# Development baseline revision

The original rules source is preserved byte for byte as rules-v1.ts (SHA-256 a65066aa1f6614517dc294233fc9af18d8439a9fb5a867d467fa538152985e5c). This is an evidence copy, not an executable adapter: its original relative import and exported version came from the v1 run source snapshot. The retained dev-v1 manifest owns the original complete source hashes and baseline version inbox-rules-1.0.

One deliberate revision used only the 40 development messages and author references, after the 40/40 Jev development calls completed. The v1 baseline matched all three fields on 15/40 dev cases. The revised rules-v1.1 matches 40/40 dev cases; that is development fit, not generalization evidence. No held-out or challenge messages/labels were inspected or used to tune the baseline. No Jev policy/question prompt was revised.

The revision removes workspace-only technical ownership, distinguishes presentation issues and bare vague requests, expands present blocking/workaround evidence, handles explicit negation/resolution, recognizes payment/access prerequisites and ordinary ownership cues, strips explicit routing commands, and abstains on unrelated mixed owners. It uses regular expressions only, with no id/annotation/reference lookup. Current rules.ts SHA-256 5adb4c2e8f6f553dc489aee0f7408d0595c8c49a9aeae9f041fa07ef7018e5d3; exported version inbox-rules-1.1.

Dev-v1 artifacts remain unchanged in the ignored research/jev-inbox/runs/dev-v1 directory and parent source snapshot. Their confidence report used legacy bin edge 0.75; the final scorer uses precommitted edges 0,0.5,0.8,0.9,1. Final dev baseline replay and held-out/challenge results must name the final frozen source/manifest hashes.

## Protocol chronology

`protocol-proposed-v1.json` is the author-handoff proposal written after development (SHA-256 `824c3078e6793f5470077ce7a6b8866ece94d9fa5ae3d371e03b00b7314313d4`; the complete digest is in `dataset-review.md`). It is not the exact protocol bytes identified by the earlier smoke/dev manifests. Those manifests retain their different protocol hashes; their historical protocol snapshots were not retained in the source snapshot. The unmet pre-development review gate remains a deviation. All final held-out/challenge/repeat/reversed manifests match the retained final frozen protocol. No historical protocol is reconstructed or relabeled as an earlier executed input.