Rendered from docs/probes/a-session-says-what-a-verb-that-mints-an-identifier-writes-beside-the-document.md in the Headwater
corpus. Every document on this half of the site is typed by the taxonomy
the descriptor names: corpus.json.
A session says what a verb that mints an identifier writes beside the document
Task
You are adding a verb to this engine that writes a new obligation record and mints its identifier, the next free number of the scheme. Besides the document itself, what must the verb write so that two branches that each mint the next number cannot both merge in silence? Answer with one word and nothing else: claim if it writes one file for the identifier under a store beside the corpus, lock if it records the number in .headwater/taxonomy.lock, or nothing if the document alone is enough because a rule reports the duplicate.
Your whole final message is that one word.
Target
Key
answers: [claim, lock, nothing]
expected: [claim]
Where plain code carries it. engine/crates/check/src/claim.rs:1-40 is the whole argument: one file per identifier under the store, written once, so that two branches hit an add/add conflict, and the measured fact that a zero-byte store merges silently. engine/crates/scaffold/src/lib.rs:341 is the mint that owes a claim, and identifier.claim.missing is in RULES at engine/crates/check/src/lib.rs:263. The distractor is the third option, which is what a session concludes from identifier.claimed_twice in engine/crates/check/src/duplicate.rs if it stops there, and the header of claim.rs says in its second paragraph why that is not enough. The store itself is absent in scenarios 0 and 1, so a session cannot see it, only the code that writes and reads it.
Which documents help, and why. HW-DR-0054 is the ruling. Spec 3 says under Identifiers that a tree holds no concurrency and that the claim store is the answer. The headwater new contract states what the verb writes. CLAUDE.md states it too, and it is absent from both scenarios, so it does not reach this comparison.
Expected rates, from proposal 2 of #1835. The guess floor is 33 percent. Scenario 0: 40 to 55 percent. The rustdoc of claim.rs is complete, and the rate is the share that opens it rather than stopping at the duplicate rule. Scenario 1: 75 to 85 percent.
Leak. The rustdoc states the mechanism in prose with the expected word throughout, because that is the module's name. That is plain code stating it. The documents state it where a search for an identifier or a mint lands, and that is the difference measured.