checkin 2026-10-18 (volt): the malformed-input class is measured on every tool -- 15 probes, 6 real defects reproduced at 461eede, 0 at HEAD

checkins/2026-10-18-volt.md

@@ -140,3 +140,97 @@ because it teaches readers that stated rules are decoration.
 The filename half needs no ruling: it is already clean since 2026-10-08 and the
 25 historical names can stay as the record. I recommend (a) leave the history,
 gate forward with `--since`.
+
+---
+
+# Check-in — Volt, 2026-10-18 (02:3xZ, appended; the block above is 00:16–00:5xZ)
+
+## Did
+
+Closed yesterday's filed carry — *"a tool that mis-codes a decodable but
+malformed input is not measured"* — by measuring it on every tool. Landed
+`tools/selftest_sweep.py` only; **no tool was edited, because no tool was
+wrong.**
+
+**The class.** Input that decodes as UTF-8 and cannot be parsed: a `.json` that
+is not JSON, a `{}` where an audit file should be, a directory that is not a git
+repo, a queue with no records. It is the class where a tool is most tempted to
+answer `0` and exit 0 — exactly the `doc_census_check --tools` bug (`76043b7`).
+
+**15 probes, each graded against the answer its own tool states**, not against a
+blanket "must be rc 2" — because that universal is false here: `check_lint`
+grades a junk log `MISSING` (rc 0 report, rc 1 under `--fail-on-fault`),
+`checkin_check` grades a heading-less check-in advisory (rc 0, its docstring
+says "never rc 1"), `queue_claim_check` reports `0 opening records` (rc 0; its
+`--fail-on-stale` is the gate). Nine tools must refuse at rc 2.
+
+    $ python3 tools/selftest_sweep.py
+    read-refusal: 17 probes, 17 ok, 0 fault
+    malformed-input: 15 probes, 15 ok, 0 fault          # rc 0, 12 tools OK
+
+**15/15 at HEAD is the honest finding, and stated as such: nothing was broken,
+nothing measured it.** Re-run against the revision's own copies
+(`git show HEAD:tools/*.py`) so the working tree is not grading itself: 15/15.
+
+**The gate reproduces the class on real history.** Same axis, tools as they were
+at `461eede` (before the rc-2 unification `6ee8fce`):
+
+    malformed-input: 15 probes, 8 ok, 7 fault (NO_TOOL, RC_WRONG, TRACEBACK)
+    read-refusal:   17 probes, 4 ok, 13 fault (NO_TOOL, RC_NOT_2, TRACEBACK)
+
+Six of the seven are defects in tools that existed then — three raw tracebacks
+(`append_only_audit` not-a-repo; `append_only_owners` malformed-json and
+wrong-shape) and three rc 1 where the contract says 2 (`census_row_delta` both
+shapes; `selftest_sweep` not-a-repo). The seventh is `NO_TOOL`
+(`checkin_check.py` did not exist at that rev) — named, not skipped.
+
+**It discriminates, which is the only thing that makes a gate worth its lines.**
+One bug reintroduced into a copy of `append_only_owners.py` (swallow
+`ValueError`, print `0 owners adjudicated`, `return 0`):
+
+    malformed-input: 15 probes, 14 ok, 1 fault (RC_WRONG)
+    RC_WRONG  rc 0, contract says 2: 0 owners adjudicated: Expecting value ...
+    # and that same tool's own --selftest: 15 cells, 15 passed, 0 failed
+
+The tool's own fixtures never covered the class, so it could regress in silence.
+That sentence is the whole unit.
+
+**The axis found its own bug on the first run** — six probes were substituting
+`{badjson}` as a bare key, leaving a literal path that refused at rc 2 for the
+wrong reason. The required-substring check caught it; without it those six rows
+would have printed OK. Fixed and re-run before any of these numbers were written.
+
+**Selftest 22 → 28 cells**, including c24 (the silent-zero shape: rc 0 where the
+contract says 2 → `RC_WRONG`), c26 (right code, wrong reason → `UNEXPLAINED`) and
+c28 (coverage: every tool with a missing-probe also has a malformed probe, so a
+new tool is visible rather than skipped):
+
+    $ python3 tools/selftest_sweep.py --selftest
+    selftest: 28 cells, 28 passed, 0 failed
+
+**Queue untouched.** `queue_index.py --check docs/queue-index.md` → rc 0
+(`49 sections, 45 ids, 4 continuation sections, 9 nested ids`) before this block
+and after it. No queue item opened or closed; open set unchanged (Q37 `:5278`
+pipe; career-lane Q53 held for Alexander).
+
+## Blocked
+
+Nothing.
+
+## Next
+
+- If a tool's malformed contract changes, its probe changes with it: the table is
+  in `tools/selftest_sweep.py:MALFORMED_PROBES`, cited by line in the review, not
+  inferred.
+- The axis grades the code, a message, and one required word — not whether the
+  message is good. That is the next honest increment if it is ever needed; it is
+  not claimed now.
+
+## Ask
+
+The four-heading ruling from the block above still stands (unchanged, still the
+only thing I need from Alexander). Nothing new.
+
+Artifact: `/root/.hermes/wizard-outputs/volt/2026-10-18-malformed-axis/`
+(`decision.md`, `selftest.txt`, `sweep-head.txt`, `head-copies.txt`,
+`ab-461eede.txt`, `discrimination.txt`).