[guijacquemet]

What

Removes the one-way migration of a legacy experiment_design.csv into a draft (inventory row PP-03), following MorphoBricks-Spec PR 25 (merged 542c936; ADR 0002 AR10 (9), experiment-v1 §7/§11/§12, conformance P30). Preprocessing never reads an output-folder experiment_design.csv to propose or prefill a draft or description. The derived file stays an output.

Changes

• Re-vendor the Spec at cc9820f5b84f01bbbb5c8e3f54e70ee13cb5e9eb with the Spec's tools/vendor.py; --check: "identical to the Spec at its pinned commit". PIN.toml updated by the tool; draft_format.SPEC_COMMIT follows the pin (an answers draft saved under 1bed024 is refused as before, spec_commit_mismatch). The tool also rewrote tests/spec_fixtures/README.md in its standard form.
• Removed (draft.py, about 165 lines): migrate_legacy_design, LegacyMigrationRefused, MigrationNote, _check_legacy_rules (and _probe, _legacy_groups, PLACEHOLDER_DATE, used only by it), Draft.origin, Draft.legacy_missing, Draft.notes, and Draft.unresolved_controls with the control_unresolved question (only the migration produced it). The legacy_folder_missing question and its texts in cli_describe.py and notebook_check_view.py; the describe "cannot be drafted from experiment_design.csv" branch; docstring mentions in draft_answers.py and run/draft_file.py. Notebook files: removals of two text-table lines only.
• Kept: design.parse_design_bytes and design_issues. design_issues is the run's reconcile check. parse_design_bytes has no src/ caller left but is exported and is the strict reader of the written file in test_design, test_design_file, test_spec_derived_design and test_dataset; docstrings now say so. Say if you want it removed with those tests.
• Tests: P30 (test_spec_derived_design) follows the amended Spec: v13 derived twice with a frozen clock moved by a day, both equal the v13 golden, every row carries its repeat's confirmed_on, clock reads 0. Deleted the migration tests (11 in test_draft.py, 3 in test_cli_describe.py, 1 in test_draft_answers.py). Added tests/test_orphan_design_csv.py and a test_cli_describe case: an orphan experiment_design.csv in the output folder (confirmed repeats, a control, odd groups, or a malformed file) is ignored by draft creation, describe (with and without --publish; output unchanged, file untouched) and the notebook (open and reopen).
• Docs: user guide, contract, AGENTS.md map, conformance P30 row (marked amended; the dated run record is not relabelled), CHANGELOG Unreleased "Breaking".

Verification

• python tools/check.py: ruff, format, code rules and pytest: 5150 passed, 4 skipped, 0 failed. The 4 skips are environmental (case-insensitive file system x3, file-mode permissions x1), none in test_spec_*.
• Evidence, main (a87f524) against this branch on the same vendored Spec fixtures: 37 valid fixtures run through run_experiment: 31 written, 6 refused identically on both (v04, v05, v06, v17, v28, v29: the test tree helper does not build their layouts). plane_histograms.csv, experiment_design.csv and the manifest minus time keys are byte-identical (SHA-256) for all. The refusal codes of the 135 invalid fixtures are unchanged. Result files: JSON of both runs compared equal.

Not changed

Another agent's claude/redesign-* notebook branches: no notebook UI file is touched beyond the two removed text-table lines.

---

Generated by Claude Code