated tests. ADDED IN REVISION C, which found both directions mandated and only one durably covered.

Pin 6 required "a score violating only invariant 21" — singular — and gate 6 asked for one score. The generator carries one direction. M6 observes both directions, but only while mutated, and a mutation is reverted: after restoration nothing in the permanent suite need exercise the second branch. A mutation demonstrates once; only a test keeps it demonstrated — this contract's own words under pin 3a, applied to itself.

Add both, named, in the shape of m40_check_invariants_dispatches_invariant_20:

Test	Fixture violates	Must satisfy
`m41_check_invariants_dispatches_invariant_21_staff_names_absent_group`	S→G: a staff whose `group` names a group whose `members` omit it	the G→S direction, and every other invariant
`m41b_check_invariants_dispatches_invariant_21_group_lists_unowned_staff`	G→S: a group listing a staff whose own `group` is not that group	the S→G direction, and every other invariant

Each fixture MUST violate its own direction only. A fixture disagreeing both ways is reported after either arm is deleted, so it signs neither — the same isolation M6 requires, made permanent. Each test must state in its doc which direction it holds and which it breaks, so a later reader cannot "simplify" the two into one.

Each test MUST assert the EXACT violation set, not membership. ADDED IN REVISION D.

m40_check_invariants_dispatches_invariant_20 — the shape this pin points at — asserts only that the target appears: check_invariants(&s).iter().any(|v| v.invariant == …). any() cannot detect a second, unrelated defect, so a fixture carrying one satisfies m40's shape, satisfies gate 6, and satisfies M6's deletion outcome — while violating pin 6a's own "only its own direction" requirement, which nothing then observes. A requirement no assertion can fail is not a requirement.

So each of m41 / m41b MUST assert:

check_invariants(&s) returns EXACTLY one violation, and its invariant is StaffGroupMembershipAgreement — an exact-set assertion, not any(). This is what proves invariants 1–20 are satisfied, which any() never touches.
The witness names the specific staff and group ids for that direction, so the two tests cannot pass on each other's fixture.
The opposite direction is satisfied on the same score, asserted directly.

Pin 6a required isolation and then prescribed a test shape that cannot check it. The model test was cited for its dispatch property — that all().len() alone passes with the arm deleted — which is real and still applies. Borrowing a test's shape imports its blind spots along with its virtue, and m40 never needed exactness because nothing about invariant 20 turned on it.

OBSERVATION HARNESS — ADDED IN RATIFICATION ROUND 5. M6 was not covered by the round-3 channel rule.

Requirement 1 says exact set and does not say how, so a conforming assert!(violations.len() == 1 && violations[0].invariant == …) satisfies it and prints nothing on failure. Under M6 that assertion is exactly what fails — and M6's required observation is that the mutated fixture went unreported, which the diagnostic would not show. M6 would fall back to "the named test failed", the signature this contract rejects.

So in each of m41, m41b and invariant_21_negative_generator_breaks_staff_to_group_only:

Bind check_invariants(&s) to a local before any assertion.
Requirement 1's cardinality check MUST be an assert_eq! on that local's length, with the full violations vector formatted into its diagnostic, e.g. assert_eq!(violations.len(), 1, "expected exactly the invariant-21 violation, got {violations:?}").
It MUST be the first assertion after the bindings — it is the one M6 trips, and the round-3 ordering rule applies: state behind a later assertion is unreachable.

assert_eq! is required rather than suggested because its default left/right diagnostic is the observation. Under M6a the S→G arm is gone, check_invariants returns an empty set for m41's fixture, and the diagnostic prints 0 against 1 with an empty vector — that is "the violation went unreported", quoted rather than inferred. assert! on the same condition prints only the message, so the identical requirement is met with and without evidence depending on a macro choice the pin had left open.

Pin 6b — the MUTATION SURFACE is pinned, not just the behaviour. ADDED IN REVISION G.

Pin 6 and pin 6a specify behaviour only, and a conforming implementation can satisfy every one of them with a single shared comparison — one walk that compares Staff.group against StaffGroup.members and emits a violation whichever way they disagree. That implementation passes m41, m41b, the generator test and gate 6, and leaves M6 with no two arms to delete. Deleting the shared check disables both directions at once, so M6's required "one test fails, the sibling passes" observation cannot be produced at all — M6 would be executable only against one implementation style, which pin 6 never required.

So the two directions MUST be two independently removable checks, following this crate's existing idiom — GraphIndex methods writing into out:

fn check_staff_names_absent_group(&self, out: &mut Vec<InvariantViolation>)   // S→G
fn check_group_lists_unowned_staff(&self, out: &mut Vec<InvariantViolation>)  // G→S

Both emit GraphInvariant::StaffGroupMembershipAgreement violations, and check_invariants calls both, in sequence with the existing idx.check_*(&mut v) calls.

This is the crate's shape, not an invention for the mutation's convenience. check_invariants (invariants.rs:257–:282) already calls 23 check_* methods for 20 invariants, so more than one method per invariant is existing precedent. The names are PINNED, because gate 12 greps for them and M6 deletes them by name — the same reason S27 had to pin synthetic_for_fixture after discovering its gate searched for a name the contract had offered only as an example.

A shared helper the two methods both call is permitted — deduplicating the walk is fine. What is forbidden is a single call site for both directions, because the deletable unit is what M6 needs.

Pin 6 is coupled to pin 5 and they split only together. M5 signs the undo hole by requiring invariant 21 to observe the residue. If invariant 21 is deferred to a later rung, then either pin 5 and M5 defer with it, or M5 must be restated to assert the leftover member directly on the materialized score rather than through check_invariants. Splitting pin 6 out while leaving M5 as written would leave the undo repair unsigned.

Pin 7 — t8b inverts. Rewrite t8b_both_permitted_stale_forms_hold as t8b_the_projection_is_maintained_and_the_spurious_form_is_refused: same two authoring orders, verdicts inverted. Its doc block must record that it previously pinned the opposite, and why the change is the ratified disposition rather than a regression. Deleting it is forbidden — the pairing of the two orders is the coverage.

Pin 7a — t8b carries an OBSERVATION HARNESS. ADDED IN RATIFICATION ROUND 3.

M1 and M2 require t8b to report state — the applied OperationEffect and minted members for M1, the still-empty members and invariant-21 verdict for M2. A #[test] emits nothing but its assertion diagnostics, so unless the state is in those diagnostics it is unobtainable, and revision E already ruled out adding print instrumentation for a report.

t8b exercises TWO authoring orders (§0.5), which are two reductions over two different groups — so there are FOUR observations, not three. CORRECTED IN RATIFICATION ROUND 4. Bind all four to locals before any assertion:

#	Binding	The mutation that needs it
1	spurious-order `OperationEffect` — `ContainerNotEmpty` when pin 1 holds	M1: becomes an applied effect
2	spurious-order `StaffGroup.members`	M1: carries the spurious membership that reached the graph
3	missing-order `StaffGroup.members` — `[s]` when pin 2 holds	M2: stays empty
4	missing-order `check_invariants`, filtered to `StaffGroupMembershipAgreement`	M2: the disagreement the empty projection leaves

Then format ALL FOUR into EVERY assertion's failure message.

Bindings 2 and 3 are different groups in different reductions, and an earlier draft of this pin collapsed them into one "resulting StaffGroup.members". A single members local satisfies the harness as written while leaving either mutation's required observation absent — whichever order it was taken from. The invariant result has the same problem: filtered from the wrong reduction it is the wrong verdict, not a missing one. A harness for two fixtures needs two sets of bindings; "the resulting value" is not a value when there are two reductions.

The ordering rule is unchanged and still load-bearing. A failing test stops at its first failed assertion, so state carried only in a later assertion's message is never printed. Under M1 the spurious-order assertion fires; under M2 the missing-order one does — different assertions, so each must carry the whole set. Putting each value in "its own" assertion is exactly the arrangement that fails, and splitting the set per order would reintroduce it one level down.

Pin 8 — the four G3a undo-repair tests are re-verified, not assumed. NAMED by draft amendment 1. All four are in crates/epiphany-ops/src/reduce.rs:

Test	`fn` at
`u2a_a_live_staff_naming_the_group_blocks_its_undo`	`:16826`
`u2bf_a_the_staff_group_guard_holds_base_free`	`:16983`
`u2tomb_a_a_tombstoned_referencing_staff_does_not_block_the_groups_undo`	`:17125`
`u3a_minting_the_group_and_its_referencing_staff_in_one_transaction_undoes_whole`	`:17333`

This pin previously identified them ONLY as :16838, :16992, :17138, :17345 — and those are interior line numbers, roughly a dozen lines into each body, not the fn lines. An interior line number is worse than a stale one: it anchors to nothing a reader can search for, and it silently drifts with every edit to the function above it. Names are stable; line numbers are a convenience. The fn line numbers above are given second and are already subject to the locator warning under pin 0.

They each construct the missing-member form, which pin 2 now makes an agreeing form. They assert on undo effects rather than on members, so they are expected to survive — expected, not known. Each must be run and its verdict reported; any that changes is a finding.

Pin 12 — the authority bump. ADDED BY DRAFT AMENDMENT 1; it had no pin, no touch row and no gate.

Bump epiphany_ops::CURRENT_REDUCTION_ALGORITHM_VERSION 0 → 1 (touch row 7), and add this rung's entry to the constant's own Bumps list in the same doc comment. The list is the only record of why a version exists; a bump without its entry leaves a number nobody can account for.

This is not optional and not a consequence — it is the obligation. This rung changes CreateStaffGroup's reduction verdict, and core_spec.tex §"Canonical Document Identity" makes such a change require a new version.

What guards it, stated precisely — corrected on review, which found this paragraph contradicting the gate added beside it:

Gate 10 is the direct guard on THIS rung's bump. It compares the constant's value against HEAD and requires the Bumps entry. A bump omitted here is caught.
Gates 11a–e are the independent wiring guards. They confirm the production path actually reads the moved authority, using operands that do not descend from it.
What remains undetectable is the general case: a FUTURE semantics change whose bump is forgotten. No gate in any contract can catch that — see Enforcement is not detection under pin 0. The discipline is the guarantee.

The sentence that stood here said "nothing in the gate set will catch its absence except the two tripwires" — written in the same amendment that added gate 10, which catches exactly that. It imported S27's true statement about semantics changes in general and applied it to this rung's bump in particular, where it is false. The same over-generalisation the enforcement/detection split under pin 0 exists to prevent, committed one section away from it.

The bump is stated in pin 0's discharge as requirement 3, which is a discharge note and not a pin. So it had no numbered pin to execute, no touch row to stage, no gate to check and no report item to confirm — the four places this repo's discipline requires a mandated change to appear. It now has all four: pin 12, touch row 7, gate 10, report item 2b.

Pin 10a — whether pin 6 or pin 10 mints a \label{req:...} MUST be decided explicitly, and stated in the report. ADDED BY DRAFT AMENDMENT 1.

This rung edits core_spec.tex (row 5, invariant 21's enumeration) and operation_catalog.tex (row 4), and crates/epiphany-testkit/tests/requirement_labels.rs counts requirements and labels in both. The two readings have different touch tables:

DECIDED IN REVISION D: NEITHER document mints a label. Row 11 is UNUSED. No counter moves.

This pin twice said "decide and report", which is not a decision. It left the staged set and the counter expectations conditional on a choice execution would make arbitrarily — and a conditional touch row is a row that can be wrong in either direction. The facts settle it, and they were readable while the pin was being written:

Pin 6 adds invariant 21 to an enumeration that already sits inside ONE requirement box. core_spec.tex:6529 opens \begin{requirement}, :6530 carries the single label req:graph:score-graph-invariants, :6533 opens the enumerate, and the box closes at :6648 — with exactly 20 \items inside. Item 21 is an \item within that box. It mints no requirement and no label.
Pin 10 rewrites existing prose in operation_catalog.tex §CreateStaff and §CreateStaffGroup, plus a Revision History row and a version bump. No new requirement box, no new label.

Therefore: CORE_REQUIREMENT_COUNT, SUITE_REQUIREMENT_COUNT and SUITE_LABEL_COUNT all stay unchanged; touch row 11 is unused and MUST NOT be staged; and the report states that it was unused for this reason rather than re-deriving the question.

If execution finds this wrong — if either edit turns out to require a new \begin{requirement} — that is a finding and a contract defect, reported under §6 item 5, not a decision to be made at the keyboard. The counters would then move per the table below, which is retained for that case and for the next rung.

Label minted in	`CORE_REQUIREMENT_COUNT`	`SUITE_REQUIREMENT_COUNT`	`SUITE_LABEL_COUNT`
`core_spec.tex`	moves	moves	moves
`operation_catalog.tex`	unchanged	moves	moves
both	moves by 1	moves by 2	moves by 2
neither — THIS RUNG	—	—	—

S27's lesson was "name all three, not one" — for a rung that touched only core_spec.tex. Carrying that conclusion across to a rung touching two documents turned a correction into a different error, and then into a deferred decision. A fix imported from another contract must be re-derived against this one's facts — and where the facts are readable, re-derived now, not delegated to execution as a "decide and report".

CLAUDE.md names this file by name as a recurring escapee, it escaped the format-epoch rung's table, and S27 had to add it mid-execution. A file that must change but is not listed silently drops out of the commit, and the failure surfaces on someone else's branch. Carrying it as a decided-unused row costs nothing and documents the decision — revision E; it read "carrying it conditionally costs nothing if unused" until revision D removed the conditionality.

Pin 9 — no byte artifact moves. spec/vectors/decode_vectors.txt, ops/src/vectors.rs:829's literal-byte vector, and textproj/src/vectors.rs:198's golden are all unchanged, because refusal is at reduction and decode is untouched (§0.2). Confirm, do not assume. No schema major or minor, no accept-set move, and no schema/binary-format companion version bump. This is narrower than "no version bump": pin 10 does require an Operation Catalog version bump and Revision History row, because the catalog's normative text about these two operations changes.

Pin 10 — specification surfaces. operation_catalog.tex §CreateStaff and §CreateStaffGroup currently state the disposition-B stale-form semantics normatively (:1265–:1278, :1531–:1540) and must be rewritten to the maintained rule, with a Revision History row and a version bump. core_spec.tex's Staff.group / StaffGroup.members docs (:5585–:5591, :4235–:4241) and the two Rust doc comments (graph.rs:842–:847, :1643–:1649) likewise.

The two Rust doc comments are grep-asserted by graph.rs:2136 and :2160 (needles at :2139, :2142, :2163, :2166, :2148, :2172, :2176, :2180). Those guards must be updated in step, and the updated needles must assert the new rule — a guard left asserting the old words would fail loudly, but a guard weakened to a substring both rules share would pass silently, which is the worse outcome.

The guards' SLICE ANCHORS must be made wording-independent. ADDED IN RATIFICATION ROUND 9 — without this, M7b cannot produce its required output.

Both guards locate the doc block by searching for a phrase from the doc text: graph.rs:2139 finds "/// Which staff group (if any) this staff belongs to.", and :2163 finds "/// A **non-authoritative denormalized projection** of group" — which is the disposition-B claim this pin rewrites. So the anchor must move with the rewrite; and once it names the new A wording, M7b's revert to B makes .find() return None and .expect("StaffGroup.members's doc comment is present") panics — naming no needle and dumping no doc_block. The mutation's required observation would be an anchor panic pointing at absent text, not a needle miss.

So locate the slice from the FIELD DECLARATION, which no wording changes — pub group: Option<StaffGroupId>, and pub members: Vec<StaffId>, — and extend backwards over the contiguous /// lines above it. The block then exists under either disposition, and the only way to fail is the needle assertion, with its message and block intact.

This is the absence-grep lesson one level down. The needles were hardened against wording that both dispositions satisfy; the anchors were left matching wording only one disposition has. A guard whose locator depends on the text it inspects fails before it can report, and failing early looks like failing correctly.

Each new needle MUST be wording the disposition-B comments cannot satisfy. Retaining only "sole authority" and "non-authoritative" is insufficient: both phrases are true under B and under A, so a guard built from them alone passes against the text it is meant to have replaced. Require phrases that are false under B — for example "maintained from Staff.group" and "must agree" — so that reverting either comment to its B wording fails the guard. M7 signs exactly this.

Pin 11 — the ledger. AMENDED 2026-08-09: the mandated state changed when S27 landed.

This pin required the P13-S16 row to read "blocked on P13-S27". S27 landed and was accepted at 4df8e25, so that state is now false, and a pin mandating a false ledger state would put this contract in contradiction with the ledger it governs. As amended, the row must read: UNBLOCKED — S27 landed; this contract is DRAFT and needs ratification before dispatch. Everything else the pin requires recorded is unchanged.

The second paragraph's instruction to file P13-S27 in the same edit is discharged — S27 has its own row, its own ratified contract, and its own RESOLVED — IMPLEMENTED 2026-08-09 (pin 10) marker. It is no longer this rung's to file.

spec/PASS13_CANDIDATES.md's P13-S16 row → UNBLOCKED, S27 landed (was: blocked on P13-S27), recording the ContainerNotEmpty reuse (and that a new reason was considered and proved unnecessary), the base-ingest hazard, the t8b inversion, the t6/t7/t9 revisions, and — per pin 0 — that canonical bases materialized before this rung must be rebuilt rather than reused.

~~File P13-S27 in the same spec/PASS13_CANDIDATES.md edit~~ — DISCHARGED 2026-08-09; do not execute. S27 was filed, contracted, ratified, implemented and landed at 4df8e25, and its row carries RESOLVED — IMPLEMENTED 2026-08-09 (pin 10). The blocking relation this instruction existed to make visible from both ends is now a resolved relation recorded at both ends. The reasoning below is retained as the record of why S27 was filed, not as work to do.

It is a prerequisite discovered by this rung, not independent ledger cleanup, and the two rows must land together so the blocking relation is visible from either end. Its claim, at the scope §0's inspection supports: the reduction-version machinery is self-referential — reduction_version_for (bundle.rs:989) sources a new superblock's version from the canonical base's own self-report, and open (bundle.rs:396) checks only that the two agree — so the current implementation has no mechanism comparing either against the semantics it actually implements, and core_spec.tex:11614's rebuild requirement is unenforced. Supporting: no constant or accessor names the current semantics, and ids.rs:288's claim that the catalog lives in epiphany-ops is false — a second instance of P13-S26's pattern.

(Kept verbatim as the filing that produced S27, not as a description of the tree. MOST of it is now false — but NOT all, and an earlier draft of this annotation said "every claim" and was itself wrong in the way this contract keeps having to correct. False now: the machinery is no longer self-referential — open compares the base's version against the injected authority, not only against the superblock it seeded; ids.rs:288 was made true by S27's pin 8; and core_spec.tex:11614 is enforced for declared-version mismatches. STILL TRUE: "no mechanism comparing either against the semantics it actually implements" — S27 compares a declared number against a declared number. Nothing anywhere compares either against the semantics the code actually implements, and nothing can. See Enforcement is not detection under pin 0.)

Do not write the stronger historical claim ("no reduction-semantics change has ever been detectable"); §0 does not establish it.

The P13-S16 row does NOT move to RESOLVED in this edit. It records the disposition-A plan, names this contract, and is marked ~~blocked on P13-S27~~ UNBLOCKED — S27 landed 4df8e25 (corrected 2026-08-09). "Not RESOLVED" still holds and is the part that matters here: this rung has not been implemented, and unblocked, dispatchable and resolved are three different states.

§2. Touch table

#	File	Change
1	`crates/epiphany-ops/src/reduce.rs`	pins 1, 1a, 2, 3, 3a, 4, 4a, 5, 7, 8; and `create_staff_group`'s own doc comment, which states the disposition-B rule
1b	`crates/epiphany-ops/src/payload.rs`	`CreateStaffGroupOp`'s doc (`:1789`ff) states that graph-aware reduction preconditions every carried member resolves to a live `Staff` and that "the mint stores `members` exactly as given and neither maintains nor trusts it" — both clauses become false
1c	`crates/epiphany-ops/src/valuegen.rs`	`staff_group`'s doc (`:372`–`:375`). Precision: the helper still carries and preserves supplied `members` exactly, and "never normalizes" stays true and MUST be preserved — the helper is unchanged. What becomes false is the framing that such a non-empty value is "the value a `CreateStaffGroup` mints" (it can no longer mint one), and the disposition-B attribution. Rewrite those two clauses only
2	`crates/epiphany-core/src/invariants.rs`	pin 6
3	`crates/epiphany-core/src/graph.rs`	pin 10 (two doc comments + their two grep guards)
4	`spec/operation_catalog.tex` (+ `.pdf`)	pin 10
5	`spec/core_spec.tex` (+ `.pdf`)	pins 6, 10
6	`spec/PASS13_CANDIDATES.md`	pin 11
7	`crates/epiphany-ops/src/lib.rs`	ADDED by draft amendment 1. Pin 12 — bump `CURRENT_REDUCTION_ALGORITHM_VERSION` `0 → 1` and add its entry to the constant's own `Bumps` list, which is in the same doc comment. The rung mandated by pin 0's discharge had no touch row, no pin and no gate
8	`crates/epiphany-core/src/generators.rs`	ADDED by draft amendment 1 — the root cause below. `violating_score` (`:498`) matches `GraphInvariant` exhaustively, so invariant 21 does not compile without a new arm. Four `all()`-driven tests (`:991`, `:1004`, `:1025`, `:1042`) then consume it, so the arm must be a real generator, not a stub
9	`crates/epiphany-testkit/src/roundtrip.rs`	ADDED by draft amendment 1. S27 test 10b (`:894`) reopens a literal-`0` base under `production_caps()`; pin 12's bump makes that path return `Err` and hit an arm that `panic!`s by design
10	`crates/epiphany-textproj/src/serialize.rs`	ADDED by draft amendment 1. S27 test 10a (`:659`) asserts the production writer supplies `ReductionAlgorithmVersion(0)`. Its own doc (`:655`) says it is expected to fail when S16 bumps and that updating it is S16 stating the authority moved
11	`crates/epiphany-testkit/tests/requirement_labels.rs`	UNUSED — DECIDED in revision D, no longer conditional. Pin 10a establishes that neither pin 6 nor pin 10 mints a `\label{req:...}`: invariant 21 becomes an `\item` inside the existing `req:graph:score-graph-invariants` box (`core_spec.tex:6529`–`:6648`), and pin 10 rewrites prose. No counter moves; this file MUST NOT be staged, and the report says so citing pin 10a. The row is retained rather than deleted because `CLAUDE.md` names this file as a recurring escapee — a row reading "deliberately unused, and why" survives review, while an absent row looks like an oversight. (Read as "CONDITIONAL — decide and report" until revision D, and as "all three counters move" until revision A.)

Regenerate the two PDFs only after their sources reach final form.

Rows 9 and 10 are S27's tripwires firing as designed, not collateral damage

S27 wrote two assertions against the literal 0 precisely so they would move when the authority moved, and serialize.rs:655 says so in as many words. Their failure is this rung's signal that the bump took effect — the only signal it gets, since no mechanism can detect a semantics change. They must be updated, not deleted or #[ignore]d, and the report must quote both before and after.

Neither file was in any touch row, and both fail at gate 1. That is the exact shape of S27's own gminor.rs failure: a file the rung must change that no surface count reached, found by a gate rather than by the table. The allowlist catching it is the allowlist working — but only if the row exists before execution starts.

Row 8 is the rung's real shape change: an enum extension is a TWO-crate change

This contract treated invariant 21 as local to invariants.rs. It is not. GraphInvariant has an exhaustive generated-consumer surface in epiphany-core:

violating_score (generators.rs:498) matches every variant — adding one is a compile error until its arm exists.
Four all()-driven tests then call it and shrink (:991, :1004, :1025, :1042), so a todo!() or trivial arm fails them. Invariant 21 needs a fixture that violates the S→G direction — a staff whose group names a group whose members omit it — and NOT the G→S direction, and survives shrinking. The direction is PINNED here, in revision C.
The direction needs its own permanent, NAMED test — REVISION D, name pinned in REVISION E. No existing all()-driven test can observe direction. negative_generators_are_reasonably_targeted (:1037) collects kinds: BTreeSet<GraphInvariant> and allows kinds.len() <= 3, but both directions of invariant 21 are the same GraphInvariant variant, so they collapse to one element and the bound is blind to the distinction; the other three loops assert only !is_empty().

Add to generators.rs's test module, named exactly:
```
invariant_21_negative_generator_breaks_staff_to_group_only
```
It MUST assert the same three properties twice — before and after shrinking. THE SHRUNK LEG IS ADDED IN REVISION F. On violating_score(StaffGroupMembershipAgreement, seed) and again on shrink(&that, StaffGroupMembershipAgreement):

(i) check_invariants returns exactly one violation and it is StaffGroupMembershipAgreement; (ii) that violation's witness names the S→G staff and group ids; (iii) the G→S direction is satisfied, asserted directly.

The shrink leg was required by row 8 and observed by nothing. The raw fixture was checked by this test; the shrunk one only by every_invariant_shrinks_to_a_small_witness (:1003), which asserts !check_invariant(&small, inv).is_empty(). That is membership in a single GraphInvariant variant, so a shrunk witness that flipped to G→S-only passes it — both directions are the same variant — and because it calls check_invariant (singular) rather than check_invariants, a shrunk witness that gained an unrelated second defect passes too.

shrink is a transformation, so its output needs the same guarantees as its input. Requiring a fixture to "survive shrinking" without asserting what survives only establishes that something still fires.

Revision D required this test and gave it no name — so nothing consumed it. Gate 6 named only m41/m41b, and §6 item 2d asks for shrink evidence. Omitting the test entirely would still compile, satisfy all four all() loops, and pass every named gate. An unnamed obligation has no consumer, which is revision D's own closing lesson applied to the requirement revision D wrote.

"Both directions" was wrong here and incompatible with M6 — corrected in revision B. violating_score returns one Score per variant, so it cannot carry two fixtures; and a fixture disagreeing in both directions is still reported after either M6 arm is deleted, which is precisely the evidence failure M6's own isolation rule forbids.

Revision B then said "one named direction" and left WHICH to execution — a design decision disguised as a reporting requirement. Either choice changes the generated witness and the shrink evidence, so reporting it afterward does not substitute for specifying it. Pinned above to S→G, because that is the smallest corruption of valid_score (drop the staff id from group.members, leave staff.group intact), matching the doctrine every other arm follows, and it is the exact shape this rung's own failure produces — pin 2's append not firing, which is what M2 observes.

The division of labour, stated once: row 8's generator covers S→G and must survive shrink; pin 6a owns two permanent direction-isolated tests, one per direction; M6 breaks those two tests, one each. Three fixtures, three purposes, none standing in for another.
shrink (:932) takes GraphInvariant but does NOT match on it — it calls check_invariant(score, inv) and shrink_candidates, so it is generic over the invariant and needs no new arm. (Draft amendment 1 first said this "MUST be checked at execution", deferring a static fact readable from the function body. Corrected on review: a draft that can decide something must decide it, or it exports its own unfinished reading as execution work.)
What the EXISTING shrink tests impose is weak, and revision F is why the named test carries a shrunk leg. generators.rs:1025–:1026 runs shrink(&violating_score(inv, 7), inv) for every variant, and shrink asserts on entry that its input violates the target; :1003 then checks the shrunk score with !check_invariant(&small, inv).is_empty(). Together these establish only that something still fires — a fixture whose violation depends on incidental structure shrink_candidates removes fails there, which is real but is the weakest of the three properties. Direction and exactness after shrinking are guaranteed only by the named test's shrunk leg.

Consequence for the mutation plan and gate 6: invariant 21's generator is itself load-bearing, so M6 (deleting each arm) now has a second signature — the negative generator must still produce a violating score for the arm under test.

§3. Mutation plan

Applied, run, output recorded verbatim, restored by hand-editing back.

A FAILING TEST IS NOT AN OBSERVATION — draft amendment 1

M1, M2 and M4 accepted "the named test fails" as their whole signature. That signs nothing: a test fails for the mutation, for a typo, for an unrelated panic, or because the harness changed — and a compile error is not a test failure at all. The evidence a mutation owes is the BEHAVIOUR it changed, not the assertion it broke. Each mutation below must now name the wrong verdict, value or state the mutated build produces, and the report must quote it. (This is S27's round-15/16 lesson, which cost four defects there; M3, M5, M6 and M9 in this contract already met the standard, so the fix is uneven-by-design rather than uniform.)

AND THE OBSERVATION NEEDS A CHANNEL — ratification round 3

Naming the state a mutation must report does not make it obtainable. A #[test] emits only the diagnostic of the assertion it fails on, and it stops there — so the rule for every mutation in this section is:

The state a mutation owes MUST be carried by the failure diagnostic of the assertion that mutation trips.

Where the observation is a single value under assert_eq!, the default left/right diagnostic already carries it — M3, M4 and M9 are satisfied this way, provided their assertions compare the verdict itself rather than assert!(matches!(…)), which prints nothing. State in the report which form each uses.

Where the observation is composite, or lives behind an assertion the mutation makes unreachable, the test needs a pinned observation harness — bind the values before the assertions and format them into every relevant message. That is pins 7a (t8b, for M1/M2), 5a (u5, for M5) and 6a (m41/m41b and the generator test, for M6 — added in ratification round 5).

The macro is part of the channel, not a style choice. assert_eq! prints left/right; assert! prints only its message. Where a mutation's observation is the compared value, the pin must require assert_eq! — otherwise the same requirement is satisfiable with and without evidence.

THE EXPECTED-OUTCOME TABLE — the CONTRACT owns it. RATIFICATION ROUND 7.

Round 6 required every mutation to state its full radius and then left M1, M2 and M5 for execution to derive. That is the discover-the-radius model it claimed to eliminate, and it makes the expected set depend on the executor's reading. Execution cannot both author the table and certify its own result.

So the table below is this contract's CLAIM. Execution re-derives it against the staged tree and reports any mismatch — in either direction — as a finding, against the contract or the implementation. No cell is a licence to ignore an outcome.

Dependency surfaces, cited so the derivation is checkable: pin 1a (t6/t7/t9), pin 3a (t8c, which asserts g.members == [s]), pin 4a (t8d), pin 5a (u5), pin 6a (m41/m41b), pin 7 (t8b's two orders), pin 8 (the four undo-repair tests), touch row 8 (GraphInvariant's four all() consumers, generators.rs:991/:1004/:1025/:1042), and check_invariant = check_invariants(…).filter(…) (invariants.rs:313). Round 6's list omitted pins 3a and 4a, which is how t8c went missing from M2.

Mutation MUST fail Notable required survivors Structural gates

M1 remove pin 1's refusal t8b spurious-order; t6's CreateStaffGroup arm; t7's inverted arm t8c, t8d, u5 and pin 8's four — all carry members: [], so the refusal never fires for them; m41/m41b/generator and the four all() consumers — no reducer path; t9 — pin 1a changes its fixture so it no longer attempts a dangling member through this op gate 8 FAILS — the refusal is what it reads for

M2 remove pin 2's append t8b missing-order; t8c — it asserts g.members == [s] (pin 3a) u5 — after the undo there is no staff, so members-lacks-s and no-violation both still hold; t8d — pin 4 seeds from the carried value with members emptied, so its re-carry still matches; m41/m41b/generator; the four all() consumers gate 8 passes (pin 1 untouched)

M3 pin 3 writes derived members t8c — AlreadyApplied → RecreateContentMismatch t8b, u5, m41/m41b —

M4 restore group.clone() at :1619 t8d — the base-seed path is the only one that reaches it t8c — non-base path —

M5 remove pin 5's strip u5 — assertion 2, members still contains s t8b, t8c, t8d; pin 8's four — they undo the group, never inspecting a live group's members; m41/m41b — constructed fixtures —

M6a delete S→G dispatch six — see M6's own table m41b; u5 (no violation either way after the undo) gate 12 FAILS — a call site is gone

M6b delete G→S dispatch m41b only m41, the generator test, all four all() consumers — every invariant-21 fixture is S→G gate 12 FAILS

M7a revert Staff.group's doc block to B wording t14_staff_group_field_doc_comment_states_sole_authority (graph.rs:2136) t14_staff_group_members_field_doc_comment_states_non_authoritative_projection (:2160) must pass that guard's needle-miss is the gate: quote its message and the doc_block it dumps

M7b revert StaffGroup.members's doc block to B wording t14_staff_group_members_field_doc_comment_states_non_authoritative_projection (:2160) t14_staff_group_field_doc_comment_states_sole_authority (:2136) must pass same, for that guard

M8 reinstate the dead loop nothing behavioural every behavioural assertion gate 8 FAILS — its only signature

M9 graph-gate pin 1's refusal t7's inverted arm t8b spurious-order — not base-free —

Cells marked with a reason are derivations, not observations — t8d under M2 and t9 under M1 in particular depend on how pins 4 and 1a are executed. They are stated so they can be falsified. A mismatch is the finding this table exists to produce.

RUNNING THE TABLE IS PART OF EACH MUTATION — RATIFICATION ROUND 8

"Report any mismatch" cannot detect a survivor that was never run. An unrun test produces no mismatch and no evidence, so every survivor cell was an unobserved claim — the table read as evidential and was advisory. Only M6 required its sibling's verdict, and that is now the general rule rather than one mutation's special case.

Under every mutation, without exception:

Run the full cargo test --workspace and report the complete set of tests that failed. This is what makes the "MUST fail" column exhaustive rather than illustrative: any failure not in that column is a finding, and so is any listed failure that did not occur.

Report the named failing tests' output, per the channel rules above — the behaviour, not the fact of failure.

Report each named required survivor's PASS VERDICT, by name. A survivor cell is discharged by a quoted verdict, never by silence.

Evaluate each named structural gate and report its output — gate 8 under M1 and M8, gate 12 under M6a and M6b, pin 10's guards under M7a/M7b.

An omitted named artifact is itself a finding, on the same footing as a wrong outcome.

The two columns have different force, and round 8 settles it. The MUST fail column is exhaustive — verified by requirement 1's full-suite run. The survivors column is illustrative: it names the ones a reader would doubt, and everything not in the failing column is required to survive whether or not it is named. "Notable" was doing that work implicitly and left the failing column's completeness unverified.

Revision E chose "quote the source assertion plus the pass verdict" for gates, which is right for a PASSING test. Mutations need the failing case, and the failing case has exactly one channel. Getting the first right does not settle the second.

Mutation	MUST fail	Notable required survivors	Structural gates
M1 remove pin 1's refusal	`t8b` spurious-order; `t6`'s `CreateStaffGroup` arm; `t7`'s inverted arm	`t8c`, `t8d`, `u5` and pin 8's four — all carry `members: []`, so the refusal never fires for them; `m41`/`m41b`/generator and the four `all()` consumers — no reducer path; `t9` — pin 1a changes its fixture so it no longer attempts a dangling member through this op	gate 8 FAILS — the refusal is what it reads for
M2 remove pin 2's append	`t8b` missing-order; `t8c` — it asserts `g.members == [s]` (pin 3a)	`u5` — after the undo there is no staff, so members-lacks-`s` and no-violation both still hold; `t8d` — pin 4 seeds from the carried value with `members` emptied, so its re-carry still matches; `m41`/`m41b`/generator; the four `all()` consumers	gate 8 passes (pin 1 untouched)
M3 pin 3 writes derived members	`t8c` — `AlreadyApplied` → `RecreateContentMismatch`	`t8b`, `u5`, `m41`/`m41b`	—
M4 restore `group.clone()` at `:1619`	`t8d` — the base-seed path is the only one that reaches it	`t8c` — non-base path	—
M5 remove pin 5's strip	`u5` — assertion 2, `members` still contains `s`	`t8b`, `t8c`, `t8d`; pin 8's four — they undo the group, never inspecting a live group's members; `m41`/`m41b` — constructed fixtures	—
M6a delete S→G dispatch	six — see M6's own table	`m41b`; `u5` (no violation either way after the undo)	gate 12 FAILS — a call site is gone
M6b delete G→S dispatch	`m41b` only	`m41`, the generator test, all four `all()` consumers — every invariant-21 fixture is S→G	gate 12 FAILS
M7a revert `Staff.group`'s doc block to B wording	`t14_staff_group_field_doc_comment_states_sole_authority` (`graph.rs:2136`)	`t14_staff_group_members_field_doc_comment_states_non_authoritative_projection` (`:2160`) must pass	that guard's needle-miss is the gate: quote its message and the `doc_block` it dumps
M7b revert `StaffGroup.members`'s doc block to B wording	`t14_staff_group_members_field_doc_comment_states_non_authoritative_projection` (`:2160`)	`t14_staff_group_field_doc_comment_states_sole_authority` (`:2136`) must pass	same, for that guard
M8 reinstate the dead loop	nothing behavioural	every behavioural assertion	gate 8 FAILS — its only signature
M9 graph-gate pin 1's refusal	`t7`'s inverted arm	`t8b` spurious-order — not base-free	—

M1 — the refusal fires. OBSERVATION TIGHTENED, draft amendment 1; CHANNEL PINNED IN RATIFICATION ROUND 3. Remove pin 1's emptiness check. Required observation: a CreateStaffGroup carrying a non-empty members is now applied instead of refused — the spurious-order OperationEffect and the spurious-order StaffGroup.members, showing the spurious membership that reached the graph. Obtain them by quoting t8b's spurious-order assertion failure verbatim, which pin 7a's harness requires to carry all four of its bindings — so this mutation's two are present whichever assertion fires. (Read "all three observations" until round 4, when pin 7a's collapsed members binding was split per authoring order.) The assertion failing is the symptom; the applied mint is the observation.

M2 — the maintenance fires. OBSERVATION TIGHTENED, draft amendment 1; CHANNEL PINNED IN RATIFICATION ROUND 3. Remove pin 2's append. Required observation: after a CreateStaff naming a live group, the missing-order StaffGroup.members is still empty, and the missing-order invariant-21 verdict on that state. Obtain both by quoting t8b's missing-order assertion failure verbatim — the same harness, a different assertion and a different reduction from M1's, which is why pin 7a binds members and the invariant result per authoring order and requires every assertion to carry all four.

M3 — the re-carry stays idempotent. Make pin 3 write maintained members into staff_group_values; a byte-identical re-carry must degrade from AlreadyApplied to RecreateContentMismatch. This mutation must be observed, not reasoned about — it is the sub-pin disposition A named.

M4 — the base-ingest hazard is real. OBSERVATION TIGHTENED, draft amendment 1. Restore :1619 to group.clone(). Required observation: the re-carry misverdict itself — name what an ingested base's CreateStaffGroup re-carry now returns and what it should have returned, in the shape M3 already uses (AlreadyApplied → RecreateContentMismatch). "Pin 4a's test must fail" was the entire signature and does not distinguish the hazard from any other breakage. If the misverdict does NOT appear, pin 4 is unmotivated and that is a finding — report it rather than keeping a guard nothing needs.

M3 and M4 are signed by pins 3a and 4a, not by manual demonstration. Run each mutation against its named test so the guard, not the transcript, is what survives the rung.

M8 — the dead loop is actually gone. Reinstate the member-liveness loop in create_staff_group after pin 1's refusal, where it cannot fire. Every behavioural assertion in this contract must still pass, and the pin-1 structural gate must fail. This is the only signature available for a deletion that no behaviour observes.

M9 — t7's inversion is real, not assumed. With pin 1 in place, run the CreateStaffGroup arm of t7 base-free and confirm it refuses; then graph-gate pin 1's refusal behind self.graph.is_some() and confirm that arm reverts to applying. Signs that the empty-container precondition is deliberately universe-independent.

M5 — the undo hole is closed. BOUND TO A NAMED TEST IN RATIFICATION ROUND 1. Remove pin 5's strip. Required:

u5_undoing_a_staff_strips_it_from_the_live_groups_members (pin 5a) must fail, and
the changed state must be reported, not the failure — by quoting the test's own failure output, which pin 5a's harness requires to carry both values: g.members after the undo, still containing s, and the StaffGroupMembershipAgreement violation with its witness ids. Quote the panic verbatim.

(M5 previously required only that undoing a CreateStaff "leave a disagreeing pair that invariant 21 flags" — an observation with no permanent test to break, and phrased as a condition rather than a named artifact. Both halves are fixed: pin 5a supplies the test, and the observation now names the state.)

M6 — invariant 21 sees both directions. FIXTURES MUST ISOLATE, tightened on review. Delete each arm in turn; each deletion must leave a distinct disagreeing fixture unreported.

M6 deletes pin 6b's two NAMED surfaces and breaks pin 6a's two named tests, one each — bound in revision C, surface pinned in REVISION G:

M6a — delete idx.check_staff_names_absent_group(&mut v); from check_invariants → m41_check_invariants_dispatches_invariant_21_staff_names_absent_group must fail, and m41b must still pass.
M6b — delete idx.check_group_lists_unowned_staff(&mut v); → m41b_check_invariants_dispatches_invariant_21_group_lists_unowned_staff must fail, and m41 must still pass.

Each half reports the following — which is §3's uniform table rule (round 8) applied, not an exception to it. M6 was the only mutation that carried these requirements before round 8 made them general:

The failing test's assert_eq! output verbatim — pin 6a's harness makes it print the cardinality it got against 1, with the violations vector, so the observation is "the mutated fixture went unreported" rather than "a test failed";
the sibling test's pass verdict, which is what shows the two arms independent;

M6a's FULL blast radius — SIX expected failures, corrected in ratification round 6, each classified. check_invariant is check_invariants(…).filter(…) (invariants.rs:313–:318), so deleting the S→G dispatcher call silences that violation for every consumer, not just the two tests M6a names:

Test	Where it fails	Why
`m41…staff_names_absent_group`	its own `assert_eq!` cardinality check	`check_invariants` returns empty for an S→G fixture
`invariant_21_negative_generator_breaks_staff_to_group_only`	same	the generator's fixture is S→G
`every_invariant_has_a_negative_generator` (`generators.rs:991`)	its own `assert!(!violations.is_empty())` (`:994`)	`check_invariant` is empty on the invariant-21 iteration
`negative_generators_are_reasonably_targeted` (`:1042`)	its own `assert!(kinds.contains(&inv))` (`:1046`)	invariant 21 absent from `kinds`
`every_invariant_shrinks_to_a_small_witness` (`:1004`)	inside `shrink`, at its entry assertion (`generators.rs:933`–`:936`)	`shrink` refuses a starting point that does not violate its target
`shrink_is_idempotent` (`:1025`)	inside `shrink`, same entry assertion	same

All six have one cause — the S→G violation is no longer reported — and all six are expected, not findings. The last two fail with "shrink starting point must violate the target invariant", raised in shrink rather than in the test, so their panic names a location the mutation never touched; report them as shrink-entry failures so they are not misread as a defect in the shrink logic.

M6b's contrast, which is positive evidence. Deleting the G→S arm leaves S→G detection intact, so m41b is the ONLY expected failure — the four all() consumers and the generator test all stay green, because every one of their invariant-21 fixtures is S→G. That asymmetry confirms touch row 8's pinned direction more directly than any assertion about the generator does.

Round 5 added this rule and enumerated the radius from memory instead of deriving it from the document. Touch row 8 already records the four all() consumers, with these exact line numbers, because draft amendment 1 added them as the two-crate root cause — and M6a's outcome was written without consulting the row that owns them. A mutation's blast radius must be DERIVED from the recorded consumer surface, not listed; the surface is in the touch table precisely so it does not have to be remembered.

Why it matters beyond bookkeeping: four unannounced failures, two of them raised inside a function the mutation did not edit, is the shape an executor resolves by deciding the contract is wrong about something. A mutation whose radius is discovered rather than specified invites exactly the improvisation the gates exist to prevent.

Each deletion is a single named call site, which is what makes the two independently removable. (M6 previously said "delete each arm in turn" while pin 6 required only behaviour — so against a single-shared-check implementation there were no two arms and M6 was unexecutable. Pin 6b fixes that at the source rather than restating the mutation.)

The surviving test passing is half the observation, and the half that proves the arms are independent rather than one arm catching everything.

Each fixture must violate ONE direction only, and the report must show that it satisfies the other. A fixture disagreeing in both directions still gets reported after either arm is deleted — the surviving arm catches it — so the invariant looks intact and the deletion is signed by nothing.

Revision C made this permanent rather than mutation-only. M6 previously named no tests, so it demonstrated both directions while mutated and left the restored suite free of durable coverage for either. Pin 6a's tests are the coverage; M6 is now their signature. The generator's fixture (row 8, S→G) is a third artifact and stands in for neither — its all()-driven consumers only ask "is 21 reported?".

M7 — the doc guards discriminate. SPLIT IN TWO by draft amendment 1. Pin 10 covers two independently guarded doc blocks — Staff.group and StaffGroup.members, each with its own grep guard (touch row 3). Reverting "one doc comment" leaves the other guard untested, and a run that reverts the stronger one passes while the weaker guard is still weak. So:

M7a — Staff.group. Revert that block to its disposition-B wording. Required observation: quote t14_staff_group_field_doc_comment_states_sole_authority's own output showing its specific needle no longer matches — the message and the doc_block it dumps — not merely that a test failed. And report t14_staff_group_members_field_doc_comment_states_non_authoritative_projection's pass verdict.
M7b — StaffGroup.members. The same, independently: t14_staff_group_members_field_doc_comment_states_non_authoritative_projection fails with its needle and block quoted, and t14_staff_group_field_doc_comment_states_sole_authority passes.

Named in full by ratification round 9. The expected-outcome table read "its own grep guard; the other guard", which recreates precisely the ambiguity M7's split existed to remove: a report can say "own guard failed" without establishing which guard failed or which survived, and the full-suite output identifies tests by name. Two independent mutations need two named failing tests and two named survivors.

Both must be run and both outputs reported. A guard that passes against both wordings is weakened, not updated — and with one mutation covering two guards, that weakening is invisible.

§4. Gate

cargo test --workspace — full pass. Baseline is 1577 passed / 0 failed / 0 ignored across 42 suites (the post-S27 figure; CLAUDE.md's Green baseline is the single origin — if it disagrees with this line, CLAUDE.md wins and that is a finding). The count will move (t8b renamed, new invariant tests, and rows 9–10's tripwire updates). Give the delta in buckets — net-new, converted-from-existing, tripwire-updated — and if they do not sum to the observed delta, that is a finding, not an arithmetic error to be papered over. Report the ignored count, which MUST be 0 — a "full pass" is satisfied by an #[ignore]d test that never runs.
cargo +1.95.0 clippy --workspace --all-targets -- -D warnings → clean. The toolchain is part of the gate: CI pins 1.95.0, this machine's default stable is 1.97.1, and the repo's CI comment records that 1.97 rejects a bare 2.0 that 1.95 accepts. A clippy result that does not say which toolchain produced it says nothing. Report the toolchain with the result.
cargo +1.95.0 fmt -p epiphany-ops -p epiphany-core -p epiphany-textproj -p epiphany-testkit --check → clean, on the same pinned toolchain. All four changed crates — corrected by draft amendment 1, which added rows 9 and 10 in epiphany-testkit and epiphany-textproj; the command named only two and would have left both unformatted while reporting clean. cargo fmt --all is forbidden — it reaches spikes/ through path dependencies.
git diff --cached --check → clean after staging. The staged list is a SUBSET of §2, not an equality — corrected by draft amendment 1. "Exactly §2" is unsatisfiable here because row 11 is deliberately unused (pin 10a decided in revision D that no label is minted): read literally, "exactly §2" fails whenever a row is correctly unstaged, or invites staging an unchanged file to satisfy it. (This said "row 11 is conditional" until revision E — true when written, and revision D made the row decided-unused rather than conditional. The subset rule is unaffected; only its rationale needed the current term.) Instead: every staged path must appear in §2, and every §2 row must be either staged or named in the report as unused, with its reason. Neither direction may be silent. (This is S27's round-17 correction; S16 carried the formulation S27 had already found unsatisfiable.)

Scope note, not a separate gate — demoted in revision C. §2 carries no prohibitions, unlike S27's §2, so gate 4 is the whole staging check here. If a later amendment adds an absence rule, gate 4 does not cover it: "appears in §2" and "is not forbidden by §2" are different questions. (This was numbered 4a., which produced no command and no output — so it could not be "a gate result" — and collided with §4a, the landing-obligation section. Demoted here in revision C, so no gate carries an a suffix and §4a names only the landing obligation. The gate count that stood in this note until revision J is gone, not corrected.)
spec/vectors/decode_vectors.txt unmodified (pin 9). Confirm by git status, not by inspection.
Invariant 21 is reached through check_invariants on a score violating only it, in the shape of m40_check_invariants_dispatches_invariant_20 (invariants.rs:6063) — for BOTH directions, by pin 6a's two named tests: m41_check_invariants_dispatches_invariant_21_staff_names_absent_group and m41b_check_invariants_dispatches_invariant_21_group_lists_unowned_staff. Both run, both verdicts reported, and each confirmed to satisfy the direction it does not break. (Revision C: this asked for one score, which left one branch with no durable coverage once M6 was reverted.) And invariant_21_negative_generator_breaks_staff_to_group_only (touch row 8), which is the only consumer of the generator's pinned direction. Three tests, all run, all verdicts reported — revision E; revision D named two and left the third with no gate. The generator test's evidence covers BOTH its legs — revision F: the raw fixture and the shrunk one, each with the exact-set, witness-direction and opposite-direction assertions. A gate that accepts only the raw leg leaves shrink free to change what the fixture proves.

EVIDENCE MODEL — CHOSEN EXPLICITLY IN REVISION E. Quote the SOURCE assertions plus the pass verdict; do NOT claim a runtime return.

Revision D said "quote check_invariants' full return … and its witness ids", which the prescribed tests cannot emit: they are assert!-style tests in m40's shape, and cargo test prints ok for a passing test, not local values. A report obeying that literally would need either unpinned --nocapture instrumentation added purely to produce it, or inference from source dressed up as observed output. Neither is evidence.

So for each of the three tests, the report gives:
1. The exact-set assertion, quoted verbatim from source — the len() == 1 and invariant-identity assertions, and the witness-id assertion. It MUST be an assert_eq! on a check_invariants local bound above it, with the violations vector in its diagnostic, and MUST be the first assertion after the bindings — pin 6a's harness, ratification round 5. Confirm the macro is assert_eq! and not assert!: M6 draws its whole observation from that macro's left/right output, so an assert! satisfies pin 6a and disarms M6.
2. The test's pass verdict from cargo test.
A passing exact-set assertion IS the observation; the assertion text says what was checked and the verdict says it held. (This follows S27's gate 6c, which quotes a struct definition rather than grepping for it: a quoted source construct is read, not inferred. Adding print instrumentation to satisfy a report would be new, unpinned code in epiphany-core written for no other purpose.)

(Revision D added the exactness requirement because this gate checked the target verdict and the opposite direction but not the absence of invariants 1–20, so a fixture carrying an unrelated second defect passed every stated check.) all().len() == 21 and the core_spec.tex enumeration ending at 21 are checked in addition, never instead.
Every test named in pin 8 runs, each verdict reported. (Read "the four pin-8 tests" until revision B — a third live tally, left standing while the mutation and gate tallies beside it were removed. Pin 8's table is the origin.)
The pin-1 structural gate: create_staff_group's production body contains the empty-members refusal and no member-liveness/TargetMissing path. METHOD PINNED IN REVISION H, BOUNDARY CORRECTED IN REVISION I.

Quote exactly pin 1's slice: create_staff_group's body, brace-matched from its fn line to its closing brace, production source only — and read it. Report the refusal's lines and state that no TypedObjectId::Staff liveness check and no TargetMissing construction remain within that slice.

Revision H said "to the #[cfg(test)] boundary", which made this gate self-defeating. create_staff_group begins at reduce.rs:4458; create_part_definition begins at :4515 and carries its own PreconditionFailureReason::TargetMissing at :4554; the next #[cfg(test)] is at :9576. So the literal read spans five thousand lines and always contains the very path the gate says must be absent, while any shorter read violates the stated boundary. The gate could not be passed and could not be honestly failed.

Pin 1 had the boundary right all along (slice the create_staff_group body, brace-matched from its fn line to its closing brace). Revision H invented a second, looser one instead of citing it — the same "imported a plausible-sounding rule rather than re-deriving from the pin" failure revision A recorded, applied to a boundary instead of a conclusion. Where a pin already defines the artifact, the gate cites the pin; it does not redescribe it.

Do not establish the absence by grep: a TargetMissing path can be spelled without either literal, so a grep for absence proves only that a chosen string is gone — S27's gate-6a lesson, and the reason its gate 6c quotes a definition rather than searching for it. This is also what M8 signs, so a vacuous gate 8 makes M8's deletion unobserved.
t8c (pin 3a) and t8d (pin 4a) both present and passing, by name.
Pin 12's bump landed, by value comparison — ADDED BY DRAFT AMENDMENT 1.
```
grep -n "pub const CURRENT_REDUCTION_ALGORITHM_VERSION" crates/epiphany-ops/src/lib.rs
git show HEAD:crates/epiphany-ops/src/lib.rs | grep -n "pub const CURRENT_REDUCTION_ALGORITHM_VERSION"
```
→ working tree = 1, HEAD = 0. Report both outputs, not the conclusion, and quote the new Bumps list entry verbatim. A bump without its entry fails this gate.
Rows 9 and 10's tripwires were UPDATED, not silenced, and remain INDEPENDENT of the authority. STRENGTHENED ON REVIEW — the first version permitted the exact tautology it exists to prevent.

It said only "updated, not silenced … not weakened to accept any version", which does not forbid replacing the literals with CURRENT_REDUCTION_ALGORITHM_VERSION. That reads as the tidiest possible update and is the one thing that must not happen: both operands would then move together with every future bump, the comparison would pass for all values, and M5a and M5b become vacuous — §0.1's tautology rebuilt inside the tests written to detect it. S27's round 3 caught this exact substitution and roundtrip.rs:882 says "do not tidy either literal into the constant".

Required, and each quoted verbatim in the report:

a. serialize.rs:663 asserts against an independent literal ReductionAlgorithmVersion(1) — never the constant, never a value derived from it. b. roundtrip.rs:901 fixture capability → synthetic_for_fixture(1), and :918's staged base → literal ReductionAlgorithmVersion(1), so the reopen under production_caps() matches and the Ok arm still runs. c. roundtrip.rs:941's success-arm assertion → literal 1. d. roundtrip.rs:947's mutation-only Err arm — assert_eq!(base, ReductionAlgorithmVersion(0)) → literal 1. Omitted from the first version of this gate. Left at 0, M5b fails at the wrong assertion: the arm is reached only under mutation, and it would abort on the base comparison before reaching the two-field panic! that is M5b's required observation. The mutation would appear to fail correctly while observing nothing. e. The literal-preservation doc comments (roundtrip.rs:872–:883, serialize.rs:648–:657) updated to name 1, with their "do not tidy into the constant" reasoning intact. The reasoning is what stops the next rung making the substitution this gate forbids.

None of a–e may be deleted or #[ignore]d. A tripwire that accepts both values, or that derives either operand from the authority, is not a weakened guard — it is no guard at all.
Pin 6b's two direction checks both exist and are both dispatched — STRUCTURAL, ADDED IN REVISION G. M6 is unexecutable without them, so this gate proves the mutation surface exists before M6 is attempted rather than discovering its absence mid-run.

FOUR INDEPENDENT CHECKS, each required to be EXACTLY ONE. REWRITTEN IN REVISION H — the aggregate count it replaced could pass with one surface missing.
```
grep -c "fn check_staff_names_absent_group"          crates/epiphany-core/src/invariants.rs   # a → 1
grep -c "fn check_group_lists_unowned_staff"         crates/epiphany-core/src/invariants.rs   # b → 1
grep -c "idx.check_staff_names_absent_group(&mut v)" crates/epiphany-core/src/invariants.rs   # c → 1
grep -c "idx.check_group_lists_unowned_staff(&mut v)" crates/epiphany-core/src/invariants.rs  # d → 1
```
Report all four counts separately. Any count other than exactly 1 is a pin 6b violation and a finding — including a count greater than 1, which means a duplicated definition or a doubled dispatch.

And quote the context, because a count is not a mapping:
- each definition with its enclosing impl GraphIndex<'_> header, proving the method belongs to the type check_invariants builds — not a free function, not a method on some other type that merely shares the name;
- each dispatch with the pub fn check_invariants header above it, proving the call is in the dispatcher M6 will edit — not in a test, a helper, or a second dispatcher.
What the previous version permitted, stated so it is not reintroduced. It ran two alternation greps and required "four lines total". Both definitions can exist (2 lines) while check_invariants calls check_staff_names_absent_group twice and check_group_lists_unowned_staff never (2 lines) — four lines, gate passes, and M6b has no call site to delete. The aggregate proved a population, never a pairing. Where a gate must establish a MAPPING, it cannot count — it has to check each element on its own, which is the same shape as this contract's rule against enumerating where completeness is required.

A grep for presence is also defeated by a rename, so pin 6b's names are pinned and this gate and M6 both use them. If the implementation chose other names, that is the finding — the gate has not "passed with zero matches", it has failed.
Pin 5a's undo regression test runs and passes — ADDED IN RATIFICATION ROUND 1. u5_undoing_a_staff_strips_it_from_the_live_groups_members, by name, with its three assertions quoted from source and its pass verdict — gate 6's evidence model. Confirm assertion 1 is present: without the still-live check the test can pass vacuously on a score where the group was undone too, which is exactly how the existing u2tomb_a fails to cover this path. And quote pin 5a's OBSERVATION HARNESS from source — ratification round 3: the bindings of g.members and of the check_invariants violations above assertion 2, and the failure messages of assertions 2 and 3 showing both values formatted in. M5 can only report what these messages print, so a harness this gate does not check is a mutation that observes nothing.
Pin 7a's observation harness in t8b — ADDED IN RATIFICATION ROUND 3, CORRECTED IN ROUND 4. Quote from source, above t8b's first assertion, all four bindings — spurious-order effect, spurious-order members, missing-order members, missing-order invariant-21 violations — and every assertion's failure message showing all four formatted in.

Confirm bindings 2 and 3 come from the two different reductions, not one value reused: they are different groups, and a single members local satisfies a four-line-shaped check while leaving one mutation's observation absent. Name which order each binding was taken from.

Both M1 and M2 draw their observations from these messages, via different assertions — so a message carrying only "its own" values silently disarms whichever mutation trips the other one. This gate must establish a mapping from each binding to its order, not a count of bindings — gate 12's lesson, one section over.

§4a. Landing obligation — files that must NOT be staged, and must be fixed after

ADDED BY DRAFT AMENDMENT 1. Pin 12's bump makes two live statements false, and both are in documents outside this rung's touch table on purpose:

File	Statement the bump falsifies
`CLAUDE.md:106`	"`CURRENT_REDUCTION_ALGORITHM_VERSION` — currently `0`", and "P13-S16 is the first rung that must move it to `1`"
`spec/HANDOFF_2026-08-07.md:25`	the POST-S27 block's "currently `0`" and the same "first rung" sentence

DO NOT STAGE EITHER DURING EXECUTION. Both describe the state of the repository after a rung has landed, and this rung is not landed while it is staged and awaiting acceptance. Staging them would have the contract assert S16 had landed at the moment it was submitted for review — a claim about the future written as present fact, which is the defect class S27's amendments 4 through 7 were spent removing.

They are a POST-ACCEPTANCE obligation. Once the owner accepts and commits this rung, reconcile both in a separate commit, together with anything else the bump falsifies. The report MUST list them as outstanding, so acceptance is not mistaken for completion.

Pin 11 already covers spec/PASS13_CANDIDATES.md, which is different and is staged: a ledger row records what a rung did, and is written by the rung. CLAUDE.md and the handoff state what is true now, and only become true at acceptance. The distinction is whether the file's claim is dated or live — the rule amendment 6 settled for S27, applied here to decide staging rather than wording.

§5. Staging and boundary

Stage only §2's files, by explicit path. Never git add -A.

A concurrent session commits here. Re-check HEAD before staging and before commit. Never git reset, git restore --staged, git checkout, or git stash.

Out of bounds — MUST NOT be read, written, or staged: the entire spikes/ tree, spec/PLAN_EDITOR_APP.md, spec/CONTRACT_EDITOR_*.md, spec/ANALYSIS_GENESIS_PERSISTENCE.md, spec/ANALYSIS_TEXT_RUN_PRIMITIVES.md, spec/DRAFT_T4_FIXTURE_RECIPE.md, crates/epiphany-render-svg/**, crates/epiphany-glyphs/**, crates/epiphany-editor-gui/**, crates/epiphany-testkit/benches/editor_pipeline.rs, the root Cargo.toml, .claude/worktrees/.

Ratified contracts MUST NOT be edited, including spec/CONTRACT_GENESIS_G3A_ENTITIES.md and …_G3A_UNDO_REPAIR.md.

Do not repair core_spec.tex's invariant-10 prose — that is P13-S26, and its evidence at invariants.rs:69–:71 must stay intact.

The executing agent MUST NOT commit. Leave the work staged.

§6. Report requirements

Every mutation listed in §3, each with its verbatim output and, where §3 names one, the behaviour it observed rather than the assertion it broke. And for each, §3's expected-outcome table discharged in full — ratification round 8: the complete observed failure set from a full cargo test --workspace under that mutation, the named failing tests' output, each named required survivor's pass verdict by name, and each named structural gate's output. An omitted named artifact is a finding, as is any failure outside the table's "MUST fail" column or any listed failure that did not occur. No count is stated here. §3's table is the single origin — read its rows. (This item read "the nine mutations (M1–M9)", a tally a report could not both enumerate and obey once mutations began splitting. The replacement then stated a count of its own, which went stale when M6 split as well as M7 — while the very sentence carrying it warned that "a count here goes stale the next time a mutation splits". Removed in ratification round 10: an explanation of a removed count must not restate a count, which is revision J's rule applied to item 1 after J applied it to item 2.)
Every gate listed in §4, each with its command and output. No count is stated here — §4 is the single origin. Where a gate has lettered subchecks, they are reported under that gate, not as separate results. (This item read "the nine gate results", then carried a corrected count, and the corrected count went stale the moment a gate was added. Revision J removed it rather than updating it again — restating a count beside the words "no count is stated here" is the defect naming itself.) 2b. Pin 12's bump and its Bumps entry, with gate 10's two outputs; and rows 9 and 10's tripwire updates, each quoted before and after, with gate 11's confirmation that neither was silenced. ADDED BY DRAFT AMENDMENT 1. 2c. Confirmation that neither pin 6 nor pin 10 minted a \label{req:...} — as pin 10a decided in revision D — that touch row 11 was not staged, and that all three counters are unchanged. If either edit did mint one, that is a FINDING against this contract (§6 item 5), reported with pin 10a's table applied; it is not a decision to be taken during execution. (This item read "decide and report" and, before that, "all three counters and their new values" — the rule pin 10a had just corrected, surviving in its own report consumer. Third site of one false claim: the pin, touch row 11, and here.) 2d. Invariant 21's negative fixture survives shrink with its properties intact — evidenced by invariant_21_negative_generator_breaks_staff_to_group_only's shrunk leg, per gate 6's model: the shrunk-leg assertions quoted from source, plus the test's pass verdict. REWRITTEN IN REVISION F — it said "quote the shrunk witness", which the test cannot emit. That is the same unproduced-runtime-evidence defect revision E fixed in gate 6 and did not carry one hop to this item; and "still violates 21" was the weak membership check that let a shrunk witness flip direction or gain a second defect. (Earlier still, this item asked whether shrink matches GraphInvariant exhaustively — a static fact the draft could read: it does not, it calls check_invariant(score, inv).) 2f. Pin 5a's u5_undoing_a_staff_strips_it_from_the_live_groups_members — its three assertions quoted from source and its pass verdict, and M5's observed state: the post-undo g.members still containing s, with the invariant-21 witness. ADDED IN RATIFICATION ROUND 1, which found pin 5 signed by a mutation and by no permanent test. 2e. The three invariant-21 tests' verdicts — m41, m41b, and invariant_21_negative_generator_breaks_staff_to_group_only — each with its exact-set and direction assertions quoted from source, per gate 6's evidence model, and for the generator test BOTH legs, raw and shrunk (revision F). No runtime return or witness dump is claimed; a passing exact-set assertion is the observation. ADDED IN REVISION E, which found the generator test required by revision D but consumed by nothing. 2a. REWRITTEN 2026-08-09 — it required the opposite of what is now correct. It read: "For pin 0: confirmation that nothing was added claiming to reject or detect stale canonical bases, and that the break is recorded only in prose." That was right while nothing rejected them. S27 now does, on both the read and write paths, so a report obeying the old text would confirm the absence of a guarantee that exists. As rewritten, the report must state:
- that CURRENT_REDUCTION_ALGORITHM_VERSION was bumped to 1, with the bump's entry added to the constant's own "Bumps" list — this rung changes CreateStaffGroup's reduction verdict, and nothing can detect a missed bump;
- that no second detection path was added. S27 owns the check and its tests; a rung that re-implements the guarantee it depends on has built a duplicate that can disagree with the original;
- that the rebuild break is recorded in operation_catalog.tex's Revision History, which is unchanged from the original requirement.
The staged file list, and the test-count delta with its cause.
The pin-8 verdicts — one per test named there, count not restated — and the t6/t7/t9 revisions with what each asserted before and after.
Anything contradicting this contract. A contract defect reported is worth more than a contract satisfied.

158 KiB Raw Blame History Unescape Escape

Contract — P13-S16: the projection gets maintained

What ratification does NOT settle, stated at the top so it is not missed

Draft amendment 1 — 2026-08-09, on ratification reconnaissance. NOT a review round.

Draft amendment 1, revision A — independent review of d06e2f7

Draft amendment 1, revision B — independent review of 3096c54

Draft amendment 1, revision C — independent review of 0a5b936

Draft amendment 1, revision D — independent review of 25473a1

Draft amendment 1, revision E — independent review of cae1d32

Draft amendment 1, revision F — independent review of abe2c35

Draft amendment 1, revision G — independent review of 2818ced

Draft amendment 1, revision H — independent review of 09d8439

Draft amendment 1, revision I — independent review of 5d2db93

Draft amendment 1, revision J — independent review of 9e43994

RATIFICATION ROUND 1 — independent, against 253249e. One blocking finding.

RATIFICATION ROUND 2 — independent, whole-artifact, against 5ec2ce0. One blocking finding.

RATIFICATION ROUND 3 — independent, whole-artifact, against e43cd34. One blocking finding.

RATIFICATION ROUND 4 — independent, whole-artifact, against 3e0a4d3. One blocking finding.

RATIFICATION ROUND 5 — independent, whole-artifact, against e48bb8d. One blocking finding.

RATIFICATION ROUND 6 — independent, whole-artifact, against 85429d6. One blocking finding.

RATIFICATION ROUND 7 — independent, whole-artifact, against 80be47c. One blocking finding.

RATIFICATION ROUND 8 — independent, whole-artifact, against 2a445d1. One blocking finding.

RATIFICATION ROUND 9 — independent, whole-artifact, against 942f261. One blocking finding, and one the sweep escalated.

RATIFICATION ROUND 11 — independent, against 63a2dc8. One blocking finding, fixed by the owner.

RATIFICATION ROUND 12 — independent, whole-artifact, against 25b4925. ZERO FINDINGS.