Lane attribution has two instruments and they disagree

The two readings

progress.py's track resolution is a cascade (~line 630):

1. Track R in the decl line (the Type + Track body bullets);
2. slug prefix — if self.slug.startswith("feature-track-t-"): return "T";
3. Track T in decl prose, if frontmatter track: is empty or T;
4. only then explicit = normalize_track(self.fm.get("track", "")), and failing that, a Track bullet from the body.

So the ranker lanes a ticket with no track: field at all, from the slug or from prose. A consumer that reads the frontmatter field directly gets nothing. Both are behaving as written; they are answering different questions.

Scale — this is not two tickets

Open tickets with no track: field:

Two are in working/ — actively claimed, and their lane is inferred rather than declared.

How it surfaced, and the prior instance

The 2026-09-02 backlog sweep read the frontmatter field. feature-t-nilpy-cpython-differential-fuzzer and feature-twatch-full-tier-coverage-age have no such field, so they were nearly swept as unowned while the ranker had been treating them as Track T throughout. They landed in low-prio/, which is recoverable — a folder that deleted, or a dispatcher that skipped them, would not have been.

progress.py:637 already records the same disagreement running the other way: a prose "Track T" mention on a fuzzer-filed bug overrode the real lane and "stranded three fuzzer-filed A/P bugs under T on 2026-07-15" (bug-t-progress-track-detection-prose-mention). That was patched by letting an explicit field break the tie — which fixes the case where the field exists. It does nothing for the 33 where it does not.

Why it is worth fixing rather than tolerating

A lane is who gets dispatched. An inferred lane is invisible to anything that is not the ranker, so every new tool that wants to ask "whose is this?" either reimplements the cascade or silently disagrees with it. That is the same shape as chore-t-split-lib-test-into-jobs-that-name-what-failed and the would_pin/permission reading: the artefact is not lying, it is answering a question other than the one being asked.

Suggested fix

Make the cascade's answer durable rather than recomputed: one command that writes the resolved track back into frontmatter for any ticket lacking it, run once over the 33, plus a check rule that a ranked ticket must carry an explicit track:. That collapses two instruments to one without changing any current attribution — the cascade stays as the resolver, it just stops being the only reader that knows.

Positive control if this is taken: feature-t-nilpy-cpython-differential-fuzzer must come out T (the ranker's current answer), and none of the 33 may change lane as a side effect of being written down.