The staged interactive-edit latency bench: reduce / engrave / scene-build /
paint measured separately, gating the core's portion against
req:perf:single-system-edit-latency's 16.7 ms frame. Criterion 2 asserts that a
toolkit verdict is uninformative while reduction dominates, and that sentence
had never been measured; the sequencing question it governs -- T4's spike now,
or T4b's incrementality first -- was resting on it.
The stage split is not invented here. It is the seam EditorSession::materialize
already walks, read off its private render_score and reproduced stage for
stage, so the bench measures the pipeline rather than a model of it. Only reduce
and engrave are gated: the requirement bounds "the core's portion" and says in
its own words that edit-to-pixel latency is a product-layer obligation, so
charging the SVG serializer and resvg against a core budget would be a category
error. They are measured and printed because the ruling asks for the stages
separately, and because today's is the path Ruling A demotes -- the number is
the baseline a canvas must beat, not a budget to defend.
Four findings, in the order they matter. Reduce is the only stage that scales
with log depth, near-linearly, and it breaks the frame at roughly ten thousand
edits -- 17.26 ms against 16.7, a three percent miss, so an order of magnitude
rather than a threshold. Engrave is flat and small at ~280 microseconds, and at
shallow depth it is the larger half of the core's portion, which qualifies
criterion 2 rather than confirming it: reduction does not dominate until about
depth five hundred. Paint is the largest single cost at every realistic depth --
2.12 ms at depth one hundred is four and a half times the entire core portion.
And scene-build is 3.5 microseconds of IR work plus about 130 of SVG
serialization, which the no-feature run separates: a canvas consuming the IR
directly skips some ninety-eight percent of today's per-edit cost, none of it in
the core.
The sequencing answer is therefore that T4 before T4b stands, for the opposite
reason to the one assumed. The dominant cost at the depths real sessions reach
is the render path Ruling A already demoted, not reduction. T4b's trigger is a
session ten thousand edits deep, and the bench now watches for it as the one
Xfail row.
Two things the bench had to survive being wrong about, both mine. The depth-1000
row was drafted Xfail on the assumption Fact 8 would already bite; it passes
with eightfold margin, the gate's XPASS notice said so, and the row is promoted
here rather than left stale -- which is the whole point of that mechanism. And
the first edit log alternated transposition direction per operation, which is
degenerate when the pitch-list length is even: every edit to a given pitch
pushed the same way, drifting it twenty-five semitones by depth 1000 and would
have been two hundred and fifty by depth 10000. That inflated engrave by a
factor of two and paint by nearly three -- a score-content change wearing a
log-depth costume. Alternating per pass instead bounds drift to one semitone.
The residual content effect is documented rather than hidden: paint is
non-monotonic in depth because pass-count parity decides how many accidentals
the score carries, and reading its dip at depth 10000 as a scaling win would be
a mistake.
Stated limitation: the testkit's largest fixture is three staves by ten
measures, so the engrave and scene-build columns are lower bounds and this
cannot prove the budget holds on the hundred-page orchestral score the
requirement contemplates. It shows where the time goes at the scale we can
build, and a row that misses at this size misses by more at a real one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RSX4zSLgKvtiXaPjnMqLGz
A golden-gate feature on epiphany-testkit: resvg + epiphany-render-svg as
optional regular dependencies (dev-dependencies cannot be optional; the
MSRV test job builds examples with --all-targets, so the raster stack must
stay out of the featureless closure — proven by cargo tree, 0 vs 1). Gate 9
re-derives the three T1a golden states headlessly and compares decoded RGBA
against the committed baselines; compare-only, failing loudly with a pointer
to the editor-gui diagnostic tests. Without the feature the suite's output
is byte-identical to before; with it, 9/9. The conformance CI job's
invocation gains the feature; CONFORMANCE.md documents gate [9] and the
split. Two mutations killed and coordinator re-verified.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Thirteen drift-locked document vectors -- four accept, nine reject, one per
implemented rejection class -- gated as conformance step [7e] beside [7d]'s 65
decode vectors.
The checker distinguishes three outcomes, and that distinction is the whole
point. `Rejected`, `AcceptedCanonical`, and `AcceptedNonCanonical` are separate
verdicts, and every one of the four wrong pairings is a failure: an accept vector
that merely normalizes fails, and a reject vector that is accepted -- whether as
canonical or after normalization -- fails. The decode corpus learned this the
expensive way, where collapsing "rejected" with "accepted but normalized" let a
corpus pass against the exact decoders it was written to catch.
Accept vectors assert the byte-checkable equation `project(serialize(parse(T)))
== T`, which quantifies over TEXTS. The comments say plainly that it must not be
replaced with bundle identity, and name the three deliberate binary losses --
duplicate blobs collapsing under derived-ordering, regenerated physical layout,
and dropped accelerators -- so the next reader does not "fix" it into a bundle
comparison and then chase a phantom.
Reach is asserted exactly, not merely non-zero: two documents each carrying an
extension, a canonical base, a custom profile, a lineage, and more than one
envelope; and every reject class must have exactly one vector the parser really
refused. A mislabelled reject drops its class to zero rather than collecting
credit from its label.
The corpus was verified by breaking the implementation, not the checker: making
`parse_document` silently tolerate a missing trailing LF is caught three ways,
and the suite reports "declared reject, but was ACCEPTED and normalized".
Review fixed one defect in the semantic harness. It claimed to check
`semantics(parse(project(B))) == semantics(B)` but compared `parse(project(B))`
against `document_from_bundle(B)` -- and `project_bundle` *is*
`project_text_document(document_from_bundle(..))`, so both sides flowed through
one function and any bug in it cancelled out. Dropping an envelope there passed.
It now compares against the envelopes the test itself serialized in, which is
independent of the projection path; the same mutation fails.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
857 workspace tests pass; clippy -D warnings, fmt, and rustdoc clean;
both new companions build with zero undefined references.
Quality Metric Catalog v0.1.0 (spec/quality_metric_catalog.tex, new):
- Formal definitions for all nine normative quality metrics, each with
a raw measurement over resolved geometry and a clamped-linear
normalization n = min(1, raw/R_worst) with pinned anchors.
- The vacuous-geometry rule (a metric over absent geometry evaluates
to 0.0; the notated-but-unrendered honesty edge is an open
question), all-1.0 default tie-breaking weights, and the per-tier
threshold table — Minimal's uniform 0.90 deliberately fails the
all-worst placeholder, forcing real measurement.
- Pins QualityMetricKind (referenced but never defined by the core
spec) and the registered SolverProfile catalog (Draft selects the
Minimal threshold column; Standard/Publication select Standard).
- QualityFloorApproached fires at 0.8x the applicable threshold and
is status-neutral by requirement.
Reference Suite v0.1.0 (spec/reference_suite.tex, new):
- Six entries referenced by deterministic builder + seed (RS-1
ten_measure_single_staff, RS-2 valid_score_rich, RS-3..6 corpus
fixtures), each with the declared A4-at-8mm-staff solve geometry
(Canvas.layout_defaults has no graph home yet, P12-I7).
- All entries required at Minimal; the same set is the pre-declared
Standard bar (no implementation claims Standard yet). Fixed-
expectation tests deliberately unused in v0.1.
Real metrics in the engraver (engrave/src/quality.rs, new;
layout-ir/src/quality.rs = the catalog constants transcribed):
- QualityMetricVector::unmeasured() replaced with computed values:
collision sweep with the catalog's same-slot-cluster and stroke
exclusions, per-system spacing CV, vertical gap deviations,
system-break slack, page fill, casting-off width CV, symbol
density; slur/beam vacuously 0.0 (no drawn geometry exists).
- Bit-identical across repeated solves (tested); floor warnings never
change solve status; malformed inputs keep unmeasured(). The two
all-worst test pins now assert real values; the StubSolver's
unmeasured() stays (Stub genuinely computes nothing).
Reference-suite harness (testkit reference_suite module + tests):
- Each RS entry asserts the four-condition Minimal pass (hard
constraints, byte/bit determinism, well-formed Minimal report,
every axis within threshold) under the F1 Pass/Xfail discipline,
with the measured table printed per run.
- HONEST FINDING, day one: RS-1 fails Minimal casting-off (measured
1.0 vs 0.90) — greedy first-fit leaves a two-measure stub last
system (width CV 0.6145). Encoded as an asserted Xfail row (fails
on XPASS) and filed as P12-I11 (engrave balance pass, or catalog
revision). P12-I12: the Standard spacing floor warns on short
scores with wide lead measures.
Multi-system click-to-insert fix (editor-core):
- Casting-off exposed two inversion breaks: position_anchors fed a
non-monotonic cross-system anchor list into a monotonic inverter
(system-2 clicks resolved to system-1 times), and
nearest_manifestation found only system 1's staff-line segment
(system-2 clicks got system-1 pitch geometry).
- Fixed with a containing-system lookup over the resolved pages tree
(containment, else nearest by vertical distance), per-system staff
resolution, and per-system anchor filtering; degenerate-geometry
fallback preserves the flat path, so all 84 pre-existing
editor-core tests pass unmodified.
- Five regression tests through the real Engraver over the wrapped
ten-measure fixture, each shown to fail without the fix; testkit
gains dev-only dependencies on editor-core and engrave.
Trackers: P12-I11/I12 filed; DECISIONS entries in engrave, layout-ir,
and testkit; Phase-3 memory updated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NEs4aYiu8MXjdYdMxw8PTd
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
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