Registers **IR-146** in `docs/INFRA-REGISTER.md`. Docs-only; no code, no REQ activation.

**Measured** (checker 0.4.1 = CI pin, isolated worktree at `6c95d691`): `traceable-reqs lint` emits **1496 `[must]` findings over 825 of 949 requirements (87%)** — `length` 819, `contains-and` 674, `tbd-todo` 3. The coverage gate `check` is exit 0 at the same sha (949 headers == 949 ids).

**Why it is an infra item and not a lane defect.** This registry's titles are deliberately spec-paragraphs — the requirement text IS the contract a lane builds against — while the linter's default criteria assume a short imperative title. So the audit's population is nearly the whole corpus and its per-lane signal is zero. The CI step is `./traceable-reqs lint || true`, so nothing fails and the only reader is whoever opens the log.

**Recorded with the mistake attributed to me:** at the #331 H3 gate I routed two such findings on `REQ-TEST-PEER-RIG-FIXTURES` to hertz as lane carry-forward before measuring the population, and withdrew it the same day. The entry states that so the next gater does not re-derive it at cost.

Fix candidates are listed but NOT ruled: (a) configure the criteria this registry wants, (b) split short `title` from a long contract field (949-row migration + upstream ask), (c) state in `docs/TRACEABILITY.md` that the findings are population-wide and never routed to a lane. (c) is the cheap floor either way.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01SgutL1VJbbSwoa7NSsKuHp