v0.85: every lane now carries a measured outcome, and the no-show decline is refuted

[avrabe]

v0.85: every lane now carries a measured outcome, and the no-show decline is refuted

Each outcome below was MEASURED before it was written and each lane's refuting
command was RUN first. With this commit every artifact in the release directory
records an outcome — none is silent.

NOSHOW3 — REFUTED. The decline this lane was asked to consider is not
available. Window stated with the result, because a window is part of a
population claim: two hundred workflow runs spanning a day and a half, jobs
fetched for ninety-five. Whole-run cancellations are their own class and
excluded. Two genuine LONE no-shows remain, each a single unstarted job inside
a run that FAILED rather than was cancelled. So "empty across two releases" was
a WINDOW-AND-CLASSIFIER ARTIFACT, not a property of the system. The decision:
do not decline.

AND THE FIRST TALLY WAS RIGHT BY ACCIDENT, recorded as a defect rather than
quietly corrected. The first classifier grouped unstarted cancelled jobs by
EXACT completion timestamp. In one run fifteen such jobs share a timestamp and a
coverage job sits fifty-five seconds later; exact grouping reported it as a lone
no-show, when that run's own conclusion is cancelled with thirty-two jobs
cancelled — one operator action whose teardown is not atomic. The classifier's
population was "jobs sharing an exact timestamp"; its subject was "jobs
cancelled by one action". Two populations, inside a classifier written FOR a
lane about classifying. After fixing it to cluster on PROXIMITY and to give
whole-run cancellation its own class, the total is unchanged but NEITHER
instance is one the broken version named.

An empty subclass is reported rather than hidden: "operator-cancel cluster
inside a run that is not itself cancelled" is zero in this window, legitimately,
so a later tick does not read that zero as potency.

PINDEBT10 — the escape is confirmed live with a byte-identical paired control.
Baseline green; a twelve-entry suppression table under a name OUTSIDE the
tripwire's pattern leaves the ledger GREEN; the SAME BYTES under a matching name
RED it, and the failing claim is the pin-debt entry itself. Only the NAME
differs. Both layers are name-scoped, so one rename defeats the explicit table
list and the tripwire at once.

RUNNERSHAPE2 — the three states are confirmed with a distribution over more
than two thousand jobs: a real name in the overwhelming majority, the EMPTY
STRING in a double-digit minority, and null-or-absent in a smaller one. All
three populated, which is why a null-coalescing default is the wrong shape: it
never fires on the empty string.

TWOSOURCE2 — not refuted, and the split is WIDER than the plan states. The
contradiction reproduces inside ONE report after a single deliberate uncommitted
change, with a clean-tree control passing. But the release-execution step alone
answers from THREE source kinds — an existence check against the ref, a
check-run conclusion from the GitHub API in retro mode, the pin sweep executed
against the working tree otherwise, and the changelog from the ref — so it is
per slot AND per mode over three sources, not two. A two-valued source label
cannot describe these slots. And the third source carries its own vacuity: an
absent check-run conclusion maps to NOT-DERIVED rather than a failure, so an API
outage makes that slot ABSTAIN.

ARCHMODEL4 — deferred, filed rather than waived. The ordinal is derived BY
ARTIFACT IDENTITY from the git tree with zero duplicate identifiers and a
non-empty population, stated with its commit. The upstream blocker is OPEN and
is that the architecture tool accepts property associations from UNDECLARED
property sets silently — a typo'd property indistinguishable from a satisfied
one, which is exactly the class this programme refuses to build evidence on. The
tag conjunction was verified by the conformance gate's own step 1-2 line naming
this artifact, not by reading the tags.

NPM4, COMPLIANCE5, LADDER7 — all three deferred and reported as BLOCKED OUTSIDE
THIS REPOSITORY rather than re-measured as progress. NPM4's one owned figure was
DERIVED from the tag list and has widened by exactly one since the previous
release's derivation, which is the only legitimate reason it could move.
COMPLIANCE5's shipping half holds, re-read at the SETTLED publication rather
than a queued one. LADDER7's corpus volume does not resolve at all — only system
volumes are mounted — and ATTEMPTING PASSWORDS STAYS DECLINED; this lane did not
attempt one.

JESSDIVERGE7 — pending, and gated on correspondence approval rather than on
measurement. The analysis is finished and the reply is drafted and HELD; nothing
was posted. The honest residual is carried as unexplained rather than guessed
at: the excess over the pure-gravity integral grows with tick count while the
constant is correct to single precision, so the growth is not the constant.

Gates at this commit: rivet validate PASS; status_evidence rc=0; claim_check
75/75; verdict-prose 0 disagreements with the population risen by exactly the
number of outcomes added — the real check, since an unclassified lead word
REMOVES an artifact from the population rather than failing, so a smaller rise
would have meant an escape.

Refs #1183, #1484, #1451, #1428, #1136, #1460, #1453, #1432, #1436.

Co-Authored-By: Claude Opus 5 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

🤖 Generated with Claude Code

https://claude.ai/code/session_01YJK5LZZEkV5smCY1jKn18L

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!