Skip to content

Human action 27

Jerry Holland edited this page Sep 29, 2026 · 4 revisions

Evaluate causal-rebase usability before expanding ranges

Canonical work item: #27.

When required: Before expanding linear rebase; work remains parked. The issue carries current status, decisions, and completion evidence.

Owner role: Maintainer and representative workflow users.

Status 2026-09-17: Done. The owner ran the kit with a pause after each PREDICT prompt against main at bd92f0b; the tool matched every prediction, so the linear v1 model is judged understandable. ADR-0032 was accepted the same day and #27 is now ordinary implementation work (decision).

Prerequisites: This feature remains parked. The examples exist; do not implement range expansion merely to collect feedback.

  1. Step 1 is done (2026-09-15): the evaluation kit, a bash script that builds a disposable repository and walks the three scenarios (clean replay, covered change, unexpected-empty blocker with both recoveries) with a PREDICT: prompt before each command, is attached to #27, and a recorded transcript shows what the tool did at 468cd06. Save the script as rebase-walkthrough.sh and run bash rebase-walkthrough.sh <checkout>/bin/vlab.js on any machine with Node and Git; it writes only under the temp directory and removes its repository.
  2. Ask participating users (maintainer and workflow representatives) to answer each PREDICT: prompt before reading the output that follows it: the predicted changes, origin/result identities, and recovery action. Record each prediction, whether the observed output matched, and any place the model was surprising.
  3. Review the feedback against the proposed contract, ADR-0032 (Proposed), and record whether the range generalization is useful and explainable, needs a revised design, or should stay deferred; then answer the ADR's remaining numbered decisions.

Verification: Concrete user feedback and the resulting scope decision must be recorded before broadening linear v1.

Failure handling and cleanup: Keep work parked if clarity is unresolved. Use examples or disposable fixtures rather than experimenting in valuable history.

Completion: attach the evidence/decision to issue #27, resolve its human-action label when satisfied, update the private board, and point the Human TODO entry to the completed outcome.

Publish only evidence suitable for the public issue/wiki. Redact private paths, repository content, and credentials from shared output. Sensitive vulnerability findings belong in a draft security advisory.

Back to Human TODO · Benchmark host baselines

Clone this wiki locally