diff --git a/spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md b/spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md
index b71fbc9..6955020 100644
--- a/spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md
+++ b/spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md
@@ -159,7 +159,8 @@ rows — read it off, do not restate it.
 | *scratch probe* | *1 falsification* | — | *execution, not review* |
 | round 11 | 3 | 2 | **yes** |
 | round 12 | 2 | 2 | **yes** |
-| **Total** | **46** | **33** | one amendment per row |
+| round 13 | 1 | 1 | **yes** |
+| **Total** | **47** | **34** | one amendment per row |
 
 **This block previously read "amended three times … fifteen findings so far,
 eight of them blocking"** — the round-2 figures, left standing through rounds 3
@@ -412,13 +413,34 @@ solely so a negative vector can carry the spelling it asserts is refused.
 widened public surface behind**, which is how a temporary harness becomes an API
 change nobody ratified.
 
-**What round 13 should weigh, stated against interest:** findings by round are
-**9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3, 2**; blocking **4, 4, 4, 4, 2, 3, 3, 2, 2, 1,
-2, 2**. Twelve rounds, **none clean**. The last two rounds found the same *kind*
-of defect — a requirement that reads as a decision but isn't one — which suggests
-scanning M7 for remaining instructions that name a constraint without naming its
-value. **Everything M7 now specifies is pinned to a number, a crate or a named
-artifact**, which is a checkable property a round can test directly. **Treat
+**Review round 13 — 2026-08-08, independent, against `bff9c9a`.** **One finding,
+blocking** — and it inverted M7's result.
+
+| # | Finding | Disposition |
+|---|---|---|
+| 1 | **M7 claimed the capability check "does not fire".** Pin 3a validates a **newly emitted** canonical base, which is exactly what **both** `B_raw` and `A` commit. The check fires on both paths and **accepts**, because the raw version equals the real authority. **As written, M7 was satisfiable by deleting pin 3a's writer check entirely** — a passing M7 demonstrating the opposite of its purpose | Claim corrected to *fires and accepts*; **three required observations**; and a **control** added — the same path with a mismatched version must be **rejected** with `CanonicalBaseRequiresRebuild` |
+
+**This is a new failure shape, and worth naming: an observation satisfiable by the
+absence of the thing it observes.** M7's earlier defects were about being
+unrunnable, or comparing the wrong artifacts. This one would have *run*, *passed*,
+and *reported success* — on a tree where the writer check had been removed.
+**"The check does not fire" cannot distinguish a check that accepts from a check
+that is not there**, and only one of those is the finding.
+
+**The control is what makes the positive result mean anything.** The matching case
+succeeding is evidence only once the mismatching case is seen to fail on the same
+path, in the same run, under the same removals. **M7's removals are now explicitly
+limited to the text refusals** — pin 3a is not among them and may not be weakened,
+because it is the thing under observation rather than an obstacle to it.
+
+**What round 14 should weigh, stated against interest:** findings by round are
+**9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3, 2, 1**; blocking **4, 4, 4, 4, 2, 3, 3, 2, 2,
+1, 2, 2, 1**. Thirteen rounds, **none clean**. Round 13 is the second single-finding
+round and the narrowest since the probe. **But it found a defect of a kind no
+earlier round had looked for** — not "can this run?" or "does this compare the
+right things?" but **"could this pass for the wrong reason?"** — and that question
+has **not** been asked of M1–M6, M5a or M5b. **Every mutation in §4 deserves the
+same check**: what else, besides the intended defect, would make it pass? **Treat
 "dispatchable" as a claim requiring evidence of convergence, not a status reached
 by running out of findings.**
 
@@ -1746,10 +1768,51 @@ conclusion is precisely the reasoning-instead-of-observing this rung exists to
 stamp out — round 1 wrote the conclusion and specified no way to reach it, and
 round 8 specified a way that could not support it.
 
-**What the observation must show, or it has not been made:** that the capability
-check does **not** fire, and that `A.image()` equals `B_fixed.image()` — the
-document launders precisely *because* its number matches, and no check can tell a
-coincidence from a rebuild.
+**What the observation must show, or it has not been made. CORRECTED IN ROUND 13
+— the previous wording inverted the result.**
+
+It read *"the capability check does **not** fire."* **That is backwards.** Pin 3a
+requires `commit`/`commit_versioned` to validate a **newly emitted** canonical
+base — and **both** `B_raw` and `A` commit exactly that. **The check fires on both
+paths. It fires and it ACCEPTS**, because the raw version equals the real
+authority.
+
+**That acceptance is the whole laundering result:** the base is not slipped past
+an absent check, it is **admitted by a check that is working correctly and cannot
+tell a coincidence from a rebuild.**
+
+Therefore M7 must show **all three** of:
+
+1. **`A.image()` equals `B_fixed.image()`** — the container records nothing about
+   how the base arrived.
+2. **Pin 3a's writer validation ran and accepted** on both commits.
+3. **The control below.**
+
+### The control — REQUIRED, added in round 13
+
+**In the same run, with the same harness, repeat the import with a base version
+deliberately NOT equal to the real authority, and observe that the commit is
+REJECTED with `CanonicalBaseRequiresRebuild`.**
+
+**Without this control, M7 is satisfiable by deleting pin 3a's writer check
+entirely** — which would produce a *passing* M7 while demonstrating the exact
+opposite of what M7 exists to show. An observation that the check "does not fire"
+cannot distinguish **a check that accepts** from **a check that is not there**,
+and only one of those is the finding.
+
+**The control is what proves the check was live.** The matching case succeeding
+means something only once the mismatching case is seen to fail on the same path,
+in the same run, under the same removals.
+
+> **M7's removals are limited to the text refusals — pin 3a is NOT among them,
+> and MUST NOT be removed or weakened.** M7 mutates the *text* boundary to make a
+> path reachable; it does not mutate the *authority* boundary, which is the thing
+> under observation. Removing both would not be a stronger mutation, it would be
+> a different and empty experiment.
+
+**Report all three observations, and the control's error variant by name.** A
+report giving only observation 1 has not made the demonstration — it has measured
+two byte strings.
 
 ### The claim M7 establishes, and the one it does NOT. SCOPED IN ROUND 11.
 
@@ -1886,7 +1949,11 @@ in §N" to a number.*
 1. **Every mutation in §4** — currently **eight** (M1, M2, M3, M4, **M5a**,
    **M5b**, M6, M7), each with verbatim output. *(M5 split in review round 2.)*
    **M4 is a recorded prohibition** (observed to compile, then reverted) and
-   **M7 is a demonstration whose expected outcome is success**. Neither is a
+   **M7 is a demonstration whose expected outcome is success — except its
+   control (round 13), whose expected outcome is a REJECTION.** A report that
+   treats the control's rejection as a problem has misread it: that rejection is
+   the evidence pin 3a's writer check was live, without which M7's success proves
+   nothing. Neither M4 nor M7 is a
    passing guard; do not report them as one.
 2. **Every gate item in §5** — currently **eight** (1, 2, 3, 4, 5, 6, 6a, 7),
    each with the command that produced it.
@@ -1907,7 +1974,7 @@ in §N" to a number.*
    | M5a | test **10a** fails, **plus** the provenance of the asserted operand |
    | M5b | test **10b** fails with both error fields asserted, **plus** the provenance of *both* operands |
    | M6 | test **5** fails (first half); test **9** fails (second half) |
-   | **M7** | **no test — its expected outcome is success.** It owes the comparison M7 specifies, and confirmation that the refusals M7 names as removed were restored. **Counts, methods and file names live in M7, not here** — this cell said "all three refusals" until round 8 and "field-by-field enumeration" until round 9, each time describing a method M7 no longer used |
+   | **M7** | **no test — its expected outcome is success.** It owes **every observation M7 specifies, including its control**, and confirmation that the refusals M7 names as removed were restored. **Counts, methods and file names live in M7, not here** — this cell said "all three refusals" until round 8 and "field-by-field enumeration" until round 9, each time describing a method M7 no longer used |
 
    **M4 and M7 were unsatisfiable under the previous wording.** Item 1 already
    said both are "not a passing guard", yet this item demanded a test each
diff --git a/spec/PASS13_CANDIDATES.md b/spec/PASS13_CANDIDATES.md
index 3cb424e..d8208e4 100644
--- a/spec/PASS13_CANDIDATES.md
+++ b/spec/PASS13_CANDIDATES.md
@@ -122,5 +122,5 @@ evidence in isolation.
 | P13-S23 | **No filed candidate owns "place any anchor pair on a common timeline and measure musical distance along it" — P13-S18 previously mis-cited a narrower capability as its gate.** Two disjoint deficiencies, both owned by this candidate. (1) **No ordering.** The pair is not comparable under any of `measure20_comparable_order`'s five shapes c1-c5 (`invariants.rs:2457`) at all — whether the failure is in the **referent** (distinct `Event` ids; distinct `Measure` ids outside c3's `Start`+`Zero` restriction), the **variant or selector** (`Event` against `Measure`, `Measure` against `Region`, differing `pos`/`edge`), or the **clock** (`Musical` against `WallClock`, including inside `measure20_offset_order`, `:2419`) — this is what invariant 20's A4 and B4 are made of. (2) **Ordering without a usable delta.** The pair IS comparable and still yields no musical distance: c3 supplies a vector index (an order, never a distance), and c5 compares two `WallClock`s, and `measure20_musical_delta` (`:2522`) never returns a `WallClock` delta (`:2527`) — this is what invariant 20's B5 is made of. Scoping this as merely "anchors of differing shapes" or "not directly comparable under c1-c5" would exclude B5 entirely — S5 (distinct-id `Measure` `Start`/`Zero`) is c3-comparable and S1 (`WallClock` measures, `WallClock` meter changes) is c5-comparable, and both still reach B5 — an earlier draft of this filing made exactly that narrower mistake. **Explicitly broader than P11-C5**: P11-C5 (`PASS11_WORKLIST.md:159`) is a re-anchoring proximity metric that resolves "when the graph-mutation phase tracks resolved positions", and covers narrowly the two-distinct-`Event`s case (`CONTRACT_GENESIS_G3B_MEASURE.md:223`, `effect.rs:139`-`:142`'s `PositionOutsideRegion` Reserved note); P13-S23 is the timeline itself, whatever positions get placed on it. Names its dependents: invariant 20's A4, B4 and B5, and `PositionOutsideRegion`'s Reserved status | `spec/CONTRACT_P13S18_MATRIX.md` pin 10 (filed 2026-07-31 during the same rung that corrected P13-S18's over-narrow P11-C5 citation) | **open.** No code owed by this rung. Closing it needs the deferred common-timeline/duration machinery — once a `Measure` end, a distinct-id `Measure`/`Event` referent, or an `Event` position on a wall-clock-placed region can be placed on a common timeline with a musical distance, invariant 20's A4/B4/B5 residue and `PositionOutsideRegion`'s Reserved status shrink together |
 | P13-S25 | **The committed decode corpus's numbered tag rows lock byte→byte, not variant→byte — one row already has the property the other thirty-nine lack.** `ops/src/vectors.rs:206`–`:209` emits one row per tag as `format!("tag_{:02}", tag.discriminant())` carrying `[discriminant]`: **both the name and the payload derive from the value alone**, so `tag_32` asserts that `0x20` round-trips and never that `SetCanvasLayoutDefaults` is 32. The `Registered` row (`:210`–`:217`) is different — its name is the hard-coded string `"registered"` while its bytes are computed from the variant, so the frozen literal at `spec/vectors/decode_vectors.txt:80` binds the association. **Disposition B of P13-S22:** give the numbered rows the same property. It **does** catch the coordinated permutation — by exactly the `Registered` mechanism, with the committed text serving as the independent statement — and it propagates the property to every implementation that reads the cross-impl corpus, which an in-crate Rust test cannot do | `spec/CONTRACT_P13S22_TAGLOCK.md` (disposition B, considered and deferred during the 2026-07-31 ruling; filed rather than left as a closing remark, per the same discipline that moved P13-S22 out of P13-S15's resolved row) | **open. Complementary to P13-S22, not a replacement for it, and not a re-litigation of it.** P13-S22 landed disposition A (`tag_wire_discriminants_are_golden`, `payload.rs:2730`), which fails **by variant name inside the crate**. B cannot supply that: its failure is still *"spec/vectors/decode_vectors.txt is stale. Regenerate: …"* (`testkit/src/vectors.rs:224`) — the misleading diagnosis P13-S22 was filed about — even though the diff text would now name variants. **What B buys is cross-implementation reach; what it costs is churn in a committed artifact other implementations pin.** Both are wanted; neither substitutes for the other. Sequencing note: run B's own signing mutation as the coordinated permutation (literals *and* declaration lines), since the literal-only form is caught today by row ordering and proves nothing |
 | P13-S26 | **A doc comment in shipped code claims a specification repair that never landed, and the claim is guarded on the code side and nowhere on the specification side.** `crates/epiphany-core/src/invariants.rs:69`–`:71` enumerates invariant 10's four reference classes and states that *“genesis tranche G3a repairs this prose to name what the check body already enforced”*. **It did not.** `core_spec.tex:6570`–`:6572`, the normative enumeration item 10, still reads only *“Every cross-cutting structure's references resolve to extant objects in the graph, except where explicit re-anchoring rules permit transient dangling states during edits”* — naming neither a staff's declared instrument, a staff's group, a staff group's members, a part's staves, a view's active layers, nor any of the meter/time-signature references the Rust doc lists and the check body enforces. The repair landed in the Rust doc comment only. **The asymmetry is the defect's sharp edge:** the Rust doc block is protected by a grep-assert, `t12_invariant_10_doc_comment_names_the_four_reference_classes` (`invariants.rs:4554`, needles at `:4562`–`:4566`), so the side that is *wrong about the other* is the side that is **locked**, while the side that is actually stale is unguarded | this file (found 2026-07-31 during P13-S16 reconnaissance, while verifying that row's invariant-10 citations; no ledger entry covered it) | **open.** **Not a live incorrectness** — the check body is correct and enforces every class; only the normative prose under-describes it, and only the doc comment lies about that. **A P13-S9 instance**, and filed deliberately as one: the loud form (a dangling citation) is caught by `requirement_labels.rs`, and this quiet form — a *true-sounding claim about another document's state* — is caught by nothing. **`invariants.rs:69`–`:71` MUST NOT be “corrected” on its own.** It is currently the only artifact in the tree pointing at the `core_spec.tex` gap; softening the Rust claim in isolation would make the specification defect invisible and convert a caught defect into an uncaught one — which is P13-S9's stated failure mode verbatim. **Repair both sides in one rung**, and consider whether the LaTeX enumeration deserves the grep-assert its Rust mirror already has |
-| P13-S27 | **The reduction-algorithm-version machinery is self-referential, so the one check that would detect a canonical-semantics change necessarily passes.** `core_spec.tex:11614`–`:11617` is normative — *"Snapshots produced under an earlier algorithm version cannot be used as canonical bases under a later one without rebuilding"* — and `:14369`–`:14372` states that replicas at differing versions *"may produce different canonical states from the same operation set."* The machinery to enforce it appears to exist: `ReductionAlgorithmVersion` (`bundle/src/ids.rs:291`) is a superblock wire field (bytes `68..72`, `superblock.rs:20`); `reduction_version_for` (`bundle.rs:989`) sets a new superblock's value; and `open` (`bundle.rs:396`–`:399`) rejects a mismatch. **But the writer sources the value from the canonical base's own self-report** (mapping the base's `reduction_algorithm_version` through `unwrap_or_default()`), **and the reader compares it only against the superblock that value seeded.** Nothing compares either against the semantics the running implementation actually implements. **The check is not vacuous** — it catches a corrupt or tampered base whose version disagrees with its superblock — but it **necessarily passes for a conformingly propagated stale base**, which is precisely the case the requirement exists to prevent. Supporting: **no constant or accessor anywhere names 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 in that crate — **a second instance of P13-S26's pattern**, a doc comment asserting a false fact about another module | `spec/CONTRACT_P13S16_PROJECTION.md` pin 0 (found 2026-07-31 while scoping P13-S16, which is a canonical reduction-semantics change and therefore the first rung to need this guarantee; filed in the same ledger edit as the row it blocks) | **UNBLOCKED 2026-08-07 — the format-epoch rung landed; dispatchable, and still blocking P13-S16.** (Was: open, BLOCKED on P13-S28.) **Scoped 2026-07-31 as `spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md` (DRAFT, not dispatchable).** Rulings taken: a typed `BundleCapabilities` required at both `Bundle::open` and `Bundle::create` and carried on the `Bundle` — no default, so every caller states the semantics it implements — and outright rejection on mismatch via a new `CanonicalBaseRequiresRebuild` error, not read-only and not an integrity anomaly. Storing the capability keeps all 57 `commit` sites unchanged; only `open` (57 sites) and `create` (32) move. **The scoping also falsified this row's first reading that the writer path was test-only:** `epiphany-textproj`'s `serialize_document` (`serialize.rs:119`) and `project.rs:936` are production paths that copy a base's `reduction_algorithm_version` verbatim into a fresh `SnapshotRef`, which `commit_versioned` then stamps into the superblock (`bundle.rs:798`) — so production mints self-consistent stale documents **without ever calling `open`**, and the capability must govern writers too. **What blocks it:** contract pin 2a. Baseline authority `0` does not preserve the corpus (`serialize.rs:327` stamps `1` and round-trips it; `vectors.rs:353`/`:363` likewise), and once P13-S16 moves the authority to `1`, a pre-S27 base that happens to carry `1` is **indistinguishable from a legitimately rebuilt one** — a raw `u32` carries no provenance. Four dispositions are recorded there; `FORMAT_MINOR` as a provenance carrier was proposed and **rejected** (the header never changes after creation, `core_spec.tex:10799`, so a legacy bundle committing a freshly validated base keeps its old minor forever; and a minor change may only append append-safe discriminants, `:12258`, not alter acceptance semantics). The surviving requirement — provenance must ride a container property **old readers cannot silently accept** and **a later commit cannot inherit unchanged** — is a format-epoch design, filed as **P13-S28**. **Scope of the claim, deliberately narrow:** this establishes that the **current implementation** has no detection mechanism. It does **not** establish that no reduction-semantics change in the project's history was ever detectable — that needs a history audit not yet done, and the stronger sentence is deliberately not written here. **What closing it requires:** an authority naming the semantics this build implements, and a rejection-or-rebuild path when a base disagrees with it. Until then any rung changing canonical reduction semantics can record its break in prose but cannot make stale bases unusable — which is why P13-S16's contract is complete, ratifiable as a plan, and **not dispatchable**. **Method note:** an earlier draft of S16's pin 0 claimed no writer path existed at all. That was false, and the way it was false is the point — the search behind it looked for `ReductionAlgorithmVersion(` constructor calls, which 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. **UNBLOCKED 2026-08-07:** the format-epoch rung landed and its pin 8 **resolves pin 2a** — reduction-version authority is meaningful only in major-1 containers, so legacy bases are refused by container epoch and never by version arithmetic. The collision pin 2a identified never has to be adjudicated: a pre-S27 base carrying `1` and a rebuilt S16 base carrying `1` are indistinguishable as numbers but can never meet, because the former exists only in a major-0 container, refused at the epoch boundary before any version is compared. The `u32` never has to carry provenance because the container does. **S27 now additionally owes three inherited items** (both interim refusals converted to validation, M8's deferred laundering demonstration, pin 3c's two suspended conformance assertions), recorded in its contract as required tests. **~~RATIFIED 2026-08-07 after review round 1~~ — RATIFICATION WITHDRAWN 2026-08-07, see round 2 below. Current status: NOT RATIFIED, NOT DISPATCHABLE, pins NOT frozen, **awaiting the next independent review round. Which rounds have closed and the running tally live in the contract's own history table and are deliberately NOT restated here — this clause carried a count through two rounds and went stale in both, and the round list went stale the same way**; no execution work may begin.** Round 1 was run because this contract had reached "dispatchable" with **zero** ratification rounds on record, against the standing rule that contracts go through adversarial review before dispatch. Round 1 returned **nine findings, four blocking**, all now carried in the contract. **Correcting this row's own preceding clause:** the three inherited items were *not* all "recorded as required tests" — obligation 2, M8's laundering demonstration, appeared in **neither** the contract's test section nor its mutation plan, while that section's preamble claimed all of them were tests. It is now **M7**, and ruled a **mutation** rather than a capability restoration: the format rung's text refusal is permanent, `COMPANION_VERSION` stays 0.14.0, and the text-projection corpus keeps `canonical_bases` reach 0. The other blocking three: §0.4's `commit`-site count counted a same-named method in `epiphany-editor-core`, **a crate with no `epiphany-bundle` dependency at all** — the third instrument failure recorded in that one section; three independent stale list-counts (the test-section header, gate 1, and three report items) all naming figures the document had outgrown; and `testkit/tests/requirement_labels.rs` missing from the touch table while pin 9 may move `CORE_REQUIREMENT_COUNT` 213 → 214 — **the escapee `CLAUDE.md` names by name**, which also escaped the format-epoch rung. Non-blocking: locator drift since `381c498` (`bc06706` grew `bundle.rs` by 338 lines; pin 5's own `:396`–`:399` confirmed unmoved), pin 2a's corpus evidence superseded by the 2 → 0 rebuild, `Bundle::open(` 57 → **60**, gate 6a's scope widened to `epiphany-testkit`, and a missing **commit-side positive** test now added as test 8. **REVIEW ROUND 2, 2026-08-07, against the frozen contract: six further findings, four blocking — and round 1's ratification was therefore premature.** (1) The call-site correction had been applied to §0.4 only, leaving the "Rung type" paragraph at **57** and touch row 2 claiming `bundle.rs` has **35** opens — a figure that was never `bundle.rs` alone and is stale besides — which made the required reconciliation impossible. (2) §0.4 called `project.rs:936` a **production** bundle writer; `#[cfg(test)]` starts at `:630` and every `Bundle` call in that file is below it, so the writer-path correction stands on `serialize.rs` alone. (3) **M5 was unexecutable**: `serialize_document` refuses bases at `serialize.rs:151`, so its output is base-free, and pin 5 with test 4 require base-free bundles to open at *any* authority — split into **M5a**/**M5b**. (4) **M6's second half was unexecutable**: `open` rejects a stale base, `create` rejects a base-bearing manifest (`bundle.rs:234`), and `commit` validates what it emits, so no caller can hold an open `Bundle` with a stale *inherited* base — replaced by broadening rather than narrowing, with the unreachability itself reported as the stronger result. Non-blocking: pin 3a's justification (*"production code mints a self-consistent stale document"*) is **false in-tree** — zero production paths stage a base — so it now rests on guarding the public `commit_versioned` API; and `serialize.rs:157` is dead code orphaned by the `:151` guard, recorded and explicitly **not** repaired here. **Two of round 2's findings were introduced by round 1** — ruling M7's refusal permanent is what made M5 unexecutable, and test 8 was added without re-deriving M6 against the same reachability. **Method note: an amendment is a change to the system, not a patch to a line**; a round must re-derive every mutation against every ruling the previous round made. **Round 3 is warranted before dispatch — the defect rate has not fallen (9, then 6), and "dispatchable" is a claim requiring evidence of convergence rather than a status reached by running out of findings.** **REVIEW ROUND 3, 2026-08-07, INDEPENDENT, against `b842975`: six findings, four blocking — and every blocking finding was a defect in text rounds 1 and 2 wrote.** (1) **Pin 3a still carried the rationale round 2 retracted** — §0.4 states there is no in-tree production base writer while pin 3a still said "§0.4 shows production code minting a stale document", so the contract asserted a claim and its negation; **the third occurrence of fix-one-site-leave-the-others**. Rewritten onto the footing that survives: `commit`/`commit_versioned` are **public API** and guard out-of-tree callers, not an in-tree path. (2) **M5a had no observation mechanism** — pin 3 required the capability be *stored* and nothing exposed it; `Bundle` carries 17 public accessors and none for capabilities, so no `epiphany-textproj` test could inspect it. **`Bundle::capabilities()` is now pinned** — new scope, flagged for round 4. (3) **M5b could not fail**: if the supplied capability and the base version both derive from `CURRENT_REDUCTION_ALGORITHM_VERSION` — the natural implementation, since `roundtrip.rs:367` currently hardcodes `ReductionAlgorithmVersion(0)` — both operands move together and the comparison passes for every value. **This is §0.1's own tautology reproduced inside the mutation built to detect it.** The base version must now come from a source that does not track the authority (persisted artifact or deliberate literal), with both operands' provenance reported. (4) **M6's replacement named a scenario with no test** — test 6 stops at opening, so nothing asserted that an unrelated commit *succeeds*, and an implementation rejecting every post-base commit passed tests 2/5/6/8 while the broadening had nothing to break; **test 9 added**. Cleanup: touch row 7 listed `generators.rs` as "call sites, real authority" though it has **zero** `Bundle::open`/`create` calls and its `rng.range(0, 8)` versions are precisely the arbitrary wire values pin 3b assigns to *synthetic* capabilities — split to **row 7a**; and §7's call-site attribution credited round 1 alone where rounds 1 and 2 are both load-bearing. **The pattern is now legible and it is not about counts: three separate mutations were unrunnable in three different ways — M5a could not observe, M5b could not fail, M6 had nothing to break. §7 item 4a now requires, for every mutation, the named test it breaks and the provenance of each operand.** **Defect rate across three rounds: 9, 6, 6 — not converging.** The newest text (pin 3's accessor, M5a, M5b, test 9, row 7a) has had **zero** adversarial passes and was written by the same agent whose previous two attempts round 3 falsified. **REVIEW ROUND 4, 2026-08-07, INDEPENDENT, against `53292f6`: five findings, four blocking.** It accepted pin 3's `capabilities()` accessor as **bounded** — the first new text any round has passed — and found the M5 pair defective a third time. (1) **M5b cited the wrong value**: `roundtrip.rs:367` sits in `assert_score_serialization_stable` (`:332`) and versions an **acceleration snapshot**, not a canonical base, while `assert_reduction_serialization_stable` has **no base at all** because pin 3c suspended it — so the value round 3 warned the implementer not to touch was irrelevant to the authority check. **The tautology diagnosis stands; only its evidence was wrong.** (2) **The instrument was left unchosen** — round 3 said "the rung picks one" and offered two, one of which does not exist for the nominated crate, since `craft_image_with_base` is a private `fn` inside `epiphany-bundle`'s `#[cfg(test)]` module (`:1648`). **Now chosen: commit-then-reopen through public API only** — build with `synthetic_for_fixture(0)`, commit a base carrying the literal `0`, reopen those bytes under the real constant. (3) **No test could assert the error fields**: `assert_reduction_serialization_stable` returns `()` and reopens with `.expect` (`:292`), so a mismatch panics rather than yielding a matchable `CanonicalBaseRequiresRebuild { base, current }`. **Test 10b added.** (4) **M5a violated §7 item 4a, the rule round 3 added in the same edit** — it named no test, and its natural assertion compares the constant with itself and cannot fail. **Test 10a added, asserting against a deliberate literal.** **Round 3's error is the one to carry: it grepped `ReductionAlgorithmVersion`, saw a `roundtrip.rs` hit, and attributed it without resolving the enclosing item — the same shape as §0.4's `.commit(` miscount that round 1 had already recorded as a lesson. Recording a defect is not the same as not committing it.** Both literals in tests 10a/10b are **load-bearing as literals**; §7 item 4b now requires confirming neither was tidied into the constant, a failure mode invisible to the suite. **Defect rate: 9, 6, 6, 5 — still not converging after four rounds, and every blocking finding in rounds 3 and 4 was in text written to fix the previous round.** **REVIEW ROUND 5, 2026-08-08, INDEPENDENT, against `df9e528`: four findings, two blocking — the first round in which blocking findings fell below four.** (1) **The status history was numerically stale again** — "amended three times … fifteen findings so far, eight blocking" were the round-2 figures, left standing through rounds 3 and 4 **while the tables recording those very rounds sat directly below them**. This is the **fifth** count-staleness defect in five rounds, and it was in the one block the author edited every round. Replaced with a **table**, so a round appends a row rather than requiring a number to be found and re-derived. (2) **Test 10b could not make the two-field assertion M5b requires**: §3 said only "assert it opens", and under mutation that yields a bare `Err` or a panic — **a `#[test] -> Result` that returns `Err` asserts nothing about that error's fields**, so M5b's required observation had no home in the test M5b names. Both `Result` arms are now pinned, plus a third for the wrong-error case, so the mutation run produces a *verified* observation rather than a stack trace. Smaller: **M5b's "cannot be tidied" claim was false** — keeping `synthetic_for_fixture` while passing `CURRENT_REDUCTION_ALGORITHM_VERSION` as both its argument and the base version preserves the fixture and fully restores the tautology, so the structure does not protect itself and the real protection is §7 item 4b; round 4 asserted a structural guarantee that undercut the procedural check actually doing the work, **which is the same error as reasoning that a mutation would fail instead of running it**. And §3's preamble still said the tests were "in `epiphany-bundle`" after round 4 added two that **cannot** be, since `epiphany-bundle` must not depend on `epiphany-ops` and reaching the real authority is their entire purpose — corrected, with each test's touch-table home named. **Blocking findings by round: 4, 4, 4, 4, 2 — the first movement in four rounds and the first weak evidence of convergence, set against the fact that every round since the third has found blocking defects in text written to fix its predecessor.** **REVIEW ROUND 6, 2026-08-08, INDEPENDENT, against `03c85dd`: three findings, ALL THREE BLOCKING, and all three in text round 5 wrote.** (1) **The amendment tally went stale inside the block round 5 restructured to prevent exactly that** — round 5 turned the review totals into a table and left "amended five times … rounds 1–4" as prose immediately above it. The amendment count is now **the number of rows**, with no separate figure to go stale. (2) **§3's test-home correction was itself false**: round 5 wrote "tests 1–9 in `epiphany-bundle`", but **test 7 *is* `assert_reduction_serialization_stable`**, which the same section names as `testkit/src/roundtrip.rs`. Two wrong versions of that sentence, both written while fixing it; replaced with a per-crate table (1–6/8/9 bundle, 7 and 10b testkit, 10a textproj). (3) **§7 item 4b protected one operand where test 10b has two** — replacing **both** `synthetic_for_fixture(0)` **and** the committed base's `ReductionAlgorithmVersion(0)` with the constant **keeps the synthetic call in place** and fully restores the tautology, and **test 10b's `Err` arm never executes in the unmutated run**, so its literal cannot detect it. Item 4b now enumerates all **three** fixture operands individually and requires each quoted verbatim. **All three findings are one defect in different clothes: a fix applied to the site named rather than to every site the claim covers** — the sixth count-staleness defect in six rounds and the third range-correction that did not check its own range. **The mechanism that works is structural, not vigilant: the review totals stopped going stale when they became a table, the amendment count did not because it stayed prose, and item 4b stopped being under-specified when it became a table.** **Demonstrated a seventh time inside round 6's own amendment**, where the new table's Total row was first written "6 amendments" — a free-standing count, three paragraphs after the sentence declaring no such count exists, and already wrong at seven rows; caught before commit and replaced with "one amendment per row". **Prose invites a number and a table does not; the defence must be the shape of the artifact, not the attention of the editor.** **REVIEW ROUND 7, 2026-08-08, INDEPENDENT, against `c0d896c`: three findings, all blocking, and all three the same defect — a claim living in two places and fixed in one.** (1) **§7 item 4a was unsatisfiable**: it required every mutation to name "the test it breaks", while item 1 four paragraphs above states that **M4 is observed to *compile*** (no test is possible — that is why pin 3's prohibition is a review rule) and **M7's expected outcome is *success***. A report obeying 4a literally could not be written, and the honest response would have been to invent a test. 4a is now a table of what each of the eight mutations owes, with M4 and M7 carved out explicitly. (2) **Round 6's three-literal correction reached §7 and not §3** — §3 still said "**both** literals … tidying **either**", so the contract carried the fixed and the broken version of the same claim, reopening exactly the narrow-scope ambiguity round 6 existed to close. §3 no longer states the count at all; it points at item 4b. (3) **"Rounds 3, 4 and 5 were independent"** went stale the instant round 6 closed, sitting in prose beside the table whose own column records it. Deleted. **Three rounds, one lesson: round 5 fixed the review totals and not the amendment tally beside them, round 6 fixed item 4b and not §3's copy of the same rule, round 7 found the classification sentence duplicating the table's column. The defect is duplication, and every previous remedy was vigilance — "check the other sites too" — which has now failed three rounds running. The remedy adopted here is deletion, not diligence: where a claim had two homes, one is removed and replaced with a pointer. A copy that cannot drift is one that does not exist.** **Findings by round: 9, 6, 6, 5, 4, 3, 3 — flattened rather than still falling. Blocking: 4, 4, 4, 4, 2, 3, 3 — rounds 6 and 7 were both 100% blocking and 100% in the previous round's text. Seven consecutive rounds, no clean round yet. The deduplication is the first structural remedy for this particular defect and therefore the first with a reason to work, but it is untested.** **REVIEW ROUND 8, 2026-08-08, INDEPENDENT, against `9829ae3`: two findings, both blocking — and the first round to reach into a mutation's mechanics rather than its bookkeeping.** (1) **M7 did not describe a runnable observation.** It instructed execution to *construct* a base-bearing `TextDocument`, which **bypasses `parse_document` entirely**, so the parser refusal it ordered removed was irrelevant and the demonstration was not the **import** laundering it is named for; `project_text_document` is the **export** direction (`&TextDocument -> Result<String, _>`) and is not on the path at all, so "all three sides, since removing one leaves the others refusing and the document never reaches the writer" was simply false for it; and "byte-indistinguishable from one whose base was genuinely validated" named **no comparison artifact and no comparison method**, leaving the central claim a conclusion rather than an observation. Now: the input must be **text and must be parsed**; only the parser (`parse.rs:138`–`:147`) and serializer (`serialize.rs:151`) refusals are removed and restored; the comparison artifact is **test 10b's construction** with the same `FileUuid` and base bytes; and the comparison is a **field-by-field enumeration** of the `canonical_base` `SnapshotRef`, the superblock's reduction version and the header's major/epoch, **reported rather than concluded** — informative in both directions, since a field that *does* differ is a provenance signal nobody knew existed. (2) The round-7 deduplication was incomplete: the status block still carried "rounds 3 and 4 are closed" while declaring the history table the sole authority. **Finding 1 is the most substantive of any round, because every earlier one was about text agreeing with other text — this one is about whether the experiment runs at all, and it did not. M7 had been in the contract since round 1 and survived seven reviews, three of which specifically re-derived mutations, because reading it never required tracing what calls what. An observation stated in the right register can look complete for a long time; "indistinguishable" was a conclusion sitting inside the rung's own demonstration, which is the exact failure mode this rung exists to eliminate.** **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2. Eight rounds, none clean. While amending, the author caught a third instance unaided — §7 item 4a's M7 row still said "all three refusals" — which is weak evidence the deduplication rule is being applied rather than merely stated. The M7 rewrite is now the newest and least-reviewed material in the contract, and its predecessor survived seven rounds while being unrunnable.** **REVIEW ROUND 9, 2026-08-08, INDEPENDENT, against `01e76d1`: two findings, both blocking, both in M7's comparator — the text round 8 had just rewritten.** (1) **Test 10b is not the "genuinely validated" reference M7 nominated**: its write-side capability is `synthetic_for_fixture(0)` and only its *reopen* uses the real authority, so M7 would have compared one synthetic fixture against another with the validated half of the claim simply absent. **This is a collision between two of the contract's own designs, not a typo** — round 4 made 10b synthetic-on-write *deliberately* so M5b's operands would be provably independent, and that is exactly what disqualifies it here. **One artifact cannot be both independent of the real authority and committed under it.** M7 now builds its own reference in `epiphany-testkit`, committing a base under `caps` derived from the real constant so pin 3a validates it on the way in. (2) **The field enumeration could not support its conclusion**: it claimed "everything that could carry provenance" while omitting `FixedHeader.file_uuid` — **the field it required to match** — plus the superblock's `generation`, `manifest_offset`, `manifest_length` and `manifest_hash`, and the whole manifest outside `canonical_base`. Replaced with **whole-`image()` byte comparison**, any difference enumerated and classified as justified nondeterminism (normalize, stating why) or as a **provenance signal** (a finding, since the refusal may then be stronger than needed). **Finding 2 retires a technique rather than an instance: a hand-written list of "every field" is a claim about a struct's contents that is wrong the moment the struct changes, and this one was wrong the day it was written. Comparing the whole artifact cannot be incomplete — the tables-over-numbers lesson applied to the experiment instead of the prose.** Three further sites were caught by the author while amending: §7 item 6 still said "M7's three text refusals", surviving round 8's correction of that exact count in two other places; §7 item 4a's M7 row still named the superseded method; and round 8's own disposition cell stated it as current. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2. Nine rounds, none clean. Rounds 8 and 9 both found defects in the immediately preceding round's rewrite of the same paragraph, so M7 has now been wrong in three distinct ways across three consecutive rounds — unrunnable, then wrong-artifact, then wrong-method. The comparator is on its third design and has never been executed.** **REVIEW ROUND 10, 2026-08-08, INDEPENDENT, against `0efd543`: one finding, blocking — the smallest round yet, and again in M7.** **The whole-image comparison had no complete construction alignment.** Round 9 named four things to align, but `serialize_document` also fixes `document_id`, `lineage_id`, `profile_declarations`, every extension's fields and preserved chunks, the envelope payloads, the **staging order** (base root → extension chunks → operation-envelope block), the manifest schema `major` and `epoch_max`, and every chunk ref, hash and offset derived from those. **So a byte difference would have had a third possible cause — "the reference was built differently" — which is neither permitted classification; the result would have been unclassifiable and the comparison meaningless. A result that cannot be classified is not an observation.** M7 is now a **round trip**: build `B` validated under the real authority, export it to text via `document_from_bundle` + the crate-private `render_text_document`, parse that text back, re-serialize as `A` with `B`'s `FileUuid`, and compare whole images. **Alignment is inherited rather than enumerated** — every input `serialize_document` reads is already `B`'s own, so no list can be incomplete and the setup-mismatch category is eliminated by construction rather than by care. It is also the realistic form of the threat: export a validated document to text, re-import it, and observe the re-imported container is indistinguishable from the original having validated only the base's number, never its provenance. **This was the third hand-enumerated "complete set" in this contract and the third wrong on the day it was written — "every field that could carry provenance" (round 8), "every field to align" (round 9), and round 9's list again. The single rule earned across rounds 5–10: where a claim requires completeness, do not enumerate, derive. Tables instead of counts, whole artifacts instead of field lists, one shared origin instead of an alignment list.** No refusal count is stated anywhere in M7 any more; three successive wordings each had a wrong one. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1. Ten rounds, none clean, and three consecutive rounds have found one paragraph — M7 — defective in a new way each time: unrunnable, wrong artifact, wrong method, incomplete alignment. Findings are falling steadily and each of the last three has been narrower than the last, the first sustained convergence signal here. Against that: M7 has never been executed and each of its four designs looked correct when written — the open question for round 11 is whether the next defect is findable by reading at all, or whether M7 must be run against a scratch branch before further paper review can add anything.** **BOUNDED SCRATCH PROBE, 2026-08-08, authorised as an explicit narrow exception in the contract's status block and run on a discarded branch: it FALSIFIED round 10.** First, recorded as evidence in its own right: **M7 cannot be executed at all until S27 lands** — `BundleCapabilities` and `CURRENT_REDUCTION_ALGORITHM_VERSION` do not exist in the tree, being S27's own deliverables, and M7 step 1 needs a base committed under the real authority. M7 is a mutation *of this rung's implementation*, so it runs after the rung. The probe therefore tested the round-trip machinery M7 depends on, base-free — which removes **no** refusal, since both `project_text_document` and `serialize_document` gate on `canonical_base.is_some()`, leaving §1.2 untouched. **Result: the round trip is byte-preserving, but only from a fixed point, and round 10's comparison did not compare from one.** Round 10 compared `A` against a `B` built from the *input* document, which is valid only when that document is already a fixed point of `document_from_bundle ∘ serialize_document`. `minimal_document(42)` happens to be one — so the first probe **passed**, and would have been reported as success — while `minimal_document(99)` was not, and the **one-extension case diverged by 295 bytes from offset 352**. Rebuilt from the fixed point, all three cases are byte-identical (1641 / 1800 / 1894). **The non-idempotent field is `envelopes`, not extensions:** diagnosed field-by-field, `document_id`, `manifest_schema_version`, `lineage_id`, `profiles`, `canonical_base`, `blobs` and `extensions` — including every `TextChunk` payload — survive exactly, while `document_from_bundle` applies a **canonical envelope ordering** (its own test says so), so any other arrival order is not a fixed point and its operation-block bytes differ. **`project_text_document` → `parse_document` proved LOSSLESS** (`b_doc == d` in every case) — the text leg was never the problem; the defect was entirely in which artifact round 10 chose as reference. **What M7 must add, for round 11 to ratify rather than for the probe to assume: an explicit fixed-point normalisation and assertion before any byte comparison, because otherwise a mismatch is round 10's own unclassifiable "third category".** Probe hygiene: the comparison was **mutation-verified** — a different `FileUuid` for `A` produced 20 differing bytes at offsets 32–47 and 60–63, observed, then restored by hand-editing, incidentally confirming round 9's point that `FixedHeader.file_uuid` is byte-visible and round 8's enumeration had omitted it; one file touched, 142 insertions, all inside `#[cfg(test)]`; no refusal removed; no canonical base carried; diff captured, branch deleted. **The methodological result: four paper rounds refined this comparison and none found that it silently depended on an unstated precondition. One execution found it in minutes, via the case a reviewer would least likely hand-pick — a document with an extension. Had the probe stopped at the case round 10 implied, the contract would have been ratified on a comparison that fails for most documents.** **REVIEW ROUND 11, 2026-08-08, INDEPENDENT, against `39f2617` (post-probe): three findings, two blocking. It confirmed the probe contained and its fixed-point result decisive, and kept M7 BLOCKED.** (1) **M7 still lacked a distinct normalised reference.** Round 10 named one artifact where the comparison needs two: build **`B_raw`** under the real authority, then **iterate derive-and-reserialize until `B_fixed` is a byte-level fixed point**, **assert that property explicitly as a hard failure**, and **compare the imported artifact only with `B_fixed`, never with `B_raw`** — otherwise an envelope-order normalisation difference remains **indistinguishable from a provenance result**, and a comparison whose failure mode cannot be told from its success condition decides nothing. Steps 1a–1c added, including a bounded convergence loop (the probe saw one pass suffice for three documents, which is not proof that one pass always suffices) and a required report of the iteration count and whether `B_raw` was already fixed. (2) **The claim was stated more broadly than any observation supports.** M7 read as though every direct bundle is byte-identical to its re-imported form; it is not, and the probe measured 295 differing bytes proving so. The contract now scopes it: M7 proves **the text path carries no provenance marker *after normalisation***, and explicitly **not** that every direct bundle is byte-identical before it — the pre-normalisation differences are `document_from_bundle`'s canonical envelope ordering and have nothing to do with provenance. **Both sentences must appear in the rung's report.** **This finding has consequences beyond M7: its conclusion is the sole evidence for a permanent capability loss — the text refusal that moved `COMPANION_VERSION` to 0.14.0 and took the corpus's `canonical_bases` from 2 to 0 — so justifying a permanent refusal from a claim broader than the result obtained is the same error as concluding instead of observing, one level up: not a false observation, but a true one asked to carry more than it can.** (3) Clarification rather than defect: **the probe cannot pre-verify M7's authority/base leg**, which needs `BundleCapabilities`, `capabilities()` and pin 3a's validation — S27's own deliverables — so it remains an **execution requirement after S27 implementation**, with the probe standing as evidence for the prerequisite and explicitly **not** as a demonstration of laundering, since it carried no base. Recorded as a standing prerequisite table: the round-trip leg is settled, the authority leg is not pre-verifiable by any review or probe. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1, 2. Eleven rounds, none clean. Round 11 broke the falling trend, and did so because the probe supplied evidence that made a previously invisible defect findable — a reason to expect the next round to find more rather than less. The M7 comparator is on its fifth design: four were falsified by reading, the fifth by execution and then rebuilt on that evidence. It is the first with a measured result behind it and the first whose precondition is asserted rather than assumed, and it still cannot be executed end to end until S27 is implemented.** **REVIEW ROUND 12, 2026-08-08, INDEPENDENT, against `74dc994`: two findings, both blocking, and both the same defect — a requirement stated without the decision it requires, leaving execution to make a design choice silently.** (1) **The convergence loop was not actually bounded**: it demanded a bound and named no limit, so execution would have chosen when non-convergence becomes failure, changing what the experiment means. **Now pinned at one normalising step** — with `B₀ = B_raw` and `B₍ₙ₊₁₎ = serialize_document(document_from_bundle(Bₙ), uuid)`, compute at most `B₁` and `B₂`, permitted maximum `n = 1`, with a three-row outcome table (`B₁ == B₀` → already fixed; `B₁ != B₀` and `B₂ == B₁` → `n = 1`, the expected case; **`B₂ != B₁` → HARD FAILURE**, reporting all three image lengths and the first differing offset). **The bound is one step because it is a property, not a tolerance:** `document_from_bundle` canonicalises, so `serialize_document ∘ document_from_bundle` must reach its canonical form in a single application, and if it does not there is **no canonical form**, no principled reference artifact, and **M7 is invalid as a whole** — a finding about the projection rather than a signal to iterate further. A loop that runs until it happens to settle tests nothing; it reports how long it took. Raising the bound needs its own amendment and review round. (2) **M7's location was unchosen**: "in a crate that can reach the real constant" is true of two crates and decisive for neither, and `render_text_document` is **`pub(crate)` to `epiphany-textproj`** (`project.rs:595`), so `epiphany-testkit` could host M7 only via **an unpinned visibility change to another crate's public API**. **The harness is now pinned to `epiphany-textproj`**, which alone has both the renderer and (via its `epiphany-ops` dependency) the real constant — under **existing touch row 9**, no new row. **`render_text_document` stays `pub(crate)`:** handoff §1.3 records it as *the one intentional hole* in the text refusal, existing solely so a negative vector can carry the spelling it asserts is refused, and widening it to host a mutation that gets reverted would leave a permanently widened public surface behind — which is how a temporary harness becomes an API change nobody ratified. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1, 2, 2. Twelve rounds, none clean. The last two rounds found the same kind of defect — a requirement that reads as a decision but is not one — so the next scan should hunt remaining instructions that name a constraint without naming its value. Everything M7 now specifies is pinned to a number, a crate or a named artifact, which is a checkable property a round can test directly** |
+| P13-S27 | **The reduction-algorithm-version machinery is self-referential, so the one check that would detect a canonical-semantics change necessarily passes.** `core_spec.tex:11614`–`:11617` is normative — *"Snapshots produced under an earlier algorithm version cannot be used as canonical bases under a later one without rebuilding"* — and `:14369`–`:14372` states that replicas at differing versions *"may produce different canonical states from the same operation set."* The machinery to enforce it appears to exist: `ReductionAlgorithmVersion` (`bundle/src/ids.rs:291`) is a superblock wire field (bytes `68..72`, `superblock.rs:20`); `reduction_version_for` (`bundle.rs:989`) sets a new superblock's value; and `open` (`bundle.rs:396`–`:399`) rejects a mismatch. **But the writer sources the value from the canonical base's own self-report** (mapping the base's `reduction_algorithm_version` through `unwrap_or_default()`), **and the reader compares it only against the superblock that value seeded.** Nothing compares either against the semantics the running implementation actually implements. **The check is not vacuous** — it catches a corrupt or tampered base whose version disagrees with its superblock — but it **necessarily passes for a conformingly propagated stale base**, which is precisely the case the requirement exists to prevent. Supporting: **no constant or accessor anywhere names 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 in that crate — **a second instance of P13-S26's pattern**, a doc comment asserting a false fact about another module | `spec/CONTRACT_P13S16_PROJECTION.md` pin 0 (found 2026-07-31 while scoping P13-S16, which is a canonical reduction-semantics change and therefore the first rung to need this guarantee; filed in the same ledger edit as the row it blocks) | **UNBLOCKED 2026-08-07 — the format-epoch rung landed; dispatchable, and still blocking P13-S16.** (Was: open, BLOCKED on P13-S28.) **Scoped 2026-07-31 as `spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md` (DRAFT, not dispatchable).** Rulings taken: a typed `BundleCapabilities` required at both `Bundle::open` and `Bundle::create` and carried on the `Bundle` — no default, so every caller states the semantics it implements — and outright rejection on mismatch via a new `CanonicalBaseRequiresRebuild` error, not read-only and not an integrity anomaly. Storing the capability keeps all 57 `commit` sites unchanged; only `open` (57 sites) and `create` (32) move. **The scoping also falsified this row's first reading that the writer path was test-only:** `epiphany-textproj`'s `serialize_document` (`serialize.rs:119`) and `project.rs:936` are production paths that copy a base's `reduction_algorithm_version` verbatim into a fresh `SnapshotRef`, which `commit_versioned` then stamps into the superblock (`bundle.rs:798`) — so production mints self-consistent stale documents **without ever calling `open`**, and the capability must govern writers too. **What blocks it:** contract pin 2a. Baseline authority `0` does not preserve the corpus (`serialize.rs:327` stamps `1` and round-trips it; `vectors.rs:353`/`:363` likewise), and once P13-S16 moves the authority to `1`, a pre-S27 base that happens to carry `1` is **indistinguishable from a legitimately rebuilt one** — a raw `u32` carries no provenance. Four dispositions are recorded there; `FORMAT_MINOR` as a provenance carrier was proposed and **rejected** (the header never changes after creation, `core_spec.tex:10799`, so a legacy bundle committing a freshly validated base keeps its old minor forever; and a minor change may only append append-safe discriminants, `:12258`, not alter acceptance semantics). The surviving requirement — provenance must ride a container property **old readers cannot silently accept** and **a later commit cannot inherit unchanged** — is a format-epoch design, filed as **P13-S28**. **Scope of the claim, deliberately narrow:** this establishes that the **current implementation** has no detection mechanism. It does **not** establish that no reduction-semantics change in the project's history was ever detectable — that needs a history audit not yet done, and the stronger sentence is deliberately not written here. **What closing it requires:** an authority naming the semantics this build implements, and a rejection-or-rebuild path when a base disagrees with it. Until then any rung changing canonical reduction semantics can record its break in prose but cannot make stale bases unusable — which is why P13-S16's contract is complete, ratifiable as a plan, and **not dispatchable**. **Method note:** an earlier draft of S16's pin 0 claimed no writer path existed at all. That was false, and the way it was false is the point — the search behind it looked for `ReductionAlgorithmVersion(` constructor calls, which 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. **UNBLOCKED 2026-08-07:** the format-epoch rung landed and its pin 8 **resolves pin 2a** — reduction-version authority is meaningful only in major-1 containers, so legacy bases are refused by container epoch and never by version arithmetic. The collision pin 2a identified never has to be adjudicated: a pre-S27 base carrying `1` and a rebuilt S16 base carrying `1` are indistinguishable as numbers but can never meet, because the former exists only in a major-0 container, refused at the epoch boundary before any version is compared. The `u32` never has to carry provenance because the container does. **S27 now additionally owes three inherited items** (both interim refusals converted to validation, M8's deferred laundering demonstration, pin 3c's two suspended conformance assertions), recorded in its contract as required tests. **~~RATIFIED 2026-08-07 after review round 1~~ — RATIFICATION WITHDRAWN 2026-08-07, see round 2 below. Current status: NOT RATIFIED, NOT DISPATCHABLE, pins NOT frozen, **awaiting the next independent review round. Which rounds have closed and the running tally live in the contract's own history table and are deliberately NOT restated here — this clause carried a count through two rounds and went stale in both, and the round list went stale the same way**; no execution work may begin.** Round 1 was run because this contract had reached "dispatchable" with **zero** ratification rounds on record, against the standing rule that contracts go through adversarial review before dispatch. Round 1 returned **nine findings, four blocking**, all now carried in the contract. **Correcting this row's own preceding clause:** the three inherited items were *not* all "recorded as required tests" — obligation 2, M8's laundering demonstration, appeared in **neither** the contract's test section nor its mutation plan, while that section's preamble claimed all of them were tests. It is now **M7**, and ruled a **mutation** rather than a capability restoration: the format rung's text refusal is permanent, `COMPANION_VERSION` stays 0.14.0, and the text-projection corpus keeps `canonical_bases` reach 0. The other blocking three: §0.4's `commit`-site count counted a same-named method in `epiphany-editor-core`, **a crate with no `epiphany-bundle` dependency at all** — the third instrument failure recorded in that one section; three independent stale list-counts (the test-section header, gate 1, and three report items) all naming figures the document had outgrown; and `testkit/tests/requirement_labels.rs` missing from the touch table while pin 9 may move `CORE_REQUIREMENT_COUNT` 213 → 214 — **the escapee `CLAUDE.md` names by name**, which also escaped the format-epoch rung. Non-blocking: locator drift since `381c498` (`bc06706` grew `bundle.rs` by 338 lines; pin 5's own `:396`–`:399` confirmed unmoved), pin 2a's corpus evidence superseded by the 2 → 0 rebuild, `Bundle::open(` 57 → **60**, gate 6a's scope widened to `epiphany-testkit`, and a missing **commit-side positive** test now added as test 8. **REVIEW ROUND 2, 2026-08-07, against the frozen contract: six further findings, four blocking — and round 1's ratification was therefore premature.** (1) The call-site correction had been applied to §0.4 only, leaving the "Rung type" paragraph at **57** and touch row 2 claiming `bundle.rs` has **35** opens — a figure that was never `bundle.rs` alone and is stale besides — which made the required reconciliation impossible. (2) §0.4 called `project.rs:936` a **production** bundle writer; `#[cfg(test)]` starts at `:630` and every `Bundle` call in that file is below it, so the writer-path correction stands on `serialize.rs` alone. (3) **M5 was unexecutable**: `serialize_document` refuses bases at `serialize.rs:151`, so its output is base-free, and pin 5 with test 4 require base-free bundles to open at *any* authority — split into **M5a**/**M5b**. (4) **M6's second half was unexecutable**: `open` rejects a stale base, `create` rejects a base-bearing manifest (`bundle.rs:234`), and `commit` validates what it emits, so no caller can hold an open `Bundle` with a stale *inherited* base — replaced by broadening rather than narrowing, with the unreachability itself reported as the stronger result. Non-blocking: pin 3a's justification (*"production code mints a self-consistent stale document"*) is **false in-tree** — zero production paths stage a base — so it now rests on guarding the public `commit_versioned` API; and `serialize.rs:157` is dead code orphaned by the `:151` guard, recorded and explicitly **not** repaired here. **Two of round 2's findings were introduced by round 1** — ruling M7's refusal permanent is what made M5 unexecutable, and test 8 was added without re-deriving M6 against the same reachability. **Method note: an amendment is a change to the system, not a patch to a line**; a round must re-derive every mutation against every ruling the previous round made. **Round 3 is warranted before dispatch — the defect rate has not fallen (9, then 6), and "dispatchable" is a claim requiring evidence of convergence rather than a status reached by running out of findings.** **REVIEW ROUND 3, 2026-08-07, INDEPENDENT, against `b842975`: six findings, four blocking — and every blocking finding was a defect in text rounds 1 and 2 wrote.** (1) **Pin 3a still carried the rationale round 2 retracted** — §0.4 states there is no in-tree production base writer while pin 3a still said "§0.4 shows production code minting a stale document", so the contract asserted a claim and its negation; **the third occurrence of fix-one-site-leave-the-others**. Rewritten onto the footing that survives: `commit`/`commit_versioned` are **public API** and guard out-of-tree callers, not an in-tree path. (2) **M5a had no observation mechanism** — pin 3 required the capability be *stored* and nothing exposed it; `Bundle` carries 17 public accessors and none for capabilities, so no `epiphany-textproj` test could inspect it. **`Bundle::capabilities()` is now pinned** — new scope, flagged for round 4. (3) **M5b could not fail**: if the supplied capability and the base version both derive from `CURRENT_REDUCTION_ALGORITHM_VERSION` — the natural implementation, since `roundtrip.rs:367` currently hardcodes `ReductionAlgorithmVersion(0)` — both operands move together and the comparison passes for every value. **This is §0.1's own tautology reproduced inside the mutation built to detect it.** The base version must now come from a source that does not track the authority (persisted artifact or deliberate literal), with both operands' provenance reported. (4) **M6's replacement named a scenario with no test** — test 6 stops at opening, so nothing asserted that an unrelated commit *succeeds*, and an implementation rejecting every post-base commit passed tests 2/5/6/8 while the broadening had nothing to break; **test 9 added**. Cleanup: touch row 7 listed `generators.rs` as "call sites, real authority" though it has **zero** `Bundle::open`/`create` calls and its `rng.range(0, 8)` versions are precisely the arbitrary wire values pin 3b assigns to *synthetic* capabilities — split to **row 7a**; and §7's call-site attribution credited round 1 alone where rounds 1 and 2 are both load-bearing. **The pattern is now legible and it is not about counts: three separate mutations were unrunnable in three different ways — M5a could not observe, M5b could not fail, M6 had nothing to break. §7 item 4a now requires, for every mutation, the named test it breaks and the provenance of each operand.** **Defect rate across three rounds: 9, 6, 6 — not converging.** The newest text (pin 3's accessor, M5a, M5b, test 9, row 7a) has had **zero** adversarial passes and was written by the same agent whose previous two attempts round 3 falsified. **REVIEW ROUND 4, 2026-08-07, INDEPENDENT, against `53292f6`: five findings, four blocking.** It accepted pin 3's `capabilities()` accessor as **bounded** — the first new text any round has passed — and found the M5 pair defective a third time. (1) **M5b cited the wrong value**: `roundtrip.rs:367` sits in `assert_score_serialization_stable` (`:332`) and versions an **acceleration snapshot**, not a canonical base, while `assert_reduction_serialization_stable` has **no base at all** because pin 3c suspended it — so the value round 3 warned the implementer not to touch was irrelevant to the authority check. **The tautology diagnosis stands; only its evidence was wrong.** (2) **The instrument was left unchosen** — round 3 said "the rung picks one" and offered two, one of which does not exist for the nominated crate, since `craft_image_with_base` is a private `fn` inside `epiphany-bundle`'s `#[cfg(test)]` module (`:1648`). **Now chosen: commit-then-reopen through public API only** — build with `synthetic_for_fixture(0)`, commit a base carrying the literal `0`, reopen those bytes under the real constant. (3) **No test could assert the error fields**: `assert_reduction_serialization_stable` returns `()` and reopens with `.expect` (`:292`), so a mismatch panics rather than yielding a matchable `CanonicalBaseRequiresRebuild { base, current }`. **Test 10b added.** (4) **M5a violated §7 item 4a, the rule round 3 added in the same edit** — it named no test, and its natural assertion compares the constant with itself and cannot fail. **Test 10a added, asserting against a deliberate literal.** **Round 3's error is the one to carry: it grepped `ReductionAlgorithmVersion`, saw a `roundtrip.rs` hit, and attributed it without resolving the enclosing item — the same shape as §0.4's `.commit(` miscount that round 1 had already recorded as a lesson. Recording a defect is not the same as not committing it.** Both literals in tests 10a/10b are **load-bearing as literals**; §7 item 4b now requires confirming neither was tidied into the constant, a failure mode invisible to the suite. **Defect rate: 9, 6, 6, 5 — still not converging after four rounds, and every blocking finding in rounds 3 and 4 was in text written to fix the previous round.** **REVIEW ROUND 5, 2026-08-08, INDEPENDENT, against `df9e528`: four findings, two blocking — the first round in which blocking findings fell below four.** (1) **The status history was numerically stale again** — "amended three times … fifteen findings so far, eight blocking" were the round-2 figures, left standing through rounds 3 and 4 **while the tables recording those very rounds sat directly below them**. This is the **fifth** count-staleness defect in five rounds, and it was in the one block the author edited every round. Replaced with a **table**, so a round appends a row rather than requiring a number to be found and re-derived. (2) **Test 10b could not make the two-field assertion M5b requires**: §3 said only "assert it opens", and under mutation that yields a bare `Err` or a panic — **a `#[test] -> Result` that returns `Err` asserts nothing about that error's fields**, so M5b's required observation had no home in the test M5b names. Both `Result` arms are now pinned, plus a third for the wrong-error case, so the mutation run produces a *verified* observation rather than a stack trace. Smaller: **M5b's "cannot be tidied" claim was false** — keeping `synthetic_for_fixture` while passing `CURRENT_REDUCTION_ALGORITHM_VERSION` as both its argument and the base version preserves the fixture and fully restores the tautology, so the structure does not protect itself and the real protection is §7 item 4b; round 4 asserted a structural guarantee that undercut the procedural check actually doing the work, **which is the same error as reasoning that a mutation would fail instead of running it**. And §3's preamble still said the tests were "in `epiphany-bundle`" after round 4 added two that **cannot** be, since `epiphany-bundle` must not depend on `epiphany-ops` and reaching the real authority is their entire purpose — corrected, with each test's touch-table home named. **Blocking findings by round: 4, 4, 4, 4, 2 — the first movement in four rounds and the first weak evidence of convergence, set against the fact that every round since the third has found blocking defects in text written to fix its predecessor.** **REVIEW ROUND 6, 2026-08-08, INDEPENDENT, against `03c85dd`: three findings, ALL THREE BLOCKING, and all three in text round 5 wrote.** (1) **The amendment tally went stale inside the block round 5 restructured to prevent exactly that** — round 5 turned the review totals into a table and left "amended five times … rounds 1–4" as prose immediately above it. The amendment count is now **the number of rows**, with no separate figure to go stale. (2) **§3's test-home correction was itself false**: round 5 wrote "tests 1–9 in `epiphany-bundle`", but **test 7 *is* `assert_reduction_serialization_stable`**, which the same section names as `testkit/src/roundtrip.rs`. Two wrong versions of that sentence, both written while fixing it; replaced with a per-crate table (1–6/8/9 bundle, 7 and 10b testkit, 10a textproj). (3) **§7 item 4b protected one operand where test 10b has two** — replacing **both** `synthetic_for_fixture(0)` **and** the committed base's `ReductionAlgorithmVersion(0)` with the constant **keeps the synthetic call in place** and fully restores the tautology, and **test 10b's `Err` arm never executes in the unmutated run**, so its literal cannot detect it. Item 4b now enumerates all **three** fixture operands individually and requires each quoted verbatim. **All three findings are one defect in different clothes: a fix applied to the site named rather than to every site the claim covers** — the sixth count-staleness defect in six rounds and the third range-correction that did not check its own range. **The mechanism that works is structural, not vigilant: the review totals stopped going stale when they became a table, the amendment count did not because it stayed prose, and item 4b stopped being under-specified when it became a table.** **Demonstrated a seventh time inside round 6's own amendment**, where the new table's Total row was first written "6 amendments" — a free-standing count, three paragraphs after the sentence declaring no such count exists, and already wrong at seven rows; caught before commit and replaced with "one amendment per row". **Prose invites a number and a table does not; the defence must be the shape of the artifact, not the attention of the editor.** **REVIEW ROUND 7, 2026-08-08, INDEPENDENT, against `c0d896c`: three findings, all blocking, and all three the same defect — a claim living in two places and fixed in one.** (1) **§7 item 4a was unsatisfiable**: it required every mutation to name "the test it breaks", while item 1 four paragraphs above states that **M4 is observed to *compile*** (no test is possible — that is why pin 3's prohibition is a review rule) and **M7's expected outcome is *success***. A report obeying 4a literally could not be written, and the honest response would have been to invent a test. 4a is now a table of what each of the eight mutations owes, with M4 and M7 carved out explicitly. (2) **Round 6's three-literal correction reached §7 and not §3** — §3 still said "**both** literals … tidying **either**", so the contract carried the fixed and the broken version of the same claim, reopening exactly the narrow-scope ambiguity round 6 existed to close. §3 no longer states the count at all; it points at item 4b. (3) **"Rounds 3, 4 and 5 were independent"** went stale the instant round 6 closed, sitting in prose beside the table whose own column records it. Deleted. **Three rounds, one lesson: round 5 fixed the review totals and not the amendment tally beside them, round 6 fixed item 4b and not §3's copy of the same rule, round 7 found the classification sentence duplicating the table's column. The defect is duplication, and every previous remedy was vigilance — "check the other sites too" — which has now failed three rounds running. The remedy adopted here is deletion, not diligence: where a claim had two homes, one is removed and replaced with a pointer. A copy that cannot drift is one that does not exist.** **Findings by round: 9, 6, 6, 5, 4, 3, 3 — flattened rather than still falling. Blocking: 4, 4, 4, 4, 2, 3, 3 — rounds 6 and 7 were both 100% blocking and 100% in the previous round's text. Seven consecutive rounds, no clean round yet. The deduplication is the first structural remedy for this particular defect and therefore the first with a reason to work, but it is untested.** **REVIEW ROUND 8, 2026-08-08, INDEPENDENT, against `9829ae3`: two findings, both blocking — and the first round to reach into a mutation's mechanics rather than its bookkeeping.** (1) **M7 did not describe a runnable observation.** It instructed execution to *construct* a base-bearing `TextDocument`, which **bypasses `parse_document` entirely**, so the parser refusal it ordered removed was irrelevant and the demonstration was not the **import** laundering it is named for; `project_text_document` is the **export** direction (`&TextDocument -> Result<String, _>`) and is not on the path at all, so "all three sides, since removing one leaves the others refusing and the document never reaches the writer" was simply false for it; and "byte-indistinguishable from one whose base was genuinely validated" named **no comparison artifact and no comparison method**, leaving the central claim a conclusion rather than an observation. Now: the input must be **text and must be parsed**; only the parser (`parse.rs:138`–`:147`) and serializer (`serialize.rs:151`) refusals are removed and restored; the comparison artifact is **test 10b's construction** with the same `FileUuid` and base bytes; and the comparison is a **field-by-field enumeration** of the `canonical_base` `SnapshotRef`, the superblock's reduction version and the header's major/epoch, **reported rather than concluded** — informative in both directions, since a field that *does* differ is a provenance signal nobody knew existed. (2) The round-7 deduplication was incomplete: the status block still carried "rounds 3 and 4 are closed" while declaring the history table the sole authority. **Finding 1 is the most substantive of any round, because every earlier one was about text agreeing with other text — this one is about whether the experiment runs at all, and it did not. M7 had been in the contract since round 1 and survived seven reviews, three of which specifically re-derived mutations, because reading it never required tracing what calls what. An observation stated in the right register can look complete for a long time; "indistinguishable" was a conclusion sitting inside the rung's own demonstration, which is the exact failure mode this rung exists to eliminate.** **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2. Eight rounds, none clean. While amending, the author caught a third instance unaided — §7 item 4a's M7 row still said "all three refusals" — which is weak evidence the deduplication rule is being applied rather than merely stated. The M7 rewrite is now the newest and least-reviewed material in the contract, and its predecessor survived seven rounds while being unrunnable.** **REVIEW ROUND 9, 2026-08-08, INDEPENDENT, against `01e76d1`: two findings, both blocking, both in M7's comparator — the text round 8 had just rewritten.** (1) **Test 10b is not the "genuinely validated" reference M7 nominated**: its write-side capability is `synthetic_for_fixture(0)` and only its *reopen* uses the real authority, so M7 would have compared one synthetic fixture against another with the validated half of the claim simply absent. **This is a collision between two of the contract's own designs, not a typo** — round 4 made 10b synthetic-on-write *deliberately* so M5b's operands would be provably independent, and that is exactly what disqualifies it here. **One artifact cannot be both independent of the real authority and committed under it.** M7 now builds its own reference in `epiphany-testkit`, committing a base under `caps` derived from the real constant so pin 3a validates it on the way in. (2) **The field enumeration could not support its conclusion**: it claimed "everything that could carry provenance" while omitting `FixedHeader.file_uuid` — **the field it required to match** — plus the superblock's `generation`, `manifest_offset`, `manifest_length` and `manifest_hash`, and the whole manifest outside `canonical_base`. Replaced with **whole-`image()` byte comparison**, any difference enumerated and classified as justified nondeterminism (normalize, stating why) or as a **provenance signal** (a finding, since the refusal may then be stronger than needed). **Finding 2 retires a technique rather than an instance: a hand-written list of "every field" is a claim about a struct's contents that is wrong the moment the struct changes, and this one was wrong the day it was written. Comparing the whole artifact cannot be incomplete — the tables-over-numbers lesson applied to the experiment instead of the prose.** Three further sites were caught by the author while amending: §7 item 6 still said "M7's three text refusals", surviving round 8's correction of that exact count in two other places; §7 item 4a's M7 row still named the superseded method; and round 8's own disposition cell stated it as current. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2. Nine rounds, none clean. Rounds 8 and 9 both found defects in the immediately preceding round's rewrite of the same paragraph, so M7 has now been wrong in three distinct ways across three consecutive rounds — unrunnable, then wrong-artifact, then wrong-method. The comparator is on its third design and has never been executed.** **REVIEW ROUND 10, 2026-08-08, INDEPENDENT, against `0efd543`: one finding, blocking — the smallest round yet, and again in M7.** **The whole-image comparison had no complete construction alignment.** Round 9 named four things to align, but `serialize_document` also fixes `document_id`, `lineage_id`, `profile_declarations`, every extension's fields and preserved chunks, the envelope payloads, the **staging order** (base root → extension chunks → operation-envelope block), the manifest schema `major` and `epoch_max`, and every chunk ref, hash and offset derived from those. **So a byte difference would have had a third possible cause — "the reference was built differently" — which is neither permitted classification; the result would have been unclassifiable and the comparison meaningless. A result that cannot be classified is not an observation.** M7 is now a **round trip**: build `B` validated under the real authority, export it to text via `document_from_bundle` + the crate-private `render_text_document`, parse that text back, re-serialize as `A` with `B`'s `FileUuid`, and compare whole images. **Alignment is inherited rather than enumerated** — every input `serialize_document` reads is already `B`'s own, so no list can be incomplete and the setup-mismatch category is eliminated by construction rather than by care. It is also the realistic form of the threat: export a validated document to text, re-import it, and observe the re-imported container is indistinguishable from the original having validated only the base's number, never its provenance. **This was the third hand-enumerated "complete set" in this contract and the third wrong on the day it was written — "every field that could carry provenance" (round 8), "every field to align" (round 9), and round 9's list again. The single rule earned across rounds 5–10: where a claim requires completeness, do not enumerate, derive. Tables instead of counts, whole artifacts instead of field lists, one shared origin instead of an alignment list.** No refusal count is stated anywhere in M7 any more; three successive wordings each had a wrong one. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1. Ten rounds, none clean, and three consecutive rounds have found one paragraph — M7 — defective in a new way each time: unrunnable, wrong artifact, wrong method, incomplete alignment. Findings are falling steadily and each of the last three has been narrower than the last, the first sustained convergence signal here. Against that: M7 has never been executed and each of its four designs looked correct when written — the open question for round 11 is whether the next defect is findable by reading at all, or whether M7 must be run against a scratch branch before further paper review can add anything.** **BOUNDED SCRATCH PROBE, 2026-08-08, authorised as an explicit narrow exception in the contract's status block and run on a discarded branch: it FALSIFIED round 10.** First, recorded as evidence in its own right: **M7 cannot be executed at all until S27 lands** — `BundleCapabilities` and `CURRENT_REDUCTION_ALGORITHM_VERSION` do not exist in the tree, being S27's own deliverables, and M7 step 1 needs a base committed under the real authority. M7 is a mutation *of this rung's implementation*, so it runs after the rung. The probe therefore tested the round-trip machinery M7 depends on, base-free — which removes **no** refusal, since both `project_text_document` and `serialize_document` gate on `canonical_base.is_some()`, leaving §1.2 untouched. **Result: the round trip is byte-preserving, but only from a fixed point, and round 10's comparison did not compare from one.** Round 10 compared `A` against a `B` built from the *input* document, which is valid only when that document is already a fixed point of `document_from_bundle ∘ serialize_document`. `minimal_document(42)` happens to be one — so the first probe **passed**, and would have been reported as success — while `minimal_document(99)` was not, and the **one-extension case diverged by 295 bytes from offset 352**. Rebuilt from the fixed point, all three cases are byte-identical (1641 / 1800 / 1894). **The non-idempotent field is `envelopes`, not extensions:** diagnosed field-by-field, `document_id`, `manifest_schema_version`, `lineage_id`, `profiles`, `canonical_base`, `blobs` and `extensions` — including every `TextChunk` payload — survive exactly, while `document_from_bundle` applies a **canonical envelope ordering** (its own test says so), so any other arrival order is not a fixed point and its operation-block bytes differ. **`project_text_document` → `parse_document` proved LOSSLESS** (`b_doc == d` in every case) — the text leg was never the problem; the defect was entirely in which artifact round 10 chose as reference. **What M7 must add, for round 11 to ratify rather than for the probe to assume: an explicit fixed-point normalisation and assertion before any byte comparison, because otherwise a mismatch is round 10's own unclassifiable "third category".** Probe hygiene: the comparison was **mutation-verified** — a different `FileUuid` for `A` produced 20 differing bytes at offsets 32–47 and 60–63, observed, then restored by hand-editing, incidentally confirming round 9's point that `FixedHeader.file_uuid` is byte-visible and round 8's enumeration had omitted it; one file touched, 142 insertions, all inside `#[cfg(test)]`; no refusal removed; no canonical base carried; diff captured, branch deleted. **The methodological result: four paper rounds refined this comparison and none found that it silently depended on an unstated precondition. One execution found it in minutes, via the case a reviewer would least likely hand-pick — a document with an extension. Had the probe stopped at the case round 10 implied, the contract would have been ratified on a comparison that fails for most documents.** **REVIEW ROUND 11, 2026-08-08, INDEPENDENT, against `39f2617` (post-probe): three findings, two blocking. It confirmed the probe contained and its fixed-point result decisive, and kept M7 BLOCKED.** (1) **M7 still lacked a distinct normalised reference.** Round 10 named one artifact where the comparison needs two: build **`B_raw`** under the real authority, then **iterate derive-and-reserialize until `B_fixed` is a byte-level fixed point**, **assert that property explicitly as a hard failure**, and **compare the imported artifact only with `B_fixed`, never with `B_raw`** — otherwise an envelope-order normalisation difference remains **indistinguishable from a provenance result**, and a comparison whose failure mode cannot be told from its success condition decides nothing. Steps 1a–1c added, including a bounded convergence loop (the probe saw one pass suffice for three documents, which is not proof that one pass always suffices) and a required report of the iteration count and whether `B_raw` was already fixed. (2) **The claim was stated more broadly than any observation supports.** M7 read as though every direct bundle is byte-identical to its re-imported form; it is not, and the probe measured 295 differing bytes proving so. The contract now scopes it: M7 proves **the text path carries no provenance marker *after normalisation***, and explicitly **not** that every direct bundle is byte-identical before it — the pre-normalisation differences are `document_from_bundle`'s canonical envelope ordering and have nothing to do with provenance. **Both sentences must appear in the rung's report.** **This finding has consequences beyond M7: its conclusion is the sole evidence for a permanent capability loss — the text refusal that moved `COMPANION_VERSION` to 0.14.0 and took the corpus's `canonical_bases` from 2 to 0 — so justifying a permanent refusal from a claim broader than the result obtained is the same error as concluding instead of observing, one level up: not a false observation, but a true one asked to carry more than it can.** (3) Clarification rather than defect: **the probe cannot pre-verify M7's authority/base leg**, which needs `BundleCapabilities`, `capabilities()` and pin 3a's validation — S27's own deliverables — so it remains an **execution requirement after S27 implementation**, with the probe standing as evidence for the prerequisite and explicitly **not** as a demonstration of laundering, since it carried no base. Recorded as a standing prerequisite table: the round-trip leg is settled, the authority leg is not pre-verifiable by any review or probe. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1, 2. Eleven rounds, none clean. Round 11 broke the falling trend, and did so because the probe supplied evidence that made a previously invisible defect findable — a reason to expect the next round to find more rather than less. The M7 comparator is on its fifth design: four were falsified by reading, the fifth by execution and then rebuilt on that evidence. It is the first with a measured result behind it and the first whose precondition is asserted rather than assumed, and it still cannot be executed end to end until S27 is implemented.** **REVIEW ROUND 12, 2026-08-08, INDEPENDENT, against `74dc994`: two findings, both blocking, and both the same defect — a requirement stated without the decision it requires, leaving execution to make a design choice silently.** (1) **The convergence loop was not actually bounded**: it demanded a bound and named no limit, so execution would have chosen when non-convergence becomes failure, changing what the experiment means. **Now pinned at one normalising step** — with `B₀ = B_raw` and `B₍ₙ₊₁₎ = serialize_document(document_from_bundle(Bₙ), uuid)`, compute at most `B₁` and `B₂`, permitted maximum `n = 1`, with a three-row outcome table (`B₁ == B₀` → already fixed; `B₁ != B₀` and `B₂ == B₁` → `n = 1`, the expected case; **`B₂ != B₁` → HARD FAILURE**, reporting all three image lengths and the first differing offset). **The bound is one step because it is a property, not a tolerance:** `document_from_bundle` canonicalises, so `serialize_document ∘ document_from_bundle` must reach its canonical form in a single application, and if it does not there is **no canonical form**, no principled reference artifact, and **M7 is invalid as a whole** — a finding about the projection rather than a signal to iterate further. A loop that runs until it happens to settle tests nothing; it reports how long it took. Raising the bound needs its own amendment and review round. (2) **M7's location was unchosen**: "in a crate that can reach the real constant" is true of two crates and decisive for neither, and `render_text_document` is **`pub(crate)` to `epiphany-textproj`** (`project.rs:595`), so `epiphany-testkit` could host M7 only via **an unpinned visibility change to another crate's public API**. **The harness is now pinned to `epiphany-textproj`**, which alone has both the renderer and (via its `epiphany-ops` dependency) the real constant — under **existing touch row 9**, no new row. **`render_text_document` stays `pub(crate)`:** handoff §1.3 records it as *the one intentional hole* in the text refusal, existing solely so a negative vector can carry the spelling it asserts is refused, and widening it to host a mutation that gets reverted would leave a permanently widened public surface behind — which is how a temporary harness becomes an API change nobody ratified. **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3, 2. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1, 2, 2. Twelve rounds, none clean. The last two rounds found the same kind of defect — a requirement that reads as a decision but is not one — so the next scan should hunt remaining instructions that name a constraint without naming its value. Everything M7 now specifies is pinned to a number, a crate or a named artifact, which is a checkable property a round can test directly.** **REVIEW ROUND 13, 2026-08-08, INDEPENDENT, against `bff9c9a`: one finding, blocking — and it inverted M7's result.** **M7 claimed the capability check "does not fire".** Pin 3a requires `commit`/`commit_versioned` to validate a **newly emitted** canonical base, which is exactly what **both** `B_raw` and the parsed `A` commit — so **the check fires on both paths and ACCEPTS**, because the raw version equals the real authority. **That acceptance is the laundering result:** the base is not slipped past an absent check, it is admitted by a check working correctly that cannot tell a coincidence from a rebuild. **As written, M7 was satisfiable by deleting pin 3a's writer check entirely** — yielding a passing M7 that demonstrated the exact opposite of its purpose. M7 now requires **three observations** (`A.image()` equals `B_fixed.image()`; pin 3a's validation ran and accepted on both commits; and the control) plus a **required control**: in the same run, same harness, repeat the import with a base version deliberately **not** equal to the real authority and observe the commit **REJECTED** with `CanonicalBaseRequiresRebuild`. **M7's removals are now explicitly limited to the text refusals — pin 3a is not among them and may not be weakened, being the thing under observation rather than an obstacle to it.** **This is a new failure shape worth naming: an observation satisfiable by the absence of the thing it observes.** M7's earlier defects were about being unrunnable or comparing the wrong artifacts; this one would have run, passed and reported success on a tree with the writer check removed. **"The check does not fire" cannot distinguish a check that accepts from a check that is not there, and only one of those is the finding.** **Findings by round: 9, 6, 6, 5, 4, 3, 3, 2, 2, 1, 3, 2, 1. Blocking: 4, 4, 4, 4, 2, 3, 3, 2, 2, 1, 2, 2, 1. Thirteen rounds, none clean. Round 13 is the narrowest since the probe, but it found a defect of a kind no earlier round had looked for — not "can this run?" or "does this compare the right things?" but "could this pass for the wrong reason?" — and that question has NOT been asked of M1–M6, M5a or M5b. Every mutation in §4 deserves the same check: what else, besides the intended defect, would make it pass?** |
 | P13-S28 | **No container property distinguishes a document produced under a validated reduction authority from one produced before any authority existed — and the two candidates that look like they would, cannot.** P13-S27 installs an authority and validates it at read and write time, but cannot state what to do with a canonical base that predates the authority: a raw `ReductionAlgorithmVersion` is a bare `u32` (`bundle/src/ids.rs:291`) carrying no provenance, and the text-projection parser accepts an unbounded one from a document (`textproj/src/parse.rs:591`), so no numeric convention — including a deliberately high epoch — is safe from a hand-authored or third-party document declaring it. **`FORMAT_MINOR` does not work either, for two independent reasons:** the header *"never changes after the file is created"* (`core_spec.tex:10799`–`:10800`) and `commit_versioned` publishes only a superblock (`bundle.rs:791`), so a legacy bundle that commits a base S27 just validated keeps its old minor **permanently** — rejecting minor-≤1 bases would then reject a base the authority itself accepted, and accepting them leaves S16's `1` ambiguous; and `core_spec.tex:12258`–`:12262` limits a minor change to appending append-safe discriminants and calls it backward-compatible, whereas making a previously-valid base newly rejectable is a **semantic acceptance change**, with current readers ignoring minor entirely (`header.rs:119` gates on major alone) so the boundary would bind only readers that already comply. **The requirement that survives:** provenance MUST ride a container property that **old readers cannot silently accept** and that **a later commit cannot inherit unchanged** | `spec/CONTRACT_P13S27_REDUCTION_AUTHORITY.md` pin 2a (filed 2026-07-31; the disposition S27 cannot make from inside itself) | **IMPLEMENTED 2026-08-07 (`bc06706`, fix `be244df`). Was the critical path; both P13-S27 and P13-S16 were blocked on it.** **This rung must own all five, and none may be deferred into S27:** (1) an **old-reader rejection boundary** — pre-boundary readers must fail closed rather than silently open a document whose safety check they do not run; (2) **provenance that survives commits correctly**, i.e. is not inherited unchanged by a later generation and is not lost by one; (3) **legacy-base rebuild/repack behaviour**, stated for real artifacts rather than assumed away; (4) **every writer path, including text projection** — `serialize_document`, `project.rs`, and the committed `.txt` vectors, since a text document can declare any version; (5) **the exact format-version and compatibility consequences**, most plausibly a **major**-version boundary or a generation-scoped attestation paired with an incompatibility boundary. **Not a sub-pin of S27 and must not drift into it** — S27's pin 2a carries an explicit prohibition against being amended into a disposition without its own ratification round. **Scoped and RATIFIED 2026-07-31 as `spec/CONTRACT_FORMAT_EPOCH_MAJOR1.md`** after four adversarial review rounds — 11 pins, 11 tests, 11 mutations, 15 touch rows, 7 gate items. **This row is now a dependency record only; the work lives there and P13-S28 does not execute as a Pass 13 rung.** Rulings taken: the carrier is the **format major** (`FORMAT_MAJOR` 0 → 1, `FORMAT_MINOR` 1 → 0), decoded three ways through a named `FormatEpoch` rather than a bool, with **no** generation-scoped attestation in this epoch; legacy resolves to **hard rejection, not read-only**; and an eight-row epoch matrix in which a major-0 bundle with no base may open, one carrying a base is rejected, and one attempting to *add* a base is rejected and told to repack — **the non-inheritance rule that `FORMAT_MINOR` could not express**. All five things this row required the rung to own are pinned: old-reader boundary (pin 2), commit-surviving provenance (pin 3), legacy repack (pins 4, 5), every writer path including text projection (pins 3b, 6), and the exact format/compatibility consequences (pins 1, 7). **Three findings from the review rounds that changed the rung's shape**, none of them visible at filing: (1) **it cannot stamp major 1 before S27's writer enforcement exists**, so pin 3a temporarily refuses *both* boundaries — opening a major-1 bundle already carrying a base, and committing one into it — through a third, temporary `ReductionAuthorityUnavailable` error that must name P13-S27 and must **not** name repack; (2) **text projection launders provenance straight through the boundary** (`serialize_document` stages a carried base into a fresh bundle and `build_manifest` writes it), resolved as **symmetric document-level refusal** — projection, parsing and a new dedicated `SerializeError` variant, none of which existed to be "retained" — which forces `COMPANION_VERSION` 0.13.0 → **0.14.0** and rebuilds the committed corpus to **20 vectors, ten rejection classes, `canonical_bases` reach 2 → 0**, a real and stated capability loss; (3) **corruption precedence binds in both epochs** — a corrupt major-1 base must still fail as malformed, never as the *temporary* authority error a user would reasonably retry. **IMPLEMENTED 2026-08-07** — amended once before dispatch (pin 3c, touch rows 10/11, gate 8) after reconnaissance found pin 3a's refusals reaching a conformance criterion through a file the touch table did not carry. All 11 tests landed under their contract names, all 11 mutations run and observed, workspace green at 1569. **P13-S27 is unblocked and P13-S16 remains blocked on S27** — pin 8 resolved S27's open pin 2a (legacy bases are refused by container epoch, never by version arithmetic), and S27 additionally inherits three obligations recorded in its own contract: converting **both** interim refusals to validation, M8's deferred laundering demonstration, and pin 3c's two suspended conformance assertions. **Two touch-table gaps found during execution, both of the same shape** — a `.tex` requirement addition moves hardcoded counts in `testkit/tests/requirement_labels.rs`, and a companion-version bump moves a second normative version literal spelled `version~0.13.0` rather than `(0 13 0)`; neither file was in any touch table, and the second was caught only because `requirements_name_only_this_companion_version` exists. **A third gap was caught in review, after the rung was committed:** pin 3b's projection refusal had been implemented only on the **bundle** side (`document_from_bundle`), leaving the public `project_text_document` free to emit a `(canonical-base ...)` line for a directly constructed `TextDocument` — text the parser then rejects. A projector that can produce what the parser refuses is exactly the asymmetry pin 3b exists to close, and the refusal is unreachable through a `Bundle` during the interval anyway, so the *only* reachable half was the unguarded one. The public projector now returns `Result` and refuses; a crate-private `render_text_document` retains the base spelling for the one legitimate caller, the `canonical_base_present` negative vector. **The lesson is the rung's own recurring one:** a guard placed on the path that happened to be named, rather than on every path a caller can reach |