The audit's fourth push: the biggest outstanding Phase-2 item plus the
performance gate. 793 workspace tests pass; clippy -D warnings, fmt,
and rustdoc (deny-warnings) clean; all three spec documents build with
zero undefined references.
Binary Format companion (spec/binary_format.tex, v0.1.0 — Agent J's
deliverable, 43 pages):
- Twelve chapters transcribed from the golden-locked implementation:
encoding conventions (the three prefix/endianness regimes, a
normative no-varint rule, reject-never-normalize decode discipline),
identifiers imported from the core spec's Canonical Byte-Layout
Reference, primitive value encodings, the whole-Score positional
codec ratified as the schema-major-0 wire form, operation wire
forms (envelope field order with the normative id-leads property,
the OperationPayload 0..=3 and OperationKind 0..=23 tables,
effects/conflict/anomaly/MaterializedState vocabulary), the bundle
physical layout (64-byte header, 256-byte superblock, chunk
preimages and framing, ChunkRef, manifest body order), the
operation-index payload, and the extension-blob/edit-barrier byte
forms.
- Ratifies P12-D1 (req:binfmt:opindex), P12-E1 (req:binfmt:ext-blobs),
P12-E2 (req:binfmt:condition-depth, MAX_CONDITION_DEPTH = 64
normative), and P12-E3 (req:binfmt:object-kind-open) — batch rows
struck through; discharges the provisional-codec notes in core
(P11-4), ops, and bundle (P11-D2/D4/D5) DECISIONS with ratification
cross-references.
- Pins the frozen-layout schema-evolution keystone: within schema
major 0 every positional struct layout is frozen; a field-set change
is a schema-major change with migration — formally grounding the
data-model-expansion staging decision. Open questions kept honest
in-document: SnapshotId derivation, index-refresh threshold, u64/u32
prefix unification at the next major.
- Not yet delivered from J's charter: the cross-implementation decoder
test and the wire-format fuzzer (follow-up harnesses).
F1 benches (crates/epiphany-testkit/benches/, per the F0 decision):
- criterion 0.5.1 (workspace dev-dependency; MSRV 1.77 respected with
documented transitive pins: clap 4.5.53, half 2.4.1).
- reduction bench at 1K/10K/50K envelopes with the Chapter-10 budget
(>10,000 envelopes/second cold) written in the bench as a Pass/Xfail
gate; bundle benches for the typical-edit commit (<=50 ms; measured
~14.7 ms on real disk after catching that tmpfs neuters fsync) and
the open/bootstrap read (<=200 ms; measured ~60 us).
- CI: quick budget gates in the conformance job, full gates nightly.
Subquadratic canonical_reduction_order (the F-surfaces/K-fixes
handshake, closing K's 10K-envelope acceptance gate):
- The bench documented the failure (50K at ~1.7K env/s, a 29 s cold
reduction; two O(n^2) loops); the fix replaces pair enumeration with
threshold/frontier readiness per replica plus explicit-dot dependent
lists and a stamp-tuple binary heap — O((n + sum(context)) log n),
never materializing covered pairs.
- Byte-identical order: same edge relation, same ready predicate, same
total order; the old implementation is retained as a test-only
oracle with element-for-element order-equality property tests over
fuzz sets, adversarial sets, and directed shapes (2,000-envelope
full-coverage chains, dot cycles, duplicate-id stamp ties),
mutation-tested for sensitivity.
- Measured: 1K 155K->674K env/s, 10K 12.5K->257K, 50K 1.7K->87K; all
three scale points now pass and the 50K row is promoted from Xfail.
Also: fixed nine rustdoc private/unresolved intra-doc links that had
accumulated across the pushes (the CI deny-doc-warnings job would have
failed on them).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NEs4aYiu8MXjdYdMxw8PTd
Two audit pushes whose code edits interleave line-by-line in the same
files (reduce.rs, bundle.rs, the DECISIONS logs), committed together so
the tree at every commit builds. Gate: 784 workspace tests pass, clippy
-D warnings clean, fmt clean.
Push 1 — the true MUST violations, all fixed:
- bundle: zstd read support on both read paths, output bounded by the
declared uncompressed_length, typed decompression errors, explicit
CompressedManifest rejection (zstd 0.13 workspace dep; write path
stays uncompressed per the Phase-3 deferral).
- ops: system-derived counter collision check — mint registry seeded
from the base graph, canonical-order pre-walk, halt via the new
PendingReason::HaltedBySystemCollision (discriminant 4, additive)
with transaction-atomicity and causal-dependent closure; neither
input set occupies a collided counter. canonical_pitch_bytes made
pub in core for the MUSCSPCH preimage.
- ops: Transpose skips tombstoned targets per the catalog; missing
targets still refuse the whole operation.
- ops: marker re-anchoring recorded as a RepairRecord in the
triggering operation's effect; ResolveConflict meta-conflicts name
both resolvers; base-free pitch-id freshness; reserved effect
vocabulary annotated.
- core: decomposition pre-pass honors authored attachments
(resolve_decomposition, spec-default precedence); inversion
tolerance typed as a TempoIntegration-class Tolerance.
- CONFORMANCE.md: the determinism conformance statement required by
Appendix D — all seven declarations.
Push 3 — wiring the types-only machinery:
- layout-ir/engrave: to_constrained emits real constraints (successive
notehead no-collision chains, per-glyph region containment, soft
user-break constraints); ConstraintStrength{Required, Preferred}
with strength-by-rule; Preferred violations surface as warnings, not
failures; StubSolver reworked honest-but-renderable. SVG goldens
byte-identical; snapshot constraint counts regenerated (0->90/15).
- layout-ir: to_logical projects user system/page breaks as anchored
EngravingOverrides with paired UserOverride-sourced decisions
(OverrideKind::SystemBreak/PageBreak carry TimeAnchor, ratified in
the spec alongside).
- layout-ir/ops/editor-core: edit-barrier bridge — decode mirrors for
the whole barrier tree (reject-never-normalize, NFC revalidation,
MAX_CONDITION_DEPTH = 64), golden-locked blob codec for the
ExtensionDeclaration fields, a barrier gate in apply and
apply_transaction backed by a Score oracle and real containment
contexts, and apply_unsafe recording the crossed extensions in
extensions_requiring_tombstone() for the next bundle write.
- ops: ResolveEquivocation meta-operation per the newly ratified
catalog entry — payload discriminant 3 (appended), set-level
earliest-resolve-governs promotion, ResolveConflict-mirrored
meta-conflicts, permutation-invariance fuzz; the missing golden
locks on the OperationKind/OperationPayload wire tables added.
- ops/editor-core: validation modes — ValidationMode + a non-canonical
advisory layer (validate.rs), an authoring gate before minting, and
reduction pinned as replay mode by construction (canonical bytes
untouched).
- bundle: the operation index (opindex.rs) — provisional golden-locked
payload, binary-search locate, staleness defined as full-ChunkRef
set equality against operation_roots, and the reject-and-rebuild
discipline (a defective index is never bundle corruption).
- ops: re-anchoring rule table completed — the four-key "nearest"
ordering computed from base-free ledger indices; markers re-anchor
to the nearest live event in the same staff instance (replacing the
Push-1 region-start stand-in); cue-source cascade; graphic-gesture
Events/Range/Free rows; comment and analytical-annotation orphaning.
Zero appended discriminants.
Spec enablers ratified with Push 3: catalog §ResolveEquivocation
(0.3.0 -> 0.4.0) and anchored break overrides; 16 new Pass-12 rows
filed (C1-C4, K5-K7, I4-I6, D1, E1-E5). The data-model payload
expansion (SlurKind, beam geometry, voltas, instrument bodies,
metadata) is deliberately staged to the Binary Format companion — the
positional graph codec has no value-level versioning, so filling those
structs is a schema-major break that should land once, with J.
Also carries the pre-existing editor-track increment: the atomic
tuplet overwrite (CascadeDeleteTuplets prunes decomposition
attachments naming the cascaded tuplet).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NEs4aYiu8MXjdYdMxw8PTd
One iteration of an interactive edit, wired end to end so a missing connection
between the crates surfaces here, not in a GUI: render a score to its RenderIR,
resolve a click on a notehead to its graph pitch through the hit-test map, apply a
real operation (sharpen -- a +1-chromatic Transpose) by reducing it onto the score
graph, re-render, and confirm the selection survives the relayout.
- run_edit_loop_with<S: ConstraintSolver>(base, solver) returns an EditLoopReport
(selected_pitch, selection, graph_changed, selection_preserved, render_changed);
run_edit_loop is the stub wrapper. render_with returns None for a
diagnostic-only (non-renderable) solver report, so the loop never hit-tests or
edits a layout the caller must not render.
- The click resolves like a GUI's: aim at each real notehead's centre (a notehead
glyph, not synthesized, pitch-backed) and select whatever the topmost hit there
is, taking the first aim whose topmost hit is a pitch-backed glyph -- faithful
to a click and robust to an occluding unison/chord notehead.
- Selection survival rests on the MUSCLOID layout id, a function of the pitch's
identity (PitchId), not its content: the sharpen changes the pitch's value but
not its id, so its layout object keeps the same stable id and the cursor does
not jump off the edited note. The prepass re-runs inside to_logical, so the
edited pitch is re-spelled and its accidental reflects the new value (why the
re-render reliably differs).
Tests drive the loop on valid_score_rich (one fixture and across 48 seeds, every
seed required), refuse a diagnostic-only solver layout, and -- in epiphany-engrave,
via its existing testkit dev-dep -- run it through the real Engraver across 16
seeds, proving the selection survives even as the Engraver re-spaces every glyph.
Full gate green: build, fmt, clippy, conformance scale 1.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Turn the Score -> layout-IR -> SVG pipeline from placeholder glyphs into real
music notation, rendered through the stub solver and the real Engraver alike.
to_constrained (the spacing pass) now dispatches each layout object to the
notation primitive that represents it, on a column-based spring model:
- Pitch -> notehead at its clef-relative staff position; chord pitches share
one column slot. StaffInstance -> clef glyph. Staff -> five staff-line
strokes (the bottom line its anchor, four synthesized). Pitched Event ->
stem stroke; Measure -> barline glyph; rest Event -> rest glyph.
- Phase 3 ornaments as synthesized glyphs: a spelling's full accidental stack
left of its notehead, a key signature's clef-relative sharp/flat zigzag in
the lead, and a measure's numerator/denominator time-signature digit pair.
- A tied decomposition draws one notehead/stem/rest per component (offsets
honored, not collapsed). Active clef and key resolve by time, not vector
order. The lead area reserves clef + key-signature width.
- Coverage/surjection preserved so the round-trip holds: each laid-out object
is covered by exactly one exact-provenance primitive, and derived primitives
(staff lines, components, accidentals, key/time glyphs) are synthesized from
a laid-out source. Engraving-coverage gaps (missing spelling, unbundled
glyph) are surfaced as ConstrainedLayoutIR diagnostics, not silently
defaulted. A measure depends on the time signature it displays.
The Engraver re-spaces glyphs AND the strokes that track them through one
collision-aware coordinate map (per-slot left/right bearings, pairwise
advances), so stems / barlines / staff lines stay attached and a note's
accidental never overlaps the previous note. validate() now rejects empty
spring slots -- the contract that map relies on -- and to_constrained never
emits one (slots are realized by glyph occupancy).
Bundle the genuine Bravura outlines and metrics for time-signature digits 0-9
(regenerated from the SHA-pinned font via tools/extract_bravura_outlines.py),
replacing an inconsistent hand-written placeholder.
Regions tile left-to-right (no page casting-off yet). Goldens regenerated into
recognizable notation. Full gate green: build, fmt, clippy, 574 tests,
conformance scale 1.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Address the five-finding review of M2d (1e4ab24) plus the two-finding
follow-up review; all fixes are code/test/doc only, no spec change needed
(the catalog/core-spec already classify metadata as advisory LWW).
- SetMetadata is now a true advisory LWW: it silently last-writer-wins in
canonical order and records no conflict, matching operation_catalog
§set-user-system-break. Dropped the StructuralFieldCollision, the
`last_metadata` working slot, and the `env` parameter; rewrote the
conflict test as `concurrent_differing_set_metadata_is_advisory_lww`
(no conflict, state stays clean, permutation-independent resolution).
- SetMetricGrid / SetUserPageBreak / SetUserSystemBreak share a new
`layout_region_slot` precondition backed by a `staff_based_regions`
index: the target must be live and staff-based (FreeGraphic regions have
neither a metric-grid nor a break slot). The index is read from base-free
state, so reduce() and reduce_onto() reach the same verdict for missing,
tombstoned, and FreeGraphic regions.
- SetMetricGrid now rejects a grid whose meter_sequence names an undeclared
time signature, rather than installing an invariant-violating grid.
- User breaks materialize under the canonical LWW key: `apply_break_lww`
drops any existing anchor resolving to the same position before adding,
so the graph break list stays in lockstep with the resolved-position
ledger map (shared `resolved_anchor_position`). Applied to page and
system breaks alike.
- Coverage: SetMetadata/SetMetricGrid/CreateVoice/DeleteVoice added to the
tag-distinctness test; layout_stub `gen_operation_kind_tag` extended to
every normative tag; the MaterializedState decode test populates
page_breaks; four direct regression tests pin each fixed bug; the stale
SetMetadata/score_metadata doc comments now say advisory LWW.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green
(533); conformance_suite scale 1 passes. Stages only core/ops/testkit; the
unrelated Agent-I working tree is left untouched.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fourth broad-K0 subsystem group — three value-typed LWW field-overwrite ops, one
per settings cluster (additive: OperationKind variants 21-23; framework frozen):
- SetMetadata { metadata: ScoreMetadata } — overwrite the score-singleton
metadata (title/composer/copyright); concurrent differing ⇒
StructuralFieldCollision.
- SetMetricGrid { region, grid: Option<MetricGrid> } — overwrite a region's
default metric grid (LWW keyed by region; concurrent differing ⇒ collision).
- SetUserPageBreak { region, anchor, present } — the page-break sibling of
SetUserSystemBreak: a canonical LWW advisory.
- core: expose ScoreMetadata + MetricGrid via CanonicalValue (no new byte layout
— they already have whole-score Codec impls).
- SetMetadata / SetMetricGrid mirror the modify ops: the resolved value lives in
the graph (reduce_onto), with new last_metadata / last_metric_grid LWW working
state for concurrent-differing detection; MaterializedState records only the
effect and conflict.
- SetUserPageBreak mirrors SetUserSystemBreak's canonical advisory: a new
MaterializedState.page_breaks map (encode + decode added, parallel to breaks),
plus the graph's region user_page_breaks.
Migration: v1-native (no lossy v0 predecessor) -> project/migrate by identity;
the round-trip identity test now covers all four M2 groups.
Coverage:
- testkit operation_payload + ops fuzz gen_payload emit the three kinds, so the
convergence / determinism / migration-equivalence and MaterializedState
decode-roundtrip gates exercise them (incl. page_breaks) at scale.
- reduce_onto graph tests: the three settings materialize in graph and ledger
(metadata title, region default metric grid, region user page break +
MaterializedState.page_breaks), invariant-clean; plus a concurrent-differing
SetMetadata conflict test.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (519);
conformance_suite scale 1 passes. Catalog sections + DECISIONS for the M2 groups
land in M2e per the staged plan. The unrelated Agent-I working tree is left
uncommitted; this commit stages only core/ops/testkit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Addresses four findings spanning the H (pre-pass) and K (reduction) seams plus
the Operation Catalog.
1. [High] A reduced RespellPitch is now visible to the pre-pass. The reducer
stored overrides only in MaterializedState.spellings, but Agent H's
derive_annotations resolves authored spellings from score.spelling_attachments
— so a real respelling accepted by reduce_onto was lost before annotation
derivation, violating manual-override precedence. respell_pitch now upserts a
user-chosen explicit SpellingAttachment into the materialized graph
(materialize_respell / graph_respell_pitch); DeleteIdentifiedPitch drops that
attachment (graph_delete_pitch) so none dangles (it does NOT tombstone the
pitch — the event survives a pitch delete and a later ModifyEvent may reuse
the id, which would make it both live and tombstoned). New testkit gate
assert_reduced_respell_is_honored reduces a real RespellPitch and proves
derive_annotations honors it as Authored(UserChosen); wired into run_all.
2. [Medium] PrePassProfile algorithm ids are now enforced, not just recorded.
derive_annotations ran the default logic and labeled the result with the
requested algorithm. It now runs each pre-pass only when its requested id is
the implemented "default"; an unknown/future id yields no annotations for that
pre-pass (the requested id stays in the result profile), so a future algorithm
can no longer silently alias the default in a derivation cache. Test:
unknown_algorithm_ids_are_not_honored.
3. [Medium/Low] The determinism gate now fingerprints canonical bytes, not Debug.
DerivedAnnotations gains canonical_fingerprint(): embedded graph values
(PitchSpelling, DecompositionAttachment, SpellingSourceKind — the latter two
added to the CanonicalValue surface) use their ratified bytes; counts/ids are
little-endian, length-framed. The pre-pass harness fingerprints with it. A
discrimination check confirms it is not a degenerate constant.
4. [Low] operation_catalog.tex K1 chapter reconciled with the implemented M2
work: the now-dispatched ops (event/pitch leaf-field, cross-cutting CRUD,
structural container CRUD) are listed as implemented-since-M2 (available under
the Phase-2 profile), and the "MUST reject" scope is narrowed to the genuinely
deferred slots (create score/canvas/staff, set metadata, metric-grid/time-sig/
tempo, layout/page-break). PDF rebuilt clean (0 undefined refs).
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (criterion
1 + the pre-pass and convergence gates); conformance scale 1 passes. The unrelated
Agent-I working tree is left uncommitted.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Third broad-K0 subsystem group — six new value-typed ops for the score-graph
containers, reusing M1's disciplines (additive: OperationKind variants 15-20,
new apply arms + reduction methods; framework frozen). Per the project lead's
call, container deletes are EMPTY-ONLY (no cascade): a precondition NoOp unless
the container has no live children, so the caller deletes contents first.
- CreateRegion / DeleteRegion, CreateStaffInstance / DeleteStaffInstance,
CreateVoice / DeleteVoice. Creates are value-typed mints of an empty container
(set-union creation); deletes are delete-wins tombstones gated on emptiness.
- core: expose Region / StaffInstance / Voice via CanonicalValue (no new byte
layout — they already have whole-score Codec impls), with round-trip coverage.
- New PreconditionFailureReason::ContainerNotEmpty (additive discriminant 10;
encode + decode), reported when an empty-only delete hits a non-empty container.
Reduction (reduce.rs):
- Two child-existence indices, region_instances and instance_voices, drive the
emptiness checks (a voice's events are read from voice_occupancy), so the
ledger projection and the graph agree on every delete result. Populated by
seed_from_graph, the create ops, and implicit voice creation in insert_event.
- CreateStaffInstance / DeleteStaffInstance maintain the region's staff_extent so
it lists exactly the manifested staves (Chapter 5 RegionExtents); valuegen's
fresh region uses a far-future wall-clock extent so it never overlaps an
existing region in both time and staff.
Migration: v1-native (no lossy v0 predecessor) -> project/migrate by identity;
group1_and_group2_kinds_round_trip_by_identity extended to cover Group 3.
Coverage:
- testkit operation_payload + ops fuzz gen_payload emit the six kinds, so the
convergence / determinism / migration-equivalence gates exercise the
bookkeeping projection at scale.
- A reduce_onto graph test materializes a region -> staff instance -> voice
subtree (invariant-clean), asserts the empty-only delete refuses a non-empty
container with ContainerNotEmpty, and verifies an ordered teardown clears the
subtree from both graph and ledger.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (criterion
1 green with the new container objects in the corpus); conformance_suite scale 1
passes. Catalog sections + DECISIONS for these ops land in M2e per the staged
plan. The unrelated Agent-I working tree is left uncommitted; this commit stages
only core/ops/testkit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Standalone follow-up to the M2b review: close the pre-existing reanchor /
graph-delete divergence, which unlocks at-scale criterion-1 coverage for the
Group-2 cross-cutting CRUD ops.
The divergence: a DeleteEvent tombstoning a slur/spanner endpoint re-anchored the
structure in the bookkeeping ledger (object stays Live) but materialize_graph_delete
removed it from the graph unconditionally — so the object was Live in
MaterializedState yet gone from the Score. Latent only because graph_edit_session
never created cross-cutting structures.
Fix (graph-materialization only; bookkeeping/convergence unchanged):
- materialize_graph_delete now mirrors reanchor_for_tombstone for slurs and
spanners: an endpoint-deleted structure re-anchors onto its surviving endpoint
(stays in the graph) and is removed only when no endpoint survives. A
two-endpoint structure collapses onto the survivor (degenerate (B,B), but
reference-clean — the cross-cutting invariant requires only live endpoints;
proximity-aware target deferred, P11-C5). Ties (cascade) and beams
(truncate-while->=2) were already consistent and are unchanged. This also fixes
a latent dangling-spanner bug (spanners weren't handled on event delete at all).
- seed_from_graph records each base-score spanner's event-anchored endpoints in
`structures`, so a seeded spanner re-anchors through the same rule as a created
one.
Coverage:
- New reduce_onto tests: deleting one slur endpoint re-anchors in both graph and
ledger (slur Live + collapsed onto survivor); deleting both cascades in both
(slur Tombstoned + removed).
- graph_edit_session now creates slurs over replica-0 events and emits
DeleteCrossCutting / ModifyCrossCutting, so criterion 1 (reduce_onto +
check_invariants, across delivery permutations) exercises cross-cutting CRUD and
slur re-anchoring at scale.
Docs: DECISIONS.md records the graph-follows-ledger re-anchoring decision and the
degenerate-collapse / P11-C5 deferral.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (ops
graph_reduction 20; criterion 1 green with cross-cutting wired in); conformance
scale 1 passes. The unrelated Agent-I working tree is left uncommitted; this
commit stages only ops/testkit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Second broad-K0 subsystem group — two new value-typed ops reusing M1's proven
disciplines (additive: OperationKind variants 13-14, new apply arms + reduction
methods; framework frozen):
- DeleteCrossCutting { structure: TypedObjectId } — delete-wins tombstone of a
cross-cutting structure (idempotent concurrent deletes; guarded to the
Tie/Slur/Beam/Spanner kinds). Drops the transient endpoint/LWW indices so a
later event-tombstone re-anchoring pass never re-processes the deleted
structure.
- ModifyCrossCutting { structure: CrossCuttingValue } — LWW field-overwrite by
the structure's id; concurrent differing => StructuralFieldCollision. Mirrors
modify_event (resolved value lives in the graph, not MaterializedState);
re-derives endpoints from the new value, and mirrors CreateCrossCutting's
beam->=2 / endpoints-live preconditions.
Graph materialization (reduce_onto): graph_delete_cross_cutting removes the
structure by id; graph_modify_cross_cutting replaces it in place by id, across
all four kinds (Slur/Tie/Beam/Spanner). New last_cross_cutting_modify LWW map,
synced through WorkingSnapshot/snapshot/restore.
Migration: v1-native (no lossy v0 predecessor) -> project/migrate by identity;
group1_and_group2_kinds_round_trip_by_identity extended to cover them.
Coverage:
- testkit operation_payload + ops fuzz gen_payload now emit both kinds, so the
convergence / determinism / migration-equivalence gates exercise the
bookkeeping projection at scale.
- Targeted reduce_onto graph tests (tests/graph_reduction.rs) cover every kind
arm of graph_delete/graph_modify_cross_cutting (Slur/Tie/Beam/Spanner) plus
the beam->=2 reject branch of modify, with check_invariants; plus two
bookkeeping unit tests (delete tombstones; concurrent differing modify
conflicts).
Not wired into graph_edit_session (criterion 1): doing so requires creating
structures in the session, which exposes a pre-existing M1 reanchor/graph-delete
divergence (a slur whose endpoint event is deleted is re-anchored in bookkeeping
but removed from the graph). That is a separate DeleteEvent fix; the targeted
reduce_onto tests above give the M2b graph paths guaranteed coverage meanwhile.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (ops lib
53, ops graph_reduction 18); conformance_suite scale 1 passes. Catalog sections +
DECISIONS for these ops land in M2e per the staged plan. The unrelated Agent-I
working tree is left uncommitted; this commit stages only ops/testkit.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
From the M2a review (no bug in the bookkeeping reduction; the gap was that the
Group-1 ops' *graph* materialization — reduce_onto — was unexercised by the
gates, which hid two invalid-graph edges). Fixes are graph-materialization only;
the bookkeeping projection, and therefore convergence/determinism, is unchanged.
- DeleteIdentifiedPitch of a single-pitch note's last pitch left an empty
(Chapter-5-invalid) PitchedEvent via EventArena::get_mut (which bypasses
insert's well-formedness guard). It now degrades the note to a Rest of the
same id/voice/position/duration; InsertIdentifiedPitch into a rest is the dual
(rest -> one-pitch note), keeping the graph consistent with the bookkeeping
that mints/tombstones the pitch object either way.
- ModifyEvent now skips placement-changing (move) and malformed-empty pitched
replacements in the graph rather than corrupting invariant 3
(VoiceEventsSortedNonOverlap) via get_mut; voice re-sort stays deferred and the
LWW bookkeeping still records the modify.
Coverage: graph_edit_session (criterion 1, reduce_onto + check_invariants) now
emits all five Group-1 kinds, so the real-Score gate exercises their graph
mutations at scale; plus two targeted reduce_onto regression tests (note->rest,
rest->note) in tests/graph_reduction.rs.
Docs: DECISIONS.md records the note<->rest equivalence and the ModifyEvent
placement deferral (catalog section prose routed to M2e); Transpose / P12-K2 note
the i8-saturation caveat; valuegen::pitch_value_nth no longer implies spelling()
is injective.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (ops
graph_reduction 13, ops lib 51); conformance_suite scale 1 passes. The unrelated
Agent-I working tree is left untouched; this commit stages only ops/testkit/spec.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
First broad-K0 subsystem group — five new value-typed ops reusing M1's proven
disciplines (additive: new OperationKind variants 8–12, new apply arms +
reduction methods; framework frozen):
- ModifyEvent { event: Event } — field-overwrite LWW by EventId; concurrent
differing ⇒ StructuralFieldCollision.
- Transpose { targets, chromatic_steps } — order-dependent; pitch ids preserved;
canonical footprint = effect-log entry; reduce_onto applies a minimal CMN
alteration shift (rich interval algebra deferred — P12-K2).
- InsertIdentifiedPitch / DeleteIdentifiedPitch — pitch-within-event mint /
delete-wins tombstone.
- ModifyIdentifiedPitch { pitch, value: Pitch } — field-overwrite LWW (the pitch
VALUE, distinct from RespellPitch's spelling-only overwrite).
Design (honesty rule): the modify/transpose ops record effect + conflict
canonically — the resolved values live in the graph (reduce_onto), since
MaterializedState is bookkeeping, not a second graph; respell stays special
because spelling is a bookkeeping-owned annotation. LWW diff uses new
`last_event_modify`/`last_pitch_modify` working maps (synced through
WorkingSnapshot/snapshot/restore).
- core: expose Pitch + IdentifiedPitch via CanonicalValue (no new layout).
- The five kinds are v1-native (no lossy v0 predecessor): project/migrate them by
identity; only the original kinds reconstruct from a lossy v0 form.
- Generators (testkit operation_payload, ops fuzz gen_payload) now emit the new
kinds, so the convergence / determinism / migration-equivalence gates exercise
them at scale; plus targeted migrate identity + reduce LWW/mint/delete tests.
Gates: build/fmt/clippy -D warnings clean; cargo test --workspace green (ops lib
51 tests); conformance_suite scale 1 passes. Catalog sections + DECISIONS for
these ops land in M2e per the staged plan.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
From the M1 review (no correctness bugs found; these are accuracy corrections to
shipped artifacts, not new feature work):
- migration gate: the `v1 == migrated` assertion's comment overclaimed a
universal inverse. It is round-trip self-consistency over the representative
corpus (which is built from the same valuegen helpers the migration
reconstructs values with); the spec-level property is the reduction-equivalence
asserted alongside it. The ReplaceWithRest rest-voice is the known
non-invertible field, recovered from the deleted event's placement at reduction.
- operation_catalog §CreateCrossCutting: document that v0→v1 migration covers the
event-anchored Tie/Slur/Beam; a Spanner (anchor-based) cannot be reconstructed
from the v0 event-reference and is reported unmigratable (read-only) under M1,
so the catalog no longer silently implies it round-trips.
Review findings deferred to M2 (per project lead): migrate.rs unit tests for the
untested reconstruction branches, and the valuegen::spelling() mod-7 domain
cleanup. Gates unchanged and green (comment + spec-text only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Foundation milestone for Track B's Operation Catalog: shift epiphany-ops from
the v0 identifier-only payload projections to durable value-typed payloads, and
scaffold the companion spec. Scope: the representative §6.10 set (7 primitives +
2 meta-ops); the slice-driven K0 expansion follows.
Core (the K↔J seam):
- epiphany-core exposes a public `CanonicalValue` trait (canonical_bytes /
decode_canonical) delegating to the existing private `Codec` machinery, for
Event/Rest/PitchSpelling/Tie/Slur/Beam/Spanner/RegionTimeModel/TimeAnchor.
No new byte layout — a value's bytes equal what the whole-score codec emits,
so all goldens / criterion 4 stay byte-identical.
Ops (value-typed payloads, frozen reduction rules):
- InsertEventOp{staff_instance,event:Event}, RespellPitchOp{pitch,spelling:
PitchSpelling}, CreateCrossCuttingOp{structure:CrossCuttingValue},
ChangeRegionTimeModelOp{...,new_time_model:RegionTimeModel},
SetUserSystemBreakOp{...,anchor:TimeAnchor}, ReplaceWithRest{rest:Rest}.
Payloads frame each value's CanonicalValue bytes behind a u32 length prefix.
- reduce.rs: read-sites only moved onto the value (rules, conflicts, ordering,
promotion, re-anchoring, undo, transactions unchanged); reduce_onto now
materializes the real event/structure instead of the C4 placeholder.
MaterializedState.spellings now stores PitchSpelling (encode + decode updated).
- v0.rs: frozen identifier-only shapes (migration regression guard).
- migrate.rs: migrate_v0_envelope(v0, &Score) + project_v1_to_v0 + MigrationError;
deterministic and equivalence-preserving. Respell spelling recovered from the
score context; irreversible case is P12-K1.
- valuegen.rs: shared value-type builders (reused by fuzz, migration, tests,
testkit). Resolves P11-C1; P11-C10 Dismiss recorded.
Testkit (Agent F merge gate):
- migration.rs: reduce(v1)==reduce(migrate(project(v1))) byte-identical, plus
migration determinism and a non-vacuity guard; wired into acceptance.rs as
agent_k_migration_equivalence_gate. Generators/harnesses build v1 payloads.
Spec:
- spec/operation_catalog.{tex,pdf}: new companion (independent semver) — framework
+ per-primitive template, the 7+2 representative primitives, the v0→v1 migration
contract, and K1 framework slots for the remaining K0 primitives. Builds clean.
- PASS12_BATCH.md: P12-K1 (respell fingerprint irreversibility).
Gates: cargo build/fmt/clippy -D warnings clean; cargo test --workspace green
(incl. criteria 1/4/5/6 and the new K gate); conformance_suite scale 1 passes.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Land the two real pre-passes as canonical *derived annotations* (pure
functions of the materialized Score + profile, recomputed on
materialization, never serialized into canonical Score bytes), exposed via
`derive_annotations`:
- Spelling: a Temperley-style line-of-fifths centre-of-gravity preference
rule (key-free, deterministic), preserving authored CMN letters and only
inferring spelling for chromatic/integer input. `resolve_spelling`
layers authored overrides above the inferred default (the RespellPitch
precedence rule). `spell` now takes `&Pitch` and delegates to
`simplest_spelling`.
- Decomposition: metric greedy-aligned splitting on a 1/4096 integer grid
(barline + dyadic-boundary ties), with exact sounding->notated tuplet
conversion before gridding. Components reconstruct the event duration
(invariant 15).
- A per-event-kind eligibility `TaxonomyReport` so "ineligible" is always
explicit and counted, never silently absent.
Test infrastructure (Agent F): a 29-fixture representative corpus +
taxonomy harness (corpus.rs), the H spelling/decomposition merge gate
(prepass_harness.rs), a discrete `tests/prepass.rs` CI target, conformance
stage [7b], a dedicated CI job, and the Pass-12 batch tracker.
Review hardening folded in (nine findings):
- Guard `decompose_metric` against a zero-length measure (was a
divide-by-zero panic; now reported ungriddable).
- Resolve spelling-override priority via `Reverse` instead of negation
(was an i32::MIN overflow).
- Verify spelling *register* (octave), not just pitch class, in the gate.
- Close the decomposition under-emission gap: the unusual-outcome
taxonomy buckets are an exact per-fixture whitelist (classify_corpus
step 5b).
- Generalize `accidental_ids` to a glyph stack so authored extreme
alterations (triple-sharp+) reconstruct exactly instead of being clamped.
- Per-fixture spread checks in the non-vacuity tripwire and broad-bucket
coverage, so no single rich fixture can carry a signal (partial-stub
resistance); added a `mixed_rhythm` fixture for margin.
- Pin the integer-grid note-value math to the canonical rational helpers
via an exhaustive test; cross-reference comments.
- Replace the O(n^2) tuplet innermost-resolution scan with an id index.
fmt + clippy -D warnings clean; 199 tests pass; conformance suite green.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_011giSRaHCFCGm1Z2SWv6JHt
Audit follow-up to b2f2e20. The ratification was byte-faithful, but the
audit found a gap between the protection the worklist claims ("a
golden-bytes test already locks every byte-layout item") and the
protection actually in place for several newly-normative tables, plus
three small inaccuracies in the ratified spec text.
Golden locks (close the round-trip-vs-golden gap):
- ChunkKind: chunk_kind_discriminants_are_golden pins the literal 0..=8.
ChunkKind::canonical_bytes() is in the chunk hash preimage, so the
prior round-trip-only test would let a coordinated renumbering silently
change every chunk content address while passing.
- CompressionAlgorithm: compression_algorithm_encoding_is_golden pins the
exact bytes (None -> [0,0], not a bare tag).
- ProfileId (load-bearing superblock field): profile_id_discriminants_are_golden
pins the u32 discriminants and the fixed 20-byte encoding.
- ResolutionAction / TransactionCategory / ObjectKind: *_discriminants_are_golden
pin the canonical discriminants (ObjectKind feeds the anomaly id;
ResolutionAction/TransactionCategory feed operation content hashes).
- IntegrityAnomalyId: integrity_anomaly_id_byte_form_is_locked golden-locks
the MUSCSANM-derived id (cross-replica agreement is a conformance
property; it previously had no byte-form golden).
Spec-text fixes (core_spec.tex):
- CompressionAlgorithm: "None = 0 (no payload)" was wrong; the code writes
a fixed two bytes (discriminant + always-present parameter byte). Text
now states the fixed-width framing.
- ProfileId: "a single discriminant followed by any variant payload" was
wrong; it is a u32-LE discriminant + a fixed 16-byte registry id (zero
unless Custom), 20 bytes total. Text now matches the only encoding.
- TupletRatio listing showed `pub` fields (freely constructible by struct
literal), contradicting req:time:tuplet-ratio-construction. Listing now
shows private fields + the checked `new`/`actual()`/`notated()`, matching
the code.
Test honesty:
- testkit resolution_action generator now emits Dismiss (rng.below(6)); it
previously skipped the variant, leaving the Dismiss path unfuzzed.
Verification: cargo test --workspace (433 pass, +7), clippy -D warnings
clean, fmt clean; spec rebuilds (lualatex/latexmk, 0 undefined refs, 254pp).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The layout time axis was inert: TimeAxisModel carried bare Vec<SpringSlotId>,
project()/affected_slots() ignored their arguments (returning the first slot /
all slots), nothing populated it, and nothing consumed it.
Now it carries ordered SlotPlacement { time, slot } entries and has real
behavior:
- project(time) returns the slot covering a time (greatest placement at or
before it; first when the query precedes them all);
- affected_slots(range) returns the slots in a half-open time range;
- slots() lists them in time order;
- with_placements populates and sorts the axis from resolved spring slots.
The spacing stage (to_constrained) now populates each region's axis from its
spring slots and carries the populated axis on ConstrainedLayoutRegion, so the
axis is a real, consumed artifact. Tests cover project/affected_slots semantics
and that spacing produces a per-region axis whose project() is a genuine
function of the queried time. DECISIONS updated.
This completes item 6 (and the whole v0 follow-up list, items 1-6 / M1-M5).
(The slot times are still the prototype's wall-clock spacing columns; mapping a
metric region's measure/beat grid to musical times is the next layer.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Chapter 8's OperationEnvelopeBlock carries dvv_summary/min_stamp/max_stamp so a
reader can select or skip a block by causal frontier / stamp range without
decoding it. These are semantic (ops-computed); the bundle carries them opaquely.
Bundle (Agent D):
- OperationBlockSummary { dvv_summary: FrontierBytes, min_stamp, max_stamp } and
Manifest.operation_block_summaries: BTreeMap<ChunkId, OperationBlockSummary>,
keyed by the block's chunk id, encoded/decoded in canonical (ChunkId-ascending)
order and accessible via Manifest::operation_block_summary. Optional and
non-canonical; preserved across reopen by the manifest round-trip.
- Round-trip + selectability test.
Testkit (Agent F, the C/D integration point):
- roundtrip::operation_block_summary computes the summary from envelopes using
ops (causal frontier + min/max OperationStamp canonical bytes).
- assert_operation_block_summary_survives_storage commits a real operation block
+ its summary, reopens, and selects the summary by block id without decoding
the payload. Wired into acceptance + the conformance suite.
bundle DECISIONS updated (summary metadata now carried, not omitted); fixed a
stale "pending item 5" doc on criterion 4 (the whole-score codec has landed).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
E-D (layout-ir, honest solver tier): add SolverTier::Stub (a non-conformance rung
below Minimal) and have StubSolver report it instead of falsely claiming the
Minimal conformance tier; the passthrough evaluates no constraints and computes
no quality metrics.
E-C (layout-ir, constraint/reference validation): ConstrainedLayoutIR::validate()
now also checks the LayoutConstraint vector — NoCollision/Align/PositionWithin
must name glyphs in the set, SystemBreakAt/PageBreakAt must name existing slots,
PositionWithin regions must be finite/non-negative — rejecting dangling
references instead of silently accepting them.
E-B (layout-ir, content-sensitive ScoreVersion): derive ScoreVersion from the
whole score's canonical bytes (Agent B's whole-score codec) rather than the
layout projection's object identities, so a pure content edit that changes no
identifier still changes the version — required for correct incremental-layout
cache invalidation.
D-A (bundle, extension-root preservation): Bundle::commit now enforces
preservation — after the builder closure runs, every prior extension declaration
it did not re-declare (by extension_id) is carried forward verbatim, so an
extension-unaware writer cannot silently orphan an unknown extension's
preserved_chunk_roots. An extension-aware writer that re-declares its id keeps
control.
Each fix has a regression test; per-crate DECISIONS updated. (Item-6 remainder:
D-B operation-block summaries next; E-A real time-axis deferred per request.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements a total, reversible canonical byte form for the entire
epiphany_core::Score graph, unblocking the byte-level full-Score serialization
gate that M3 had to defer.
epiphany-core/src/codec.rs:
- Score::canonical_bytes() / Score::decode_canonical() with a validating
ScoreDecodeError and a forward-only Reader cursor.
- A local Codec trait with generic combinators (Option/Vec/BTreeSet/BTreeMap/
tuple) and macros (struct_codec!/cstyle_enum_codec!/unit_codec!/
catalog_id_codec!) so encode and decode stay symmetric across ~110 types
spanning graph.rs, event.rs, pitch.rs, time.rs, tempo.rs.
- Uniform form: LE integers, one discriminant byte per tagged union, u32
counts/length-prefixes, every variable-width leaf length-prefixed, raw UTF-8
for free text (so decode(encode(x)) == x for any valid score; catalog ids are
already NFC). EventArena round-trips via iter_canonical + insert.
- Two pub(crate) accessors added for the codec: EventOrderingDAG::edges_ref,
SpellingPrecedence::order_ref.
- Tests: generator-score corpus (valid_score + valid_score_rich), exotic
event/pitch variants the generators omit, distinctness, and decoder
rejection of trailing/truncated/empty bytes.
epiphany-testkit:
- roundtrip::assert_score_serialization_stable: encode the real Score, store it
as a bundle Snapshot, reopen + hash-verify, decode to an equal Score, and
assert a byte-identical re-encode.
- convergence::materialized_score builds a real ~50-bar reduce_onto
materialization for the gate.
- criterion_4_full_score_byte_roundtrip flips from #[ignore] to a live gate;
wired into the conformance suite. Docs (lib.rs, README, core DECISIONS P11-4)
updated to reflect the landed codec.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The audit flagged Agent F's criterion 1 (convergence) and criterion 4
(serialization) as testing the reducer-bookkeeping projection
(MaterializedState) while claiming to be full-Score gates. This makes them
honest, per item 4 of the v0 follow-up plan.
Criterion 1 — real-Score convergence:
- generators::graph_edit_session builds a real ~50-bar, two-voice edit
session targeting a base Score's actual voices (so it survives reduce_onto,
which rejects unknown voices).
- convergence::{assert_graph_convergence, run_graph_convergence} reduce that
session onto a real epiphany_core::Score via OperationSet::reduce_onto and
assert the entire GraphMaterialization (graph + bookkeeping) is identical
across delivery orders, passes check_invariants, and genuinely grows both
voices (non-vacuity).
- acceptance criterion_1_convergence now drives this; the former bookkeeping
convergence is retained and renamed reducer_bookkeeping_convergence.
Criterion 4 — honest serialization tiers:
- criterion_4_canonical_serialization_stability keeps the real typed/manifest
round-trips; the MaterializedState round-trip is split out as
reducer_bookkeeping_serialization.
- full_score_materialization_is_reproducible asserts the materialized Score is
reproducible across orders (the determinism precondition for a byte codec),
achievable without the codec.
- criterion_4_full_score_byte_roundtrip is #[ignore]'d pending item 5's
whole-score codec (visible as ignored, never falsely green).
Negative regression guards (src/negative.rs): one guard per audited M1 defect
(inverted causal/HLC order, missing predecessor via vector, HLC 100/200/50
quarantine-from-0, tx rollback of member conflicts, causally-ordered
same-position non-promotion, partial-duration overlap), driven through the real
epiphany_ops API with explicit negative controls. Wired into acceptance and the
conformance suite. Crate/README docs updated to match.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Implements Agent E per spec/QUICKSTART.md — the layout intermediate
representation (Chapter 7) and the constraint-solver interface (Chapter 9):
* Four IR stages: LogicalLayoutIR -> ConstrainedLayoutIR ->
ResolvedLayoutIR -> RenderIR (interface only), with the composite-object
taxonomy, spring slots/constraints, vertical-band model, pages/systems,
engraving decisions + overrides, and the incremental dependency/cache model.
* TimeAxisModel tagged enum (Metric/Proportional/Aleatoric/Registered).
* Provenance back-references with manifestation- and synthesis-aware ids
((source, region) and (source, kind, ordinal)), so multiply-manifested and
synthesized objects never collide.
* In-tree Bravura GlyphCatalog (Send+Sync, metrics + render-data interface),
MUSCFNTM-tagged metrics hash with anchors hashed as a name-keyed map.
* Edit-barrier types keyed on OperationKindTag, with precise EditContext /
EditOracle scope/condition evaluation.
* StubSolver: returns SolveStatus::Solved with the input geometry verbatim;
spec-compliant SolveReport, Minimal tier + all-worst (unmeasured) metric
vector (no false conformance claim); rejects ill-formed input.
* f32 staff-space IR coordinates, quantized to the 1/1024 grid only at
canonical ResolvedLayoutIR serialization (Appendix D); non-finite geometry
is rejected, not normalized. Canonical encoding is injective in glyph
identity, provenance, engraving decisions, and catalog identity.
Re-points Agent F's testkit layout harness from its in-tree stub to the real
crate (v0 acceptance criterion 6) and expands its generators to E's public
surface. Expands Agent C's OperationKindTag to the full normative variant set
so edit barriers can prohibit every operation class.
Workspace gates green: fmt, clippy -D warnings, 377 tests, doc tests, rustdoc
-D warnings, and the conformance suite at scale 1. Decisions and Pass 11
candidates recorded in crates/epiphany-layout-ir/DECISIONS.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Levi Neuwirth