Rendered from docs/probes/a-session-says-whether-a-command-or-a-hand-writes-a-governs-line.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 whether a command or a hand writes a governs line

Task

An agent in this repository writes a new document. The document must declare a governs edge onto a source file, in its front matter. Can a command of the headwater tool write that line into the new document? Or must the line be typed by hand?

Answer on one line with one word, command or hand, and with nothing else. Your whole final message is that word.

Expectation

The terminal answer is command, the one expected value of the closed set below.

answers: [command, hand]
expected: [command]

This probe replaces the probe that the campaign of 2026-10-03 ran, and only the tree is different (#1709). The task and the key are word for word the same. The key held. The recount of that probe missed one current near copy, and #1709 repaired it, so the tree that a session searches changed. A result of this probe does not compare with the results of that campaign, which ran the old identifier.

This probe has the distractors-at-scale shape (#1472). The record in force is HW-DR-0104, at status: current. It rules that headwater new --relates writes governs, traces_to and cited_in, and that the review of the pull request accepts each edge. On 2026-10-10, headwater new probe --relates traces_to=HW-SPEC-ai-integration wrote this document's own traces_to edge, which declares created_by: agent as governs does.

Near copies state the other answer. HW-DR-0083 is superseded by HW-DR-0104. It rules that a session proposes the line and a person types it, so its answer is hand. Three other documents still state its answer. None of them is a current record: one is a discharged obligation, and two are generated pages that print the summary of HW-DR-0083. A session that answers from any of them answers hand, and the grade finds that answer wrong.

The count, and how it was measured. A near copy here is a document that holds the two key terms of the task, governs and created_by. The tree is the tree of #1709, extracted with git archive and sealed with tools/probe/seal.sh for the five probes of the HARDER table in engine/crates/probe/tests/corpus.rs, this probe in place of its predecessor. In that tree, grep -r -l 'created_by' docs | wc -l prints the documents that hold the first term, and xargs grep -l governs over that list prints the near copies. Then grep -n -i -E "no verb (of this engine )?writes|person types|is hand entry|types the line" over the near copies prints each candidate sentence. A person reads each candidate and keeps the documents that state the answer hand about governs.

Why the search has one more term. The search of the predecessor did not have types the line. So it did not find the governs-edges evaluation, a current near copy that said "A person or an agent that has read the file types the line." On the tree of origin/main at f4a7c908, the wider search prints that sentence. So the true count on the tree of the campaign was 5 of 39, not 4. #1709 repaired the sentence, and the evaluation now cites HW-DR-0104.

The recount of 2026-10-10. The first search printed 52 documents, and 39 of them are near copies. The wider search prints no sentence of the governs-edges evaluation. These 4 of the 39 state the answer hand about governs:

  • HW-DR-0083 itself, which is superseded.
  • HW-OBL-0105, at discharged: "No verb of this engine writes either edge".
  • The generated index docs/decisions/README.md and the tombstone docs/spec/09-open-questions.md, which each print the summary of HW-DR-0083 beside the mark superseded.

The other candidate sentences are about another relation or another artifact, or they state the expected answer. docs/taxonomies/brd-prd/doctrine.md says that a person types the line of a relation between a business need and a product document, and it cites HW-DR-0083. It does not name governs, so the count leaves it out. Spec 13 says that a person typed each line at the time of HW-DR-0083, and then that HW-DR-0104 supersedes it. HW-DR-0104 and the contract of headwater new state the expected answer.

The distractors are not built for this probe. Each one is a document of this corpus that went stale when HW-DR-0104 landed. A campaign over this probe therefore measures the corpus as it is. If a later change repairs more of these documents, the count above falls, and this probe must count again before it runs.