docs: the handoff reports what the trees actually contain

[ShocOne]

handoff.md stopped at PR #86; main is at #109. Re-measured against the three generated trees at 30b23ac and rewritten to what they contain.

Stacked on nothing — independent of #110, though both came out of the same audit.

The headline was false

All fifteen merged PRs improved coverage, not correctness. … the second has not been started.

It started at #91, #93, #94 and #96. Constraint keywords parse, bounds and patterns become validators, format: password / writeOnly become Sensitive, deprecated becomes a DeprecationMessage, and a documented default fills the response.

The census was taken at the wrong layer

The old file listed ten framework symbols "verified against current main" as zero. Measured in the generated trees, four were not:

Symbol	Claimed	Actual
Sensitive	0	53
DeprecationMessage	0	171
int64validator.Between	0	180
RegexMatches	0	21
UTF8LengthBetween	0	42
LengthAtLeast / LengthAtMost	0	138
Default	0	18

The census had been taken by grepping this repo for the symbol. internal/emit/render_constraints.go spells a pair of bounds as fmt.Sprintf("%sBetween(%v, %v)", …), so the string never appears here and always appears in the output. That lesson is now written into the file, and into CLAUDE.md in #110.

Still genuinely zero, and each is a mapping.md row: WriteOnly, SetAttribute, SetNestedAttribute, SemanticEquals, RequiresReplaceIfConfigured, and any config validator at all.

What has still never happened

No audit has run against any of the three pilots. All three revised specs contain zero x-tfpfgen-* extensions, and no scratch tree has an audit/ or spec/corrections/ directory. Every observation-driven shape is unexercised — including the conditional validators, which is why row 5 reads "emitter exists, no tree contains one".

Counts

Refusal counts are gone; docs/emittance_tracker.md owns those, and the old file's 1600 disagreed with the tracker's 1507.

The presence census stays, because it is not a tracker row — the tracker holds tree files, entities by kind, and refusals by stage. It is stamped with the commit and date, and it moved the right way for the first time:

	optional	computed_optional	ratio
first measured	2,221	64	35×
after coverage work	3,216	65	49×
now	10,574	670	15.8×

Absolute figures are not comparable across rows — union variants brought each branch's own fields into the tree. The ratio is the number that means something.

Other corrections

• mapping.md "is not in the repo" — it is, and has been since docs: commit the behaviour-to-schema mapping #87.
• docs/contract.md:168 now resolves to line 175; the citation is replaced with the quote.
• prune.go "is at 799 lines, the next change fails the gate" — decomposed in refactor: decompose prune.go by protocol #97, now 351.
• Path-parameter type mismatch down from 250 refusals to 1 (feat: parse an integer path parameter out of a string attribute #88); union refusals from 90 to 13 (feat: a union becomes one attribute per variant where nothing writes it #109).
• Owner-blocked list: variant naming is settled (feat: a union becomes one attribute per variant where nothing writes it #109, glossary variant attribute); the undeclared-response-schema observation and mapping.md rows 1 and 8 are still open, and compile.go:116 still refuses for want of row 8.

Verification

Trees confirmed current before measuring — tfpfgen provider verify on ThousandEyes reports 1912 files, no drift, matching the tracker's row. Refusal totals independently recomputed from unsupported.json: 1507, split exactly as the tracker records.

bash scripts/repo_hygiene_gate.sh   # clean

One question for you

Should the presence census and the emitted-symbol census move into docs/emittance_tracker.md? The tracker currently defines itself as tree files, entities by kind, and refusals by stage, so I left them here rather than widen it without asking. They have the same staleness problem the tracker exists to solve.

🤖 Generated with Claude Code