[guijacquemet]

What

Stage 1 design PR 7 (conformance P30 and the proposal part of P32). Additive: no run, notebook or CLI command uses it yet, and nothing is written to disk.

• draft_ids: a folder label is proposed as a condition, technical-repeat or repeat ID only when the reader would accept it as written (Spec 4.6a, extended to repeat IDs as in Q6). control, files, z and the like get no proposal. Nothing is trimmed, re-cased, numbered, derived or renamed. The rules applied are the reader's own (description_identity, description_rules), with no second copy. A raw header name is proposed as a channel ID only if it is a valid ID as it stands (Q4). R, G and B are never proposed, and names equal ignoring case within one layout get no proposal.
• draft: draft_from_discovery and Draft. IDs stay proposed until confirm_id. Open questions are listed, and a draft is not publishable while any is open. unit_rows() gives condition_id, repeat_group and technical_repeat_id beside the unchanged source labels (Q8). The draft is not a description and writes no file.
• migrate_legacy_design: reads experiment_design.csv strictly, so yes, no and empty are refused. repeat_group is carried as the proposed repeat ID. A group with several confirmed_on dates takes the latest, and a note lists each folder's date (never the clock). A confirmed group with no date stays open. 0, 1 or 2+ marked controls give no reference, one reference per other condition, or an open question. A design that breaks a Spec rule is refused with the reader's code (found by running the reader's own phases on a probe document, for example E_REPEAT_ID_SHADOWS_SOURCE). A legacy condition such as control migrates with no proposed ID and is not refused. New folders are pending and reopen their group. Missing legacy folders block publication.
• Owner decision (2026-11-07), Spec 4.6a: a legacy repeat_group that is not an allowed repeat ID (for example z, files, overrides) is not refused. It migrates with no proposed repeat ID and an open id_pending question, so the user names the repeat. Nothing is repaired or derived, the legacy confirmation and date are kept, and confirm_id still refuses the forbidden value unchanged (E_FORBIDDEN_FIELD). In the rule-check probe the repeat now gets a stand-in ID like conditions and technical repeats do. This replaces the earlier "refused with E_FORBIDDEN_FIELD" behaviour (commit 2 of this branch).
• design._parse_strict is renamed parse_design_bytes (public, same behaviour).

Not decided (skipped)

• For a forbidden condition or technical-repeat folder (control, files) the Spec says preprocessing proposes "a different ID that is allowed", but gives no rule for making one. Deriving one would be a repair, so there is no proposal. The user names the ID. P32's "never control, files, overrides" holds. The derived CSV part of P32 is PR 10.
• Wiring raw header names into the draft needs the raw-name field of the discovery PR (Discovery and header additions for the experiment description (Stage 1, PR 6) #190, now merged). propose_channel_ids takes the names as input and is tested alone.

Part of the Stage 1 design (PR 7). Main was merged into the branch (CHANGELOG and AGENTS.md conflicts resolved by keeping both entries).

Tests

tests/test_draft.py, 35 tests. On origin/main the file fails at collection (ModuleNotFoundError: morphobricks_preprocessing.draft). Regression for the owner decision: test_a_legacy_repeat_group_that_is_not_an_allowed_id_opens_a_question[z|files|overrides]. On the previous branch tip (source reverted) it fails for all three with LegacyMigrationRefused ... (E_FORBIDDEN_FIELD); on this branch: 35 passed.

python tools/check.py after merging current main: 3879 passed, 4 skipped, code rules clean. The 4 skips are environment ones (case-insensitive file system and file-mode tests, unrelated to this change).

---

Generated by Claude Code