Repository navigation
Conversation
Signed-off-by: dada-yan <BinjunYann@gmail.com>
WaylandYang
left a comment
There was a problem hiding this comment.
The path you draw — a reason-classified unbound signature becomes an evidence-backed property proposal that a person adopts on the workbench — is the shape 0044 reserves for cut 5, and the seed cases with expected directions are the part I would have asked for. Naming #802 as the prerequisite rather than folding it in was the right call, and that one is merged.
The argument I made on #804 applies here: docs/design/proposals/ is a third place for design in a repository that already has two, and a draft that is not accepted and linked from nothing is a document that rots. This one belongs on #725, or as a decision record with a number and cuts if it is ready to be decided.
One thing that is specific to this draft rather than to its location. ontology-property-cases.jsonl is a bench fixture without a bench: nothing reads it, nothing scores it, and its expected directions cannot fail. If the evaluation is real, it wants to live beside the thing that runs it — scripts/bench/ has the corpora and the scoring for exactly this, and a case file there is checked by running it. As a file under docs it is a table of intentions.
|
Closing as requested in review. The findings and unresolved design questions are now on the existing issue: #725 (comment). No new draft-document directory or unexecuted fixture is being proposed for merge. Thank you for clarifying the repository’s design and test conventions. |
This draft defines a narrow path from reason-classified unbound signatures to evidence-backed relationship proposals and human adoption, reusing the existing workbench. It includes three-domain seed evaluation cases with explicit expected directions and reuse outcomes.
Design discussion only: no agent, migration or adoption endpoint is implemented. Decisions remain on proposal identity, evidence/version checks, rejection lifecycle and atomic adoption. The separate statement-count prerequisite is #802; this draft does not claim that fix implements the proposal flow. Related to #725 and ADR 0044.
Validation: documentation/link checks and JSONL parsing. Real-model quality and the new adoption lifecycle are NOT_RUN.