Guard every generic sweep over MAPPINGS against a keyless mapping

[bbfrederick]

What this fixes

Making Mapping.ascconv_key optional for the preview-only 1st Signal/Mode entry needed three guards in mappings.py — display reads the preview instead, resolve refuses with a reason, is_churn is not asked. Those are the three the feature itself traverses. Four other sites read the field generically, and only one of them failed.

The full suite was red at 1 failed, 3742 passed before this.

The four sites, by how they failed

Site	Behaviour	Fix
test_every_enum_choice_agrees_with_the_corpus	Failed. Selected enums as m.choices and m.bit is None, so it asked the ASCCONV block for a key that by construction does not exist — reported the mapping as disagreeing on 418 readings over 261 scans	Excludes keyless mappings, naming the preview-side sweep that covers them
test_no_two_mappings_claim_the_same_ascconv_key	Passed, would invert. Entry identity is (ascconv_key, sequences, when, bit); unique with one keyless mapping, a false collision with two	Identity is the preview_path where there is no key
test_no_protocol_has_two_mappings_writing_one_key	Passed, would invert. Same shape on (ascconv_key, bit) — would report a write race between two mappings that write nothing at all, resolve refusing both	Leaves out a mapping that never writes
probe._is_mapped	Passed vacuously. _is_mapped("") became True; the empty key joined the claimed set and the empty stem joined stems. Safe only because no real ASCCONV key splits to an empty stem	Answers for real keys only

build.stored_display was saved by accident — it returns early on mapping.choices and never reaches the key. A keyless mapping without choices would have gone straight through.

The two that passed are the expensive ones. A check that inverts on the second instance of a thing is worse than one that fails on the first, since the first instance is when someone is looking.

Verification

• Full suite green: 3743 passed, 26 skipped with -n auto (12:15)
• The enum sweep still bites: mutating a real choice (Averaging 1=None → 1=99) fails it, so the exclusion did not gut it. Mutation reverted in try/finally and the file verified clean afterwards
• Duplicate-key fix falsified both ways: clears two distinct preview-only mappings, still catches two sharing a preview_path
• black (91 files), isort, codespell clean

CLAUDE.md

Three entries: the finding itself (a scoping rule enforced in the table and not in the sweeps over the table is enforced nowhere — the sibling of the round-5 widening bug); that piping a suite run through tail discards pytest's exit code, which is why the failing run was first read as green; and a one-clause pin on the probe section's "115 to 249" so a future reader doesn't reconcile it against a MAPPINGS count of 250.

Two figures in the write-up were measured after being drafted from guesswork — 418/261 and 49 grep lines replaced invented "~320" and "30".

🤖 Generated with Claude Code