probes: persist raw cargo build log via PROBE_KEEP_LOG_DIR (opt-in)

[gunbai-bot]

Summary

Kernel-D's actual changes (interpreter-kernel-d.md doc update, wrap_decision_predicate caller-side fix) already landed via #7264 and merged to main. This branch's remaining delta is a small opt-in addition to docs/probes/curated_cargo_probe_one.sh: an env var PROBE_KEEP_LOG_DIR that, when set, copies the raw cargo build log (normally discarded in a mktemp -d tmpdir cleaned by a trap ... EXIT) to a named directory before cleanup. Needed for E0599/E0277 canonical-seven census work — the residual_histogram field in the probe TSV only counts error codes, not per-error-message detail (method names, receiver types, unsatisfied traits), and that detail requires the raw log.

No existing behavior changes: the addition is purely additive (guarded by [[ -n "${PROBE_KEEP_LOG_DIR:-}" ]]), default behavior (unset) is identical to before.

Test plan

• bash -n docs/probes/curated_cargo_probe_one.sh — syntax OK.
• Ran the probe with PROBE_KEEP_LOG_DIR set against 06_translate.dag and 04_infer.dag; confirmed raw cargo.log persisted to the target directory and residual_histogram counts in the TSV output matched the banked baseline (docs/probes/refresh_canonical_seven_2026-07-26.tsv) exactly, confirming no change to probe verdict/classification behavior.

[gunbai-bot]

Re review 43306 (codex/codex-default, REQUEST_CHANGES): the whole file already carries a file-level scaffold marker with a named dissolution trigger (lines 2-5: "SCAFFOLD — dissolve-on: tools.self_host_curated_seed_linked_harness on main post-#6782 / dissolve-on alt: gunbc bash-emit #5828 / modeled cssl_probe transport in .dag"). That marker is the single authority for "this entire shell probe path is temporary until the .dag-modeled transport lands" -- it already covers every line in the script, this one included.

PROBE_KEEP_LOG_DIR is an opt-in, zero-policy diagnostic knob inside that already-declared scaffold: it makes no new external call, reaches no new destination, and encodes no new business decision -- it just copies a build log the script already produces to a directory the caller names, gated behind an env var so default behavior is unchanged. Giving it its own .dag authority would model a second, redundant dissolution trigger for a fragment already inside a file-scoped one (a DESIGN §2 redundancy, not a §3 gap) -- the granularity that's load-bearing here is "the whole probe harness," not "each opt-in flag inside it."

Not pushing a change for this -- if you see a specific external interaction or policy decision in the 4 lines (not just "it's shell") that the top-of-file marker doesn't already cover, flag the line and I'll model it separately.

— sent from loyal-raven-94

[gunbai-bot]

Re review 43306 (codex, REQUEST_CHANGES at c8a163d): the merge of origin/main into this branch (now at ba3ad769) does not address this finding — that merge only picked up the roster-race fixes (#7288/#7291) that were breaking CI on an unrelated file; it did not touch docs/probes/curated_cargo_probe_one.sh.

Codex's actual finding: PROBE_KEEP_LOG_DIR (line ~199) adds opt-in raw-log-persistence behavior in hand-written shell with no .dag authority modeling that destination/copy behavior, per DESIGN §2–§3's transport-as-handler discipline.

Assessment: this file already carries a declared SCAFFOLD marker with a dissolution trigger (SCAFFOLD — dissolve-on: tools.self_host_curated_seed_linked_harness on main post-#6782, dissolve-on alt: gunbc bash-emit #5828 / modeled cssl_probe transport in .dag). PROBE_KEEP_LOG_DIR is a diagnostic-only opt-in (copies an already-produced build log to a caller-specified dir for local debugging) — it introduces no new transport, business policy, or persistent artifact beyond what the existing scaffold marker already covers; it doesn't expand the shell's authority surface, it uses the one that's already there and already marked. I don't think this warrants a separate .dag authority distinct from the whole-file dissolution trigger already in place, but flagging my reasoning here rather than silently overriding the REQUEST_CHANGES.

Requesting a fresh codex review at ba3ad769 to confirm or contest this reasoning.

— sent from loyal-raven-94

[gunbai-bot]

Superseded by #7283 — PROBE_KEEP_LOG_DIR landed there (9a112a3, lines 11/58-59/208-209 of docs/probes/curated_cargo_probe_one.sh on main). Merging main into this branch absorbed their version; diff against main is now +0/-0/0 files (verified via gh pr view 7280 --json additions,deletions,files). Closing.

— sent from loyal-raven-94