Architect drift review over 2add97e..acd3d2c, plus the regression gate.
What holds, confirmed against the diff and the contracts:
- **C1 bit-identity, structurally.** `SeriesFold` is the single arithmetic source
for both sides of the migration, and the entry point's new positional read is
line-for-line what the member path already does. `RunMetrics` carries exactly
the four fields read — no value silently dropped.
- **Ledger discipline.** The superseded C08 clause sits verbatim in its history
sidecar, appended newest-last, with nothing above it rewritten. C10, C18 and
C22 were renames carrying no superseded substance, so they owe no sidecar
entry; the asymmetry follows the substance rather than being an oversight. No
further live contract was falsified: the remaining `--trace` mentions are
absence framings, and the remaining `reduce` mentions name the run mode, which
the cycle's rule excludes explicitly.
- **C8 / C27 seams untouched.** Under folded delivery the r-equity branch is not
built at all, so no zero-output node is left dangling; the declared-tap bind
and its shared wiring pair are unchanged, and the trace channel still has one
consumer.
Drift found and resolved in this commit:
- **[high] A twin comment the sweep could not see.** `crates/aura-runner/src/measure.rs`
distinguished the bare measurement run by claiming it is "where the measured
O(cycles) retention is removed" — a contrast against the single-run path that
this cycle made false. Corrected: the R leg is *absent* there rather than
merely folded. Notably the cycle's prose sweep could not have found it, since
the sentence contains none of the words the sweep keyed on. That is the third
instance this cycle of a keyword-driven enumeration under-covering prose a
change falsified; the lesson is that such a sweep needs a semantic pass beside
it, not a longer word list.
- **[medium] A deferral recorded only in ephemeral places.** The two auxiliary
senders left dead in the wrap's signature were documented in a git-ignored spec
and an issue comment, so `closes #308` would have dropped them from the forward
queue. Filed as its own tracker item instead.
- **[low] Out of scope, agreed.** A CLI test's two-part property claim describes
grammar its body no longer drives. The site is now retirement-framed and
spec-conformant; the rest is a test-scope correction, not a prose fix.
Regression gate: `cargo run --release -p aura-bench -- run` exit 0, all five
fingerprints OK — `engine_throughput`, `ingest_throughput`, `campaign_sweep`,
`campaign_heavy`, `cli_fixed_cost`. No baseline updated, so no ratify statement
is owed. Timing deltas and a `campaign_sweep` peak-RSS move of +9.0% rode under
a `WARNING: loadavg 14.9` emitted by the harness itself; the campaign path is
untouched by this cycle and its fingerprint is unmoved, so the figure is host
noise rather than a metric shift.
Suite 1652 passed / 0 failed (exit 0); clippy exit 0.
refs #308
claude
7c27c070f4