neuwirth
/
epiphany
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Contract revision 6: round 1 was testing the wrong property on the wrong glyphs 

Building round 1's oracle exposed two defects in the round as written, both
mine, and both invisible until something tried to satisfy it.

The glyph set was chosen by subpath count. I wrote that 19 of the 37 bundled
outlines have more than one subpath and named gClef, fClef, timeSig8 and
accidentalFlat as the useful ones -- conflating multi-subpath with has-a-hole.
fClef is the counter-example: its three subpaths are a bowl and two SOLID,
disjoint dots, nested in nothing. Measured across all 37 by point-in-path,
exactly twelve carry a bounded hole; fClef, cClef, barlineFinal and every
repeat glyph carry none.

Round 1 now runs five glyphs in two classes testing two different properties.
Hole checks -- gClef, timeSig8, accidentalFlat and noteheadHalf, which is new
and earns its place by being frequently repeated and semantically consequential
(a filled counter renders half notes as quarter notes, a notation error rather
than an artifact). Disjoint-component check -- fClef alone, no background
requirement, instead requiring one ink point inside EACH of its three filled
subpaths, tagged with its subpath index. That second class catches a
tessellator that keeps only the largest contour, which would pass every hole
check ever written. Hard failure is now stated as either: a bounded hole
painted as ink, or a required filled subpath omitted.

The oracle's status model is now specified rather than inferred from an
absence. fClef passing with zero background points is a SATISFIED result under
its own requirement class; recording it only as background_satisfied = false
would make a correct outcome indistinguishable from a failed one.

The second defect was the criterion itself. Ruling A said epaint does not
implement even-odd/nonzero fill for paths with holes -- framing criterion 1
around the fill RULE. Bravura's contours are correctly oppositely wound
(signed ring areas gClef [8.702, -0.691, -1.803, -0.509]; fClef
[2.534, 0.153, 0.148], all positive, the same fact from the other side), so
even-odd and nonzero AGREE on every bundled hole. The rule is not load-bearing;
preserving every filled contour and every bounded counter is. The criterion is
amended to compound-path / inner-subpath fill correctness, with a more accurate
reason for excluding raw egui shapes than the one it replaces: PathShape is a
single point loop documenting "Fill is only supported for convex polygons", so
it cannot express compound-fill or subtractive-hole semantics -- a Shape::Vec
can group loops, but grouping paints them, it does not subtract a counter from
its enclosing contour.

Signed areas are recorded from the oracle's adaptive flattening rather than an
earlier coarse fixed-step measurement, with the note that magnitudes are
flattening-dependent and the SIGNS are the claim.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RSX4zSLgKvtiXaPjnMqLGz
This commit is contained in:
Levi Neuwirth 2026-07-28 21:24:47 -04:00
parent 3af242d457
commit b724e94faf
2 changed files with 87 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

90
spec/CONTRACT_EDITOR_T4_SPIKE.md
View File

@ -1,13 +1,18 @@
# Contract: Editor T4 — the toolkit spike
**Revision 5** (2026-07-28), after four reviews. Revision 1 left discretion in
**Revision 6** (2026-07-28), after five reviews. Revision 1 left discretion in
the measurement and elimination mechanics; revision 2 reordered the ladder and
pinned the deciding numbers, but its new structure carried its own defects —
a damage oracle that compared the wrong thing, correctness failures dressed as
environmental absences, and an escalation branch with no continuation.
Revisions 3 and 4 closed those and introduced smaller ones of their own — a
damage gate covering one rung, a censored median that flattered silence, and a
latency threshold with no timer origin. This revision closes those.
latency threshold with no timer origin. Revision 5 closed those. **Revision 6 amends Round 1** after building its
oracle exposed two defects in the round as written: its glyph set was chosen by
subpath count rather than by measured holes (`fClef` has none), and it framed
criterion 1 around the fill *rule* when Bravura's oppositely-wound contours make
even-odd and nonzero agree. Both are corrected below, and Ruling A criterion 1
is amended to match.
Repo root `/home/jeans/Repos/active/epiphany`. Governed by
`spec/PLAN_EDITOR_APP.md` Ruling A ("What this ruling does **not** pin: the
@ -46,10 +51,31 @@ two make it possible at all.
`PathCommand::{MoveTo, LineTo, CurveTo, Close}` in staff-space units
(`glyphs/src/catalog.rs:21,39`; `layout-ir/src/glyph.rs:283`). No candidate
needs to parse SVG `d` strings, and no glyph work belongs in this packet.
* **Glyphs with holes are in the bundled set, and can be named rather than
assumed.** 19 of the 37 bundled outlines have more than one subpath;
`gClef` (4), `fClef` (3), `timeSig8` (3), `accidentalFlat` (2) are the
useful ones. Criterion 1 has real material.
* **Glyphs with holes are in the bundled set — but subpath count does not
identify them.** Revisions 1–5 named `gClef`/`fClef`/`timeSig8`/
`accidentalFlat` on the strength of "19 of the 37 bundled outlines have more
than one subpath". That conflates *multi-subpath* with *has a bounded hole*,
and `fClef` is the counter-example: its three subpaths are a bowl and **two
solid, disjoint dots**, nested in nothing. Measured over all 37 by
point-in-path (2026-07-28), exactly twelve carry a **bounded hole** —
`gClef`, `timeSig8`, `accidentalFlat`, `accidentalSharp`,
`accidentalNatural`, `noteheadHalf`, `noteheadWhole`, `noteheadDoubleWhole`,
`timeSig0`, `timeSig6`, `timeSig9`, `dynamicPiano` — while `fClef`, `cClef`,
`barlineFinal` and every repeat glyph carry none. Criterion 1 has real
material, and Round 1 below names it by that measurement rather than by
subpath count.
* **Bravura's contours are correctly oppositely wound**, so even-odd and
nonzero *agree* on every bundled hole. Signed ring areas, as measured by the
Round 1 oracle's adaptive flattening (tolerance 0.0005 staff-space, the
authoritative figures — magnitudes are flattening-dependent, the **signs**
are the claim): `gClef` `[8.702, −0.691, −1.803, −0.509]`, `timeSig8`
`[2.674, −0.435, −0.515]`, `accidentalFlat` `[1.040, −0.257]`,
`noteheadHalf` `[0.903, −0.368]`; and `fClef` `[2.534, 0.153, 0.148]` —
**all positive**, which is the same fact from the other side: no counter,
three filled components. The
fill **rule** is therefore not the load-bearing property here; **preserving
every filled contour and every bounded counter** is. Ruling A criterion 1 is
amended to say so.
* **Per-system ownership is populated only by the real solver.** `Engraver`'s
casting module builds one `PrimitiveIndices` per system
(`engrave/src/casting.rs:1185`); the **stub solver publishes everything
@ -474,22 +500,42 @@ is a hard
criterion, and discovering its absence in round 3 would waste three rounds —
which is the whole claim of "cheapest disqualifier first".
**Round 1 — criterion 1, fill correctness (HARD).** Each candidate draws
`gClef`, `fClef`, `timeSig8`, `accidentalFlat` from `BravuraGlyphCatalog`'s
typed outlines under pin 4's configuration. **Precommitted oracle (pin 13):**
for each glyph, ≥3 must-be-ink and ≥3 must-be-background sample points in
staff-space coordinates, **derived programmatically** by an even-odd
point-in-path test over the `PathCommand` outline — not chosen by eye — each
required to lie **≥8 device pixels from any outline edge** at the pinned
render transform, so antialiasing cannot explain any result.
**Every background point must lie inside a bounded hole** — enclosed by the
glyph's outer silhouette yet unfilled under even-odd — asserted by the
derivation, not assumed. Background points merely outside the silhouette test
nothing: a renderer that fills holes solid passes them trivially, which was
revision 2's false-pass path. The oracle file, the render transform, and the
expected class per point are committed before any candidate renders.
*Fails hard:* holes that are not holes — the disqualification Ruling A
anticipated.
**Round 1 — criterion 1, compound-path fill correctness (HARD).** Each
candidate draws five glyphs from `BravuraGlyphCatalog`'s typed outlines under
pin 4's configuration, in **two check classes testing two different
properties**. All sample points are **derived programmatically** by
point-in-path over the `PathCommand` outline — never chosen by eye — and every
point must lie **≥8 device pixels from any outline edge** at the pinned render
transform, so antialiasing cannot explain any result.
* **Hole checks — `gClef`, `timeSig8`, `accidentalFlat`, `noteheadHalf`.**
≥3 must-be-ink and **≥3 must-be-background points, each inside a bounded
hole** — enclosed by the glyph's outer silhouette yet unfilled — asserted by
the derivation, not assumed. A background point merely outside the silhouette
tests nothing, since a renderer that fills holes solid passes it trivially.
`noteheadHalf` is in this set deliberately: it is a frequently repeated and
semantically consequential glyph — a filled counter renders half notes as
quarter notes, a notation error rather than a cosmetic artifact. (It is not
the single most-drawn glyph; `noteheadBlack` is, and it has no counter.)
* **Disjoint-component check — `fClef`.** It has **no bounded hole** (bowl plus
two solid dots), so it carries **no background requirement** and tests the
other half of compound-path correctness: **≥1 ink point inside each of its
three filled subpaths** — body and both dots — at the same clearance floor,
each tagged with its `subpath_index` so the oracle proves coverage of every
component rather than three generic ink points that could all land in the
bowl. A tessellator that keeps only the largest contour fails here and would
pass every hole check.
**The oracle's status model must be explicit, not inferred from an absence.**
Each glyph carries a requirement enum (or at minimum `background_required:
bool`) *plus* an overall satisfied status. `fClef` passing with zero background
points is a **satisfied** result under its own requirement class; recording it
only as `background_satisfied = false` would make a correct outcome
indistinguishable from a failed one.
The oracle file, the render transform, and the expected class per point are
committed before any candidate renders. *Fails hard:* **any bounded hole
painted as ink, or any required filled subpath omitted.**
**Round 2 — criterion 3, text (HARD).** W3 §5's five checks against pin 8's
stand-in and pin 10's reference emitter: (1) **faithful consumption** — draws

24
spec/PLAN_EDITOR_APP.md
View File

@ -532,11 +532,25 @@ result rather than whether it can shape correctly.
What this ruling does **not** pin: the toolkit/tessellation stack. The T4
spike decides it, bounded by these recorded criteria:
1. **Fill correctness:** `epaint`'s filled-path tessellator does not implement
even-odd/nonzero fill for paths with holes — and glyphs have holes — so
"just egui shapes" is likely out; the candidates are lyon-tessellated
meshes inside egui, or a vector renderer (e.g. Vello) behind a window
shell.
1. **Compound-path / inner-subpath fill correctness** (**amended 2026-07-28**;
was "fill correctness", framed around the fill *rule*): `epaint`'s
`PathShape` is a **single point loop** documenting "Fill is only supported
for convex polygons" (`epaint-0.35.0/src/shapes/path_shape.rs:14`), so it
cannot express **compound-fill / subtractive-hole semantics** — a
`Shape::Vec` can group several loops, but grouping paints them, it does not
subtract a counter from its enclosing contour. Glyphs are multi-contour with
bounded counters, so "just egui shapes" is out; the
candidates are lyon-tessellated meshes inside egui, or a vector renderer
(e.g. Vello) behind a window shell.
**The fill rule is not the load-bearing property, and the original wording
implied it was.** Bravura's contours are correctly oppositely wound (measured
signed ring areas, e.g. `gClef` `[8.702, −0.691, −1.803, −0.509]`), so **even-odd
and nonzero agree on every bundled hole**. What a candidate must preserve is
**every filled contour and every bounded counter**: a tessellator that keeps
only the outer contour paints counters solid, and one that keeps only the
largest drops disjoint components such as `fClef`'s two dots. The T4 spike's
Round 1 tests both properties separately
(`spec/CONTRACT_EDITOR_T4_SPIKE.md`).
2. **Staged latency, not end-to-end latency:** reduce / engrave / scene-build
/ paint measured **separately** on large multi-system fixtures against
Chapter-10-style budgets (Fact 8). A toolkit verdict from an end-to-end