v0.2.0-beta.2

v0.2.0-beta.2 — the freshness gate can fail

A survey of this repository by its own engine found seven defects in the
conspectus machinery, all of one shape: a gate reporting green because it had
nothing to measure.

Staleness lived on a table no code path ever wrote, so every derived signal was
pinned at zero and the published report told its readers "the conspectus is
fresh" while the survey sat 23 commits behind HEAD with 47% of files
unclassified. The file ledger was never reconciled against the working tree, so
additions were invisible and rows for deleted files kept asserting they had been
examined. The release gate checked only that the roadmap agreed with itself, so
beta.1 went on asserting it was unpublished while npm served it as latest.

Staleness now derives from the file ledger, which the survey always populates,
and each row is compared against the commit at which it was examined — so the
signal has a denominator by construction and re-derives in both directions.
Scope reconciles against git ls-files, and a ledger row whose file is gone can be
retired without discarding the subsystem's survey record. Activation probes are
time-bounded. The evidence vocabulary is unified across the tools and the
methodology. Release state is reconciled against the repository's own tags.

Every repair was gated by a check written first and observed failing.

This build's own conspectus reports 84 stale files against 502 scoped. That
number is the feature: yesterday it was a structural zero that could not move.

Product efficacy remains unestablished, and the supported operating envelope
remains the Codex CLI exec surface measured under A25.