Rendered from docs/probes/a-session-says-where-a-check-gets-the-version-that-stood-before-a-change.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 where a check gets the version that stood before a change

Task

You are extending this engine so that lifecycle.transition.not_permitted can judge a status change, which needs the version of the document that stood before the change. In this engine, how does a check obtain that prior version? Answer with one word and nothing else: git if the check runs a version control command itself, history if the engine walks the repository history through a library, or manifest if a caller hands the run a file that names the prior bytes.

Your whole final message is that one word.

Target

Spec 12, Temporal inputs: the clock and the prior version.

Key

answers: [git, history, manifest]
expected: [manifest]

Where plain code carries it. engine/crates/check/src/change.rs:1-10: "It walks no history, it runs no version control tool, and it opens no file that the caller did not name." engine/crates/vcs/src/lib.rs:1-20: "The one crate of this workspace that runs a version control command", and the check-evaluation path "runs no version control command". engine/crates/cli/src/lib.rs carries --change <manifest> on check. The distractor is real and in plain code: headwater change and the vcs crate run version control commands. A session that finds the vcs crate first and stops gives the wrong answer.

Which documents help, and why. Spec 12 states the rule and its boundary in one section, with the reason. HW-DR-0072 rules which verb may run version control plumbing and why. The headwater change contract says that the verb is a producer outside the check loop.

Expected rates, from proposal 2 of #1835. The guess floor is 33 percent. Scenario 0: 45 to 60 percent. The right answer is in the rustdoc of two modules, and the wrong one is in the rustdoc of a third, so the rate is the share that reads change.rs or the vcs header rather than stopping at the first place that runs a command. Scenario 1: 80 to 90 percent.

Leak. The comments state the rule in prose and name spec 12. None of them holds the expected word beside a marker that names this probe. The expected word is also the name of the flag's operand, which is the one place plain code states it as a word, and that is the fair case.