docs: run the itd-88 coverage experiment and close spc-3

[REPPL]

Runs the itd-88 (spc-3) lifeboat coverage experiment and closes the record behind it — the final item of the current run queue. No new implementation code: the probe/coverage/pack surface ships; the experiment and its record are the deliverable.

Chained PR: base is fix/iss-112-114-116-walkfiles-class (#145) — deliberate: the queue ordered the walk fixes first precisely so this experiment's verdicts would be trustworthy, and the probe ran on the fixed walk. Merge order: #136 → #145 → this.

The experiment

• Probed this repository plus two public, record-less foreign repositories — spf13/cobra (Go) and psf/requests (Python) — and aggregated with disembark coverage. Both foreign trees byte-identical after probing; same-repo double-probe byte-identical (determinism).
• The headline number: this repository grounds 21 of 23 brief sections; cobra grounds 4, requests 3 — an abcd-native record is worth roughly 17–18 grounded sections over git-plus-conventions alone.
• The surviving section list is all 23 (nothing is blank in every repo), with product/personas carried as the one human-owned blank — confirming the intent's stated open question. One genuinely surprising finding: graveyard inverts the rich-vs-poor delta (busy foreign git histories ground it; the curated history here scores only partial) — the record premium is real but lives elsewhere.
• Full reading in the dated research note .abcd/development/research/2026-07-26-itd-88-coverage-experiment.md, including the corpus-size caveat (two foreign repos is a first reading, not a population).

The record close

• abcd spec close spc-3 shipped itd-88 and parked fidelity receipt rcp-4d07032fc6ab; the Role-1 fidelity review judged all 8 acceptance criteria with live command and test evidence: 6 MET, 2 MET_WITH_CONCERNS, 0 NOT_MET (concerns: graveyard grounding is narrower than the intent's four-signal parenthetical; the dry-run/pack parity holds via a hash chain and one shared code path, but the verbs ship as plan/pack rather than the press release's dry-run/to). The verdict is ingested into the Audit Notes with full provenance, diverged and missing claims included, for the maintainer's disposition — the verifier selects, the gate decides.
• impact: additive is hand-stamped in the ship commit (the Reconcile path cannot set it — known gap iss-126, captured earlier in this run).

Verification

• Gate run independently by the orchestrator after every commit: make preflight exit 0, record-lint exit 0.
• Step 0 (abcd intent ready itd-88) exited 0 before the work began.
• Review handling: the fidelity review is this item's review (record-only diff otherwise); no code changed, so no code reviewers ran — stated openly.