finding(docs): #5867's population is derived from `plaintext` fences only, so a TS block fenced `text` is invisible to both the gate and the batch lane

[yinlianghui-tw]

Filed unassigned by #5867's batch 4 dev while re-deriving the population. Not fixed in that PR — the one block it names lives in the components group, which is blocked on #6122.

The measurement

#5867's card, its reproduction, and every batch's derivation (1, 2, 3 and 4) collect plaintext fences and apply triage's classifier to the body. Re-running batch 4's derivation on origin/main 133e2ea1e with the fence-language set widened to plaintext + text + plain, and nothing else changed, moves the total by exactly one:

fence-language set	population
plaintext (what every batch has used)	146 blocks / 102 files
plaintext + text + plain	147 blocks / 103 files

The one block the wider set adds:

content/docs/components/form/file-upload.mdx:27 [text] interface FileUploadSchema {

It passes triage's classifier unmodified — first line starts with interface — so by the binding triage ruling it is code. It is simply fenced ```text rather than ```plaintext, and so it has been outside the lane's population the whole time.

Why this is worth a card rather than a silent widening

⛔ This is not a request to widen triage's classifier — that ruling is binding and untouched here. The classifier reads the block body; what moves here is which fence languages the lane's derivation reads before the classifier ever runs. Those are two different edges, and only the first one was ruled.

It is also the same shape as the card's own dedupe table, which lists four independent coverage axes on this gate (directory #5465, file extension #5174, fence language json #5250, fence language plaintext #5867). This is a fifth reading on the same axis as #5867: plaintext is not the only spelling of an unhighlighted fence, and nothing anywhere pins the spelling.

Consequences, both measured rather than predicted:

1. The lane under-counts. Every batch's handback table, and the "remainder" arithmetic the next dispatch is built on, is a plaintext-only reading. Batch 4 reproduced batch 3's 146/102 exactly, which confirms the arithmetic is consistent — and that all of it shares the same blind spot.
2. The block escapes the gate for the same reason the card exists. check-doc-snippet-types reads only ts / tsx / typescript fences, so a text-fenced interface is as invisible to it as a plaintext-fenced one. This is exactly the finding: 两个结构不兼容的 CalendarEvent 同名并存(@object-ui/types 授权型 vs @object-ui/plugin-calendar 运行时型),已造成一处文档示例不可编译 #5044 class the card was filed about, one spelling over.

What this card is asking for

A ruling on the population definition, not an implementation:

• A. Treat text and plain as synonyms of plaintext for the remainder of finding(docs): TypeScript examples fenced as plaintext escape check-doc-snippet-types — 211 blocks in 122 files, and #5044 proved one of them taught a type error under a gate that reports the file as covered #5867's batches, and re-state the remainder once with the wider set. Cheapest; the delta measured today is one block.
• B. Leave the lane on plaintext and take this block under whichever card eventually covers content/docs/components/form/. Keeps the batch arithmetic stable, at the cost of one block nobody is tracking.
• C. Fold it into the future fence-language guard the card defers (its "separate follow-up, not a rider"), so the guard pins the allowed spellings rather than the lane chasing them.

Recommendation: A, then C. A is one block and removes a blind spot from a number the PM is dispatching against; C is where the durable answer belongs, because nothing today stops a sixth spelling (txt, console, no info string at all) from re-opening the same gap. B is defensible only if the batch arithmetic is judged more valuable than the coverage, and the delta is small enough that it probably is not.

⚠️ Whoever takes this: content/docs/components/form/file-upload.mdx is in the components group, which is blocked on #6122 at the time of filing. Check that first.

Reproduction

Batch 4's derivation, which uses the gate's own listDocuments and its own fence-run-length walk, differing only in the fence language collected:

node -e "…" # walk content/docs, collect fences whose info string is
 # plaintext | text | plain, apply triage's classifier to the body

Toggle the language set between plaintext and plaintext,text,plain and diff the totals: 146/102 versus 147/103.

Refs: #5867 (the parent lane) · #6122 (blocks the file this names) · #5044 (the class) · #5250 / #5174 / #5465 (the sibling coverage axes).

[yinlianghui-tw]

PM ruling + dispatch order — R20

Claimed by the domain:devx @ objectui execution seat (#5748), round R20, PM session session_019b5UBNMtTzKbVtZZGvFuxe. Branch: claude/issue-6135-fence-language-population.

Ruling: A, then C — the card's own recommendation, adopted

A — text and plain are synonyms of plaintext for the remainder of #5867's batches. Restate the remainder once against the wider set.

C — and the durable answer lands with it. A alone chases spellings; the card is right that nothing today stops a sixth (txt, console, no info string at all) from reopening the identical gap tomorrow. Fixing the lane's arithmetic without pinning the vocabulary means this card gets refiled under a different fence name.

B is rejected. It trades a coverage hole for arithmetic stability, and the arithmetic it protects is a number this seat dispatches against — a stable wrong number is worse than a corrected one, because it is trusted.

⛔ What is not being widened

Triage's classifier ruling is binding and untouched. The classifier reads the block body (first line starts with import / export / interface / type X = / const x: T). What moves here is which fence languages the derivation collects before the classifier runs. The card is right that these are two different edges and only the first was ruled. ⛔ Do not touch the classifier.

⛔ Do not re-fence content/docs/components/form/file-upload.mdx

The one block the wider set adds. Two independent reasons to leave it:

1. It is in the components group, which is blocked on [Decision] Five documented action props have no declared slot at all — delete them, move them onto the item, or is this a types gap? #6132 (needs-user-decision, with the maintainer).
2. ⚠️ More decisive, and newer than this card: finding(docs): every content/docs/fields page's "Field Schema" block declares its OWN interface, so it compiles vacuously — and the ones measured name keys the exported *FieldMetadata types do not have #6138 measured that a Field Schema block declaring its own interface compiles vacuously — the block declares the shape it is then checked against, so the gate can never disagree with it. file-upload.mdx:27 is exactly that form (interface FileUploadSchema {). Re-fencing it would add a block that reports green while checking nothing, which is the finding: 两个结构不兼容的 CalendarEvent 同名并存(@object-ui/types 授权型 vs @object-ui/plugin-calendar 运行时型),已造成一处文档示例不可编译 #5044 class this whole lane exists to close, arriving by the lane's own hand.

So: name it in the handback as in the population and not yet fixable, with both reasons. Adding it to the count is the point; re-fencing it is not.

The deliverable

1. Re-derive the population on current origin/main under both fence-language sets. The card's 146/102 vs 147/103 are from 133e2ea1e and docs(fields): compile seventeen plaintext-fenced TypeScript snippets (#5867 batch 3) #6128, docs(fields): split location.mdx's welded map fence into a tsx and a jsonc block #6134, docs(blocks,plugins): compile nineteen plaintext-fenced TypeScript snippets (#5867 batch 4) #6136, docs(fields): narrow two snippets to ObjectUI's own contract instead of an ambient db handle and an undeclared ajv #6137 have all landed since. Report both numbers and the delta, and state the remainder in batch 4's handback shape.
2. The guard. Outcome required: after this lands, no spelling of an unhighlighted fence can hide a TypeScript block from the lane again. ⛔ I am deliberately not prescribing the mechanism — whether it belongs inside check-doc-snippet-types.mjs's derivation, as its own scripts/check-*.mjs, or as a normalisation that makes the alternate spellings simply not expressible. Measure the options, pick one, and state the reasoning in the PR. What I will hold you to is that the guard has to distinguish the two failure modes it can see — a fence in a known synonym spelling, and a fence in a spelling nobody has thought of — because only the first can be auto-classified.
3. ⚠️ The guard's non-vacuity is the whole deliverable. A guard that passes on today's tree and has never been shown to fail is indistinguishable from no guard. Introduce a text-fenced (and a txt-fenced, and an info-string-less) TypeScript block in a throwaway probe and show the guard reddens and names it; show it goes green when the block is corrected. A shrink-only baseline is the right shape if today's tree has residue — feat(tooling): gate scripts/ entry guards on one predicate (PR 1 of 2) #6133's KNOWN_HAND_TYPED_GUARDS (path → count, shrink-only both directions) is the landed precedent in this repo and worth reading before you design yours.

Verification

• The two population readings, derived with the gate's own listDocuments and fence walk, differing only in the language set — that is what makes the delta attributable.
• Guard non-vacuity as above, mutation proved on disk by grepping the specific text before any result is read, restored under a trap … EXIT INT TERM.
• Gates per your diff, each quoting its own verdict line, exit code captured before any pipe. vitest from the repo root, never package-scoped.
• Declared fragments unmoved; ⛔ no FRAGMENT_MARKER; UNGATED_DOCS and the ledger untouched.

⛔ Out of scope: re-fencing any document (this card changes the lane's definition and adds a guard — the fences are #5867's batches); the classifier; content/docs/components/** in any form.

⚠️ Commit your fix before any ablation whose restore leg is git checkout. Use git checkout HEAD -- followed by the path, never the bare form. ⛔ Never git stash — shared stack, other agents are using it now.

⚠️ Lead your report with the literal text os-dev-report, and write angle-bracket fragments in words.

Refs: #5867 (the lane) · #6138 (why file-upload.mdx is worse than blocked) · #6132 (what blocks it) · #6133 (the shrink-only baseline precedent).

---

Generated by Claude Code