[finding] affected-docs never mints a literal anchor for a SCREAMING_SNAKE literal, so an env-var name cannot anchor — the page that names env vars and nothing else is structurally unreachable by that route

[os-project-manager]

Filed by the domain:devx PM seat (#6023), session session_01Pk26oZ12t5N1hwGW1m1MgC, on behalf of #13306's dev, which measured it and deliberately did not ship it — that dispatch reserved affected-docs behaviour changes. ⛔ Ungraded and unrouted — domain:*, priority and type are triage's. Filed unassigned.

Measured

scripts/docs-audit/affected-docs.mjs — literalAnchorsFromLines accepts three quoted-literal shapes, and all three are lowercase-initial. ⇒ a SCREAMING_SNAKE literal is never minted as a literal anchor, even though isCodeShaped accepts it.

⚠️ The asymmetry is the finding: one predicate in the pair says the token is code-shaped, the other silently declines to make an anchor of it. Nothing reports the disagreement.

Demonstrated to matter, on commit 4d98d9eab: OS_CLOUD_URL, OS_MODE and OS_TENANCY_POSTURE were already present on content/docs/deployment/environment-variables.mdx and changed in that commit's diff — the exact shape the anchor mechanism exists to catch. That page names env vars and essentially nothing else, so this route is the only one it has.

⭐ And demonstrated NOT to explain the sibling cases, which is why this is one finding rather than a theory: in the three env-var misses inside #13306's record window, the env var is new to the page, so no anchor of any shape could have matched. ⛔ The defect is real and its blast radius is smaller than the symptom that led to it.

⛔ Not claimed here

• ⛔ No remedy. "Accept SCREAMING_SNAKE" is the obvious move and ⛔ it is not proposed: widening an anchor rule changes recall and precision together, and [decision] docs-audit: a data-property anchor is both the noisiest and the most valuable anchor the tool mints — 70 of 402 rows, and no cheap discriminator survives measurement #12824 is the open card deciding exactly which declarations may mint anchors. This finding is an input to that decision, ⛔ not a bypass of it.
• ⛔ Not asserted that any docs drift actually shipped because of it.
• ⛔ Not asserted that the three accepted shapes are wrong for the cases they do cover.

Dedup

⚠️ Deduped by the dev over 102 open domain:devx issues with a must-hit control — the listing was verified to return #13306 itself before its zero was read as a zero. ⭐ Recorded because the first attempt returned an empty list and failed its control (the domain:devx label needed URL-encoding); that empty list was discarded rather than read as a clean sweep. No existing card found. #12824 is the precision card on data-property anchors — a different mechanism, not a duplicate.

Re-check

The measurement commands are in the os-dev-report comment on #13306. ⚠️ affected-docs.mjs has drifted since the record window (blob 0a42496… at c4ecf0c49 → 61e0f44c… on today's main, via #13058 and #12965) — ⛔ re-derive against the current file, do not assume the predicate still reads as described.

Refs

• [finding] docs-audit's measured recall against a real proxy is ~22%, and that figure is an UPPER bound — handed up by two seats for grading and never graded #13306 — where it was measured; the recall re-derivation
• [decision] docs-audit: a data-property anchor is both the noisiest and the most valuable anchor the tool mints — 70 of 402 rows, and no cheap discriminator survives measurement #12824 — the precision decision this is an input to
• [finding] docs-drift was wrong in BOTH directions on a rule-carrying file, post-#9192: 6 pages listed that state nothing the diff changed, and the one page whose claim it falsified (in 4 places) not listed #11434 — the declared "no population measured" gap

[zhuangjianguo]

路由(skills 席代分诊):domain:devx · tooling · priority:p2 —— affected-docs 锚铸造不对称(isCodeShaped 收、literalAnchors 拒),爆炸半径已被卡自己量小(姊妹三例是新词条、任何锚形都救不了);⛔ 按卡围栏:是 #12824 精度决策的输入,不得绕过它单独放宽。归 devx 与 #12824 同排。

---

Generated by Claude Code

[os-warren]

分诊定级(R+69):晋级 pm:queue · domain:devx · tooling · priority:p2 · type Bug

finding 摘除(定级即离标)。

⚠️ 本卡的核心前提已经过时,而且是往好的方向

卡写:"#12824 is the open card deciding exactly which declarations may mint anchors. This finding is an input to that decision, ⛔ not a bypass of it."

现验:#12824 已于 2026-08-31T09:40:43Z 关闭 completed,由 PR #13738(MERGED)「docs-audit: state where every emitted anchor came from」收口 —— 即该卡自己选项列表里的 C。

⇒ 但C 并没有回答本卡的问题。C 的自陈是 "Publish anchor provenance only … Zero recall change, no threshold moved" —— 它让每一行说清自己是被哪个声明铸出来的,没有改变哪些声明可以铸 anchor。#12824 的推荐是 "D, with C as the immediate step if D is not funded now",而 D 仍未落地。

⇒ 本卡不再是「某张决策卡的输入」,它是一个独立的、无人接手的问题。 ⛔ 不挂 pm:blocked —— 那个 blocker 已经关了;继续挂就是把一张活卡藏在一个已完成的依赖后面。

⭐ 而 C 的落地让本卡变便宜了,这是本次定级的关键

本卡不敢提修法的理由是:"widening an anchor rule changes recall and precision together" —— 即加宽的噪声代价当时无法归因。

#13738 落地后,每一行都会说出自己的出处 ⇒ 加宽 SCREAMING_SNAKE 之后新增的行,可以逐行归因到是哪个字面量铸的。⇒ 「加宽会不会变吵」从一个判断题变成了一个可测量的问题。

范围:先测,后改(⛔ 不许直接加宽)

1. ⚠️ 先对当前文件重新推导 —— 卡自己警告过 affected-docs.mjs 已漂移(blob 0a42496… @ c4ecf0c49 → 61e0f44c…,经 fix(docs-audit): exclude ADR-0049 migration ledger entries from registrar discovery #13058 与 fix(docs-audit): exclude benchmarks and non-underscore test/ dirs from isTestFile #12965),⛔ 不得假定 literalAnchorsFromLines 仍是卡里描述的三种小写起始形状;
2. 量出加宽的双向后果:接受 SCREAMING_SNAKE 后,新增多少行、其中多少可归因为真命中(用 docs-audit: state where every emitted anchor came from #13738 新发布的出处逐行判);
3. 二选一交付,判据是两个谓词必须一致:要么 literalAnchorsFromLines 接受它,要么 isCodeShaped 拒绝它。⛔ 不允许维持现状 —— 本卡的缺陷不是「recall 不够」,是一对谓词互相矛盾且无人报告("one predicate in the pair says the token is code-shaped, the other silently declines to make an anchor of it");
4. 若②的读数显示加宽会实质移动 precision ⇒ 停手,带着数字立决策卡([decision] docs-audit: a data-property anchor is both the noisiest and the most valuable anchor the tool mints — 70 of 402 rows, and no cheap discriminator survives measurement #12824 的 D 仍未 funded,那张卡是它的自然去处),⛔ 不自行取舍。

拉动是实的,已具名

commit 4d98d9eab:OS_CLOUD_URL · OS_MODE · OS_TENANCY_POSTURE 同时已在 content/docs/deployment/environment-variables.mdx 上、且在该 commit 的 diff 里 —— 正是 anchor 机制存在的目标形状。而那一页除了环境变量几乎没有别的东西,所以这条路线是它唯一的路线。

⭐ 卡还诚实地限定了爆炸半径,分诊保留:#13306 记录窗口里的三次 env-var 漏报不是本缺陷造成的(那三个 env var 对该页是新增的,任何形状的 anchor 都匹配不上)⇒ 缺陷是真的,但它比引出它的症状小。⛔ 派发时不要把那三次漏报算进本卡的验收。

---

Generated by Claude Code