Next Feature Decision¶
Status: waiting for launch feedback. Do not start a large feature merely to fill the gap while the poll is running.
Decision window¶
Open the launch poll using
_drafts/launch-v0.5.4/github-poll.md,
then review it after 14 days or 10 votes, whichever comes first. Record the
opening date, closing trigger, links, and raw counts below.
| Signal | ML preset | Provenance manifest | PyInstaller guide | MCP integrity |
|---|---|---|---|---|
| Poll votes | — | — | — | — |
| Substantive comments | — | — | — | — |
| Launch-channel requests | — | — | — | — |
| Existing issue reactions | — | — | — | — |
Selection rule¶
- Prefer described workflows and reproducible pain over bare votes.
- Require at least two independent users describing the same need before a large implementation; otherwise choose the smallest reversible experiment.
- Break ties by expected user value, then implementation cost, then fit with pyobfus's AST + AI-debuggable scope.
- Publish the result and rationale before implementation starts.
Candidate experiments¶
- ML/model-serving preset: first add a failing fixture for one real serving framework and validate preserved reflection points.
- Signed provenance manifest: first define a versioned JSON schema and an unsigned deterministic manifest; add signing only after the contract settles.
- PyInstaller guide: first prove one clean obfuscate-then-bundle example in CI before expanding into a cookbook.
- MCP integrity: first threat-model which metadata can be trusted locally and demonstrate a changed-tool-surface detector.
Decision record¶
- Poll opened: pending external publication
- Review trigger reached: pending
- Selected candidate: pending evidence
- First bounded experiment: pending
- Rationale/link: pending