Plan de Migración (Coveris-e → main)

MIGRATION-PLAN.md — Coveris-e → main

Task list for the main worktree terminal. Migrate file-by-file from Coveris-e into main in well-curated, atomic phases. Do NOT git merge. Use git checkout Coveris-e -- <paths> per phase, test, commit cleanly, move on.

Source of truth & references

Source branch: Coveris-e (HEAD f371f20, 13 commits ahead of origin/Coveris-e, never pushed)
Target branch: main
Background reading (in Coveris-e worktree):
- docs/INTEGRATION-HANDOFF.md (situation summary + 3 blockers)
- docs/volatile-docs/00-08-*.md (full analysis archive — NOT to be migrated)
- docs/plans/anesthesia-rotation-mock-data.md (was the plan for C0 below — NOT to be migrated)
- docs/volatile-docs/04-branch-comparison.md line-by-line file diff
- docs/volatile-docs/05-merge-readiness.md blockers detail

Hard rules

✋ Never git merge Coveris-e or git cherry-pick. Only git checkout Coveris-e -- <path>.
✋ Never bring docs/volatile-docs/, docs/plans/anesthesia-rotation-mock-data.md, frontend/design_handoff_scheduling_redesign/.
✋ Never commit .env (it is gitignored — verify).
✋ Never hardcode SUPERUSER_PASSWORD in any doc; use ${SUPERUSER_PASSWORD} or “(see .env)”.
✅ One phase = one PR-grade commit on main. Conventional message + Co-Authored-By trailer.
✅ Each phase: tests must pass before committing. Migrations must apply on an empty DB.
✅ Update CHANGELOG.md once per phase.
✅ If a phase needs new ADRs to make sense, do A0 first.

Tier A — Spec & contract (no runtime code)

[ ] A0 — ADRs 025–034 + README index

git checkout Coveris-e -- docs/adr/adr-025-*.md docs/adr/adr-026-*.md docs/adr/adr-027-*.md docs/adr/adr-028-*.md docs/adr/adr-029-*.md docs/adr/adr-030-*.md docs/adr/adr-031-*.md docs/adr/adr-032-*.md docs/adr/adr-034-*.md
Verify file list with ls docs/adr/
git checkout Coveris-e -- docs/adr/README.md (index)
Verify ADR-029 status = Accepted (per ADR-029 promotion in Coveris-e); if not, edit by hand
Commit: docs(adr): add ADRs 025-034 (weekly structure, scheduling, slot anchoring)

[ ] A1 — PRD.md + API.md + api-conventions (resolve blockers B1, B2)

B1: Edit PRD.md §3 line ~44 to mention Fase G (scheduling, exceptions, PDF, calendar); §8.4 mark “¿Cuándo entra programación de turnos?” as resolved by ADR-029.
B2: git checkout Coveris-e -- API.md (brings full updated API.md including PATCH /api/v1/assignments/{id}/slot_index/).
Verify API.md footer “Last Updated” matches today’s date; verify Scope line includes Scheduling + Holidays + AssignmentExceptions + ShiftInstances.
git checkout Coveris-e -- docs/api-conventions.md (if changed).
git checkout Coveris-e -- docs/07-drf-endpoints.md docs/08-primeng.md docs/09-tailwind4.md (only if these changed since main).
Commit: docs: PRD §3+§8.4, API.md scheduling endpoints, conventions (resolve B1+B2)

Tier B — Backend, in dependency order

[ ] B0 — `demand/` foundation: OrgUnit depth + manager_position

git checkout Coveris-e -- backend/apps/demand/enums.py backend/apps/demand/models.py
git checkout Coveris-e -- backend/apps/demand/migrations/0003_add_org_level.py through 0010_alter_orgunit_manager_position_and_more.py
git checkout Coveris-e -- backend/apps/demand/management/commands/seed_org_units.py
git checkout Coveris-e -- backend/apps/demand/tests/test_orgunit_depth.py backend/apps/demand/tests/test_orgunit_manager.py (whatever exists)
Inspect serializers/views for breaking changes; bring backend/apps/demand/serializers.py and backend/apps/demand/views.py only if needed
cd backend && pytest apps/demand/tests/ -v → all green
docker compose down -v && docker compose up -d → migrations apply from empty DB
Commit: feat(demand): fixed-depth OrgUnit taxonomy + manager_position (ADR-027)

[ ] B1 — `tags/` + `business_rules/`

git checkout Coveris-e -- backend/apps/tags/
git checkout Coveris-e -- backend/apps/business_rules/
git checkout Coveris-e -- backend/apps/tags/management/commands/seed_tags.py
Run pytest apps/tags/tests/ apps/business_rules/tests/
Commit: feat(tags,business_rules): regime tags + hours_source rules (ADR-028)

[ ] B2 — `api/` (Employee) + `fsm/` + `users/` (minor changes)

Diff first: git diff Coveris-e..main -- backend/apps/api/ backend/apps/fsm/ backend/apps/users/ — bring only what changed
Bring updated seed_anesthesia_nomina.py, seed_therapy_nomina.py, seed_mock_users.py
Run pytest apps/api/tests/ apps/fsm/tests/ apps/users/tests/
Commit: feat(api,fsm,users): Employee FSM updates + seed commands

[ ] B3 — `assignments/` core (without slot_index — that’s B7)

git checkout Coveris-e -- backend/apps/assignments/enums.py backend/apps/assignments/models.py backend/apps/assignments/services.py backend/apps/assignments/serializers.py backend/apps/assignments/views.py
Migrations 0003-0006: git checkout Coveris-e -- backend/apps/assignments/migrations/0003_*.py 0004_*.py 0005_*.py 0006_*.py
Tests: git checkout Coveris-e -- backend/apps/assignments/tests/
Skip migration 0007_assignment_slot_index.py for now (B7)
Inspect models.py and manually remove the slot_index field for this phase (it lands in B7). Same with any service code that uses it.
Run pytest apps/assignments/tests/
Commit: feat(assignments): hours_source + effective_weekly_hours + recurrence_rule (ADR-028, ADR-029 partial)

[ ] B4 — `weekly_structure/` full module

git checkout Coveris-e -- backend/apps/weekly_structure/ (everything)
Migrations 0001-0012 all of them
Run pytest apps/weekly_structure/tests/
Verify reconciler: create board → placements → reconcile → check Position.source_subrow_key shape
Commit: feat(weekly_structure): PositionTemplate, WeeklyBoard, Placement + reconciler (ADR-025, ADR-026, ADR-031, ADR-032)

[ ] B5 — `demand/` Position scheduling fields (ADR-030)

git checkout Coveris-e -- backend/apps/demand/migrations/0013_position_time_range.py 0014_position_rotates_weekly.py 0015_position_pinned_slot_indexes.py 0016_remove_position_pos_time_range_coupled_and_more.py
Re-apply backend/apps/demand/models.py (Position now has start_time, end_time, pinned_slot_indexes) — already brought in B0, but verify the post-merge state matches Coveris-e
Update backend/apps/demand/serializers.py and views.py for new fields
Tests for time-range validation, overnight wrap, ADR-032 pinned indexes
Run pytest apps/demand/tests/
Commit: feat(demand): Position start_time/end_time + pinned_slot_indexes (ADR-030, ADR-032)

[ ] B6 — `calendar_api/` (Holidays AR cache)

git checkout Coveris-e -- backend/apps/calendar_api/
Migration 0001
Run pytest apps/calendar_api/tests/
Smoke: curl http://localhost:8000/api/v1/holidays/?country=AR&year=2026
Commit: feat(calendar_api): holiday cache for AR (ADR-029 Q12)

[ ] B7 — `scheduling/` Fase G + `assignments.slot_index`

git checkout Coveris-e -- backend/apps/scheduling/ (full app)
git checkout Coveris-e -- backend/apps/assignments/migrations/0007_assignment_slot_index.py
Restore Assignment.slot_index field in backend/apps/assignments/models.py (was held back in B3)
Restore PATCH endpoint for slot_index in assignments/views.py
Run pytest apps/scheduling/tests/ apps/assignments/tests/
Smoke: curl /api/v1/scheduling/<org_unit_id>/2026/5/ returns nested structure
Commit: feat(scheduling): monthly projection + ShiftInstance + AssignmentException + PATCH slot_index (ADR-029, ADR-034)

Tier C — Data, runtime, bug fixes

[ ] C0 — Mock data infra + anesthesia rotation

git checkout Coveris-e -- mock-data/scripts/ (generators + validator)
git checkout Coveris-e -- mock-data/notti-anesthesia/ (7 YAMLs)
git checkout Coveris-e -- backend/apps/scheduling/management/commands/seed_anesthesia_rotation.py backend/apps/scheduling/management/__init__.py backend/apps/scheduling/management/commands/__init__.py
Decide: does main keep its existing mock-data/clinic-bienestar/? If yes, leave it. If no, drop in favor of notti-anesthesia.
git checkout Coveris-e -- docker-compose.yml — pulls the ./mock-data:/mock-data:ro mount line
git checkout Coveris-e -- backend/entrypoint.sh (if changed)
Smoke: docker compose down -v && docker compose up -d && docker compose exec api python manage.py seed_anesthesia_rotation
Commit: feat(mock): notti-anesthesia rotation scenario + YAML SSOT scripts

[ ] C1 — coverage-summary empty-state fix

Already merged into B0 path via backend/apps/demand/views.py. If not picked up there, cherry-pick manually:

total_required = sum(d["required_hours"] for d in unit_data.values())
total_assigned = sum(d["assigned_hours"] for d in unit_data.values())
total_required = sum((d["required_hours"] for d in unit_data.values()), Decimal("0"))
total_assigned = sum((d["assigned_hours"] for d in unit_data.values()), Decimal("0"))

Verify with empty DB: curl /api/v1/demand/coverage-summary/ returns 200 with "0.00"
Commit (only if standalone): fix(demand): coverage-summary 500 on empty positions

Tier D — Frontend (out of scope for this sprint)

Backend-first per user’s directive. Frontend migration is a separate plan to be drafted once B0–B7 + C0 land cleanly on main. Anticipated sub-phases:

D0 org-chart improvements (manager dialog, worker drawer)
D1 roster dialogs (assignment, contract, tag-assignment)
D2 structure component + tabs + filters
D3 weekly-structure (19 files + WeeklyStructureStore 818 lines)
D4 scheduling-month-grid (1142 lines)
D5 staffing-plan
D6 coverage page polish
D7 help system (5 components)
D8 theme/design tokens diff (only if changed)
D9 NG8113 cleanup + minor lint warnings

Do not start D0 until all of A0..C0 are green and CHANGELOG reflects them.

Verification per phase (mandatory)

# 1. tests
cd backend && pytest apps/<app>/tests/ -v

# 2. clean migrations from empty DB
docker compose down -v
docker compose up -d
# wait for health: curl http://localhost:8000/api/health/

# 3. lint (optional but recommended)
cd backend && ruff check apps/<app>/

# 4. CHANGELOG
echo "- <date> feat(<scope>): <one-line summary>" >> CHANGELOG.md

# 5. commit
git add -A
git commit -m "<conventional message>" \
  -m "Co-Authored-By: ..."

Push policy (recordatorio)

Push de main SOLO con consentimiento explícito del usuario.
Push de Coveris-e ya fue autorizado en la sesión pasada (último push de archive).
Toda otra rama → local-first.

Estado al cierre de la sesión que armó este plan

Coveris-e HEAD: f371f20 (luinloder integrado, mock data anestesia, handoff doc)
main HEAD: fdf0a90 (intacto)
Worktrees activos:
- /Dev/Coveris/cotton-coveris-mvp → Coveris-e
- /Dev/Coveris/cotton-coveris-mvp-main → main (este archivo vive acá)

INTEGRATION FINDINGS — ADR review + models/API diff (2026-05-31)

Read-only assessment by the coveris-adr-review Agent Team (3 Opus business-logic experts) + direct diff. Source: Coveris-e @ 6bfd1bd · Target: coveris-b-integration @ 8c6f479 · merge-base ed3b4c8. ⚠️ Per AGENTS.md, ADR integration is HUMAN-IN-THE-LOOP. Nothing below is adopted until the user picks. The team assesses; the user decides.

1. ADR divergence (bidirectional)

Coveris-e adds 9 net-new ADRs: 025, 026, 027, 028, 029, 030, 031, 032, 034 (there is no 033).
Drift on shared ADRs is mostly cosmetic / a downgrade — keep OURS:
- Coveris-e lacks our adr-000-adr-conventions.md and adr.base.
- Coveris-e still carries the monolithic adr-001.md (we split it into 001-a/b/c).
- Its adr-019 differs by ~75 lines but the change is cosmetic only (stripped YAML frontmatter + wikilinks, heading style) — a downgrade for our vault tooling, zero semantic change. ADR-028 does not depend on it. → Keep our adr-019.

2. Supersession map (the load-bearing part)

New ADR	Effect on our spec
ADR-027 (Accepted)	Supersedes ADR-026 (already marked Superseded) and supersedes our ADR-018 (UNIT-only positions). Reshapes the nucleus: drops `OrgUnitType` enum (CLINIC/DEPARTMENT/SERVICE/UNIT) → `depth` 0-4 (CLINICA/AREA/DEPTO/SERVICIO/UNIDAD, adds AREA = 5 levels); Positions allowed at any depth; manager slot via `OrgUnit.manager_position` OneToOne. Rewrites ADR-001-a OrgUnit rows + the whole org-units/coverage API. Spanish enum names (`CLINICA`/`AREA`/`DEPTO`) are a possible ADR-001-b (English-in-code) snag.
ADR-025 (Accepted)	Its core “Semanal isolated, no Position materialization, four models incl. `WeeklyTier`” is already reversed by 026/027 — auto-sync reconciler materializes Positions per-day; `WeeklyTier` is dropped (confirmed: `weekly_structure/models.py` has no WeeklyTier). 025 stands only as a historical record.
ADR-026 (Superseded by 027)	Bring in as a Superseded record for completeness only — no active force.
ADR-028 (Accepted)	Silently contradicts our ADR-024 (`effective_hours == required_weekly_hours` + binary `is_reinforcement`) and extends ADR-001-c (adds violet PRESTACION band, redefines “blue”) without amending either. Coupled edits to 024 + 001-c are mandatory before adoption.
ADR-029 (Accepted)	Monthly scheduling LAYERS ON ADR-011 weekly-only (011 pre-authorizes “monthly = read-time aggregation”). No contradiction; ADR-011 stays Accepted/untouched. Caveat: keep `ShiftInstance.planned_hours` / `AssignmentException.hours_delta` projection-only — never feeding the weekly ledger.

3. Per-ADR verdict (team recommendation — user picks)

ADR	Title	Src status	Verdict	Coupled work
025	Weekly design surface	Accepted	ACCEPT (historical)	note isolation/WeeklyTier superseded by 027
026	Configurable org levels	Superseded	ACCEPT as Superseded only	none (no active force)
027	Fixed-depth taxonomy + manager slot	Accepted	NEEDS-DISCUSSION	supersede ADR-018; rewrite ADR-001-a + org-units API; resolve Spanish enum vs ADR-001-b
028	Staffing `hours_source`	Accepted	NEEDS-DISCUSSION	reconcile/ supersede ADR-024; amend ADR-001-c (violet band)
029	Monthly scheduling	Accepted	ACCEPT-WITH-MODS	state publish/staleness invariant in-ADR; ADR-001-a glossary rows; fold Q1–Q15
030	Position time-bands	Accepted	ACCEPT	(optional) rename file to match scope; band must divide `required_weekly_hours`
031	Closed-on-holidays	Accepted	ACCEPT	lowest risk; default-safe
032	Weekly rotation	Accepted	ACCEPT-WITH-MODS	pin ISO-week TZ per ADR-015; decide year-boundary anchor
034	Assignment slot anchoring	Accepted	ACCEPT (gated)	land AFTER 032; consider INFO on silent slot-collision

4. Backend `models.py` comparison (coveris-b-integration → Coveris-e)

Net-new apps: scheduling, weekly_structure. Unchanged: api (Employee/BaseModel byte-identical), fsm, business_rules, offer, users.

File	Δ	What changed
`demand/models.py`	+336	OrgUnit: −`unit_type`, +`depth`, +`max_weekly_extra_hours`, +`manager_position`(OneToOne). Position: +`plan_line` FK, +`source_subrow_key`, +`day`, +`start_time`, +`end_time`, +`pinned_slot_indexes`, time-range `clean()`/`derived_hours`/constraints. NEW models `StaffingPlanLine`, `StaffingPlanLineTag`. (ADR-026/027/030/032 + staffing-plan)
`assignments/models.py`	+73	+`effective_weekly_hours`, +`hours_source`, +`recurrence_rule`(JSON), +`slot_index` (ADR-028/034)
`tags/models.py`	+52	TagCatalog +`allows_productividad`, +`sourced_from_template`; NEW model `PositionTemplateTag` (ADR-028/025)
`scheduling/models.py`	NEW 171	`AssignmentException`, `ShiftInstance` (ADR-029)
`weekly_structure/models.py`	NEW 244	`PositionTemplate`, `WeeklyBoard`, `Placement`, `DayOfWeek` — no `WeeklyTier` (ADR-025/031/032)

5. `API.md` comparison

Footer is STALE in Coveris-e: Last Updated: 2026-03-26 and the Scope: line are byte-identical to ours despite ~+464 lines of new endpoints. (MIGRATION-PLAN steps A1/B2 “verify footer/scope” → FAIL in source.)
New endpoints (body): GET /employees/{id}/capacity/, GET /demand/{id}/eligible-employees/, GET /calendar/holidays/, GET /scheduling/{ou}/{y}/{m}/, POST …/publish/, GET …/pdf/, GET/POST/GET/PATCH/DELETE /assignment-exceptions/; Position write adds start_time/end_time/pinned_slot_indexes.
🔴 API.md ↔ ADR-027 INCONSISTENCY (must fix before adoption): Coveris-e API.md §“Org Levels (ADR-026)” still documents the full /api/v1/org-levels/ CRUD + reorder, the OrgUnit.level nested object, and unit_type “survives one release”. ADR-027 deleted all of that (depth model, no OrgLevel, no unit_type) and added POST/DELETE /org-units/:id/manager — which API.md does not document. The code is already on depth. → Coveris-e’s org-units section cannot be adopted as-is; it must be rewritten to the depth + manager-slot contract.

6. Human-in-the-loop decision points (asked via the pick interface)

Org nucleus reshape (025/026/027): adopt ADR-027 depth+manager model (supersede ADR-018, rewrite ADR-001-a + org-units API) — load-bearing: scheduling/staffing/Semanal all sit on it — vs defer vs reject.
Staffing 028 vs 024: 028 supersedes 024 (hours_source replaces is_reinforcement) + amend ADR-001-c · vs coexist · vs defer.
Scheduling stack (029→031→030→032→034): adopt with MODS · as-is now · defer.
Doc drift + API.md: keep our adr-000/adr.base/adr-019/split-001 and rewrite org-units API.md to depth+manager (discard the stale Org-Levels section, bump footer) · vs take Coveris-e docs wholesale · vs per-file.

EXECUTION PLAN — ADR curation + API.md fix (⏳ AWAITING USER APPROVAL — 2026-05-31)

Autonomous /goal from the user (AFK). Hard gate: the user approves THIS plan before any ADR/API.md write begins. Nothing below is executed until then.

Guardrails (non-negotiable)

Current ADRs are never destroyed or edited destructively. No Status flips, no rewrites on adr-000…adr-024. (Old-ADR diff = cosmetic-only, except Coveris-e’s ADR-009 Draft→Accepted promotion, which we do not auto-adopt.)
Every imported ADR lands as Status: Proposed. Our branch’s code is still the OLD model (4-level unit_type, no scheduling/weekly_structure/hours_source); “Accepted” would contradict our code. Promotion to Accepted happens per backend phase when the code lands, with the user’s express OK (AGENTS.md HITL rule).
Drafting via the project’s cotton-coveris-backlog ADR-draft pipeline (Draft/Proposed only).
Keep Coveris-e’s ADR numbers (025–032, 034) — no collision with ours (we end at 024); matches the incoming code’s references.
No backend/, no frontend/, no push.

Old-ADR comparison (DONE — read-only Haiku pass)

All shared ADRs 000–024 changed COSMETIC-only (frontmatter/wikilink/heading/callout downgrade) → keep ours wholesale.
Exception: Coveris-e flips adr-009 / 009-a / 009-b Draft→Accepted → separate minor ruling; default = keep ours Draft unless the user says adopt.
Coveris-e lacks our adr-000 + adr.base, carries monolithic adr-001.md → ignore (our split + conventions win).

Phase 1 — New ADRs (per-ADR disposition; all writes = new files, Status Proposed)

ADR	Disposition	Notes / conflict handling (written INTO the ADR, not into ours)
025 weekly design surface	KEEP (Proposed)	Historical banner: its isolation + `WeeklyTier` are superseded by 027/reconciler
026 configurable org levels	DISCARD	Dead/superseded, never shipped; 1-paragraph history folded into 027’s Context
027 fixed-depth + manager slot	KEEP (Proposed)	In-ADR notes: “supersedes ADR-018 when depth code+API land” (not now). Flag open rulings: (a) `0.01h` manager-hours sentinel, (b) `AREA` = new 5th level vs ADR-021’s 4-level reasoning, (c) Spanish enum `CLINICA/AREA` vs ADR-001-b, (d) glossary rows pending. Carries a “Glossary additions (pending ADR-001-a)” block instead of editing ADR-001-a
028 staffing `hours_source`	KEEP (Proposed)	Prominent RECONCILE-BEFORE-ACCEPT block: must reconcile ADR-024 (`is_reinforcement` vs `hours_source`) + amend ADR-001-c (violet band). Neither edited now
029 monthly scheduling	KEEP (Proposed)	+ MODS: publish/staleness invariant stated in-ADR; explicit “`planned_hours`/`hours_delta` never feed the weekly ledger (ADR-011)”; glossary rows; fold/link the Q1–Q15 design decisions
030 position time-bands	KEEP (Proposed)	Rename file → `adr-030-position-time-ranges.md` (title/scope honesty; current filename says “schedule-architecture”)
031 closed-on-holidays	KEEP (Proposed)	Clean, lowest-risk
032 weekly rotation	KEEP (Proposed)	+ MODS: pin ISO-week tick to browser-local per ADR-015; decide year-boundary anchor
034 slot anchoring	KEEP (Proposed)	Gated after 032; note 033 intentionally absent

Net: write 8 Proposed ADRs, discard 1 (026), update docs/adr/README.md index. Zero edits to existing ADRs.

Minimalist alternative (user prefers discard): DISCARD ALL 9 now; re-introduce each ADR alongside its backend code phase (B0–B7). The §INTEGRATION FINDINGS above already preserves the full assessment, so nothing is lost. → reply “discard-all” to take this path instead.

Phase 2 — API.md fix (only after Phase 1 lands)

Bring API.md to the integration target, done correctly (fixing Coveris-e’s own mistakes), reflecting exactly the Proposed ADRs — via the cotton-coveris-live-docs pipeline:

Org-units: unit_type → depth + depth_name; add POST/DELETE /org-units/:id/manager; nesting rules → depth 0-4; do NOT import the dead /org-levels surface; coverage-summary.org_unit_type → depth_name.
Positions: add start_time / end_time / pinned_slot_indexes (+ day).
Assignments: hours_source; slot_index PATCH.
Staffing: GET /employees/{id}/capacity/, GET /demand/{id}/eligible-employees/.
Scheduling: GET /scheduling/{ou}/{y}/{m}/, POST …/publish/, GET …/pdf/; assignment-exceptions CRUD; GET /calendar/holidays/.
Bump footer Last Updated + Scope line. Mark new endpoints as target/planned per ADR-005-b (code catches up during migration).
Conservative alternative: add endpoints phase-by-phase as backend code lands → reply “api-phased”.

Will NOT do without further OK

Promote any ADR to Accepted · edit/supersede any existing ADR (018/024/001-a/001-c/009…) · touch backend/ or frontend/ · push.

Verification

After Phase 1: git diff coveris-b-integration -- docs/adr/adr-000*…adr-024* is empty (existing ADRs untouched); only new adr-025/027/028/029/030/031/032/034 files + README index appear; no adr-026.
After Phase 2: contract-coverage guard parses every ### METHOD /path; footer bumped; zero /org-levels references.

Three quick confirmations for the user (defaults in bold)

ADR set: keep-8-as-Proposed + discard-026 · or discard-all.
API.md: target-state rewrite (correct, no org-levels) · or api-phased.
ADR-009 Draft→Accepted from Coveris-e: keep ours Draft · or adopt-009.

Reply “approve” for all defaults, or override any of the three.

DRY-RUN SIMULATION RESULTS — 2026-05-31 (Agent Team `coveris-migration-dryrun`)

Read-only simulation of every phase of this plan against the real code in BOTH worktrees. No git checkout, no migrate, no edits to code — verified only via git ls-tree / git show <ref>:<path> / git diff. Three Opus teammates (backend-dryrun, docs-dryrun, frontend-dryrun) each dry-ran a tier; the orchestrator independently corroborated every load-bearing migration-dependency claim against the actual migration files.

This section is the executable-reality complement to the two ADR/docs-focused sections above (§INTEGRATION FINDINGS and §EXECUTION PLAN). Those decide which ADRs/API contract to adopt; this decides whether each migration phase mechanically applies, and ends with the consolidated user questionnaire.

Source: Coveris-e @ 6bfd1bd (⚠️ has uncommitted working-tree edits — see §6) · Target: coveris-b-integration @ 8c6f479 · merge-base: ed3b4c8. Coveris-e is 178 commits ahead of the merge-base; target is 42 ahead; 617 files diverge (+52.8k / −104.1k).

0. Headline corrections to the plan’s premises

#	The plan assumes	Reality (verified)
0.1	Target branch is `main` (`fdf0a90`)	Target is `coveris-b-integration` @ `8c6f479`. The header + §“Estado al cierre” are stale.
0.2	Source `Coveris-e` HEAD `f371f20`	Source HEAD is `6bfd1bd` with a dirty working tree (uncommitted edits to API.md, ADRs 030/032/034, api-conventions, 6 backend files, 4 frontend files).
0.3	`git checkout Coveris-e -- X` brings the intended content	It brings the committed `6bfd1bd` content and silently reverts any file dirty in the source worktree (§6). The single biggest trap in the plan.
0.4	Migrations can be cherry-picked per phase (“0003 through 0010”, “skip 0007”)	Django migrations are an indivisible linear per-app chain. Several phase splits are mechanically impossible (§2).
0.5	`.env` hard rules	✅ Verified safe: `.env` gitignored on both branches, tracked nowhere; no `SUPERUSER_PASSWORD` exposure.
0.6	(the §EXECUTION PLAN guardrails)	✅ Real — they live in the §EXECUTION PLAN section of THIS doc (no separate file). This dry-run aligns with them: every imported ADR lands `Proposed`; 026 discarded; existing ADRs untouched.

1. Phase verdict matrix

Legend: 🟢 GREEN (works as written) · 🟡 NEEDS-FIX (works after a correction) · 🔴 BLOCKED (cannot work as written).

Phase	Plan intent	Verdict	Load-bearing problem	Required correction
A0	ADRs 025–034 + README	🟡	ADR-file glob is safe (033 truly absent; monoliths not dragged). But wholesale `checkout docs/adr/README.md` drops our `adr-000`/`adr.base`/frontmatter/wikilinks and reintroduces the monolithic `adr-001` row. Committed statuses ship 030/032/034 as Proposed (not Accepted).	Keep the ADR-file glob. Do NOT checkout README — hand-append rows into OUR README (per §EXECUTION PLAN Phase 1).
A1	PRD §3/§8.4, API.md, conventions, docs 07/08/09	🟡	PRD live file is root `PRD.md` (`docs/PRD.md` is 0 bytes); wholesale checkout strips frontmatter/wikilinks, inlines a 23-row tag table, reverts the 001-c ref, and leaves §8.4 open. API.md footer byte-identical/stale, committed copy lacks `slot_index`, org-units section contradicts ADR-027 (§3). Paths `docs/07/08/09-.md` don’t exist in either branch*; `api-conventions.md` moved to `docs/stack/`.	Keep B1 as a surgical §3+§8.4 hand-edit on root PRD.md. Rewrite API.md per §EXECUTION PLAN Phase 2. Drop the 07/08/09 steps; reconcile `docs/stack/api-conventions.md` by path.
B0	demand: checkout “0003 → 0010” + models.py	🔴	`demand/0003` deps `('demand','0002_staffing_plan_line')`; target has only `0001`. Skipping 0002 = broken graph. models.py brought is the final state (StaffingPlanLine, max_weekly_extra_hours, Position time fields) whose migrations are 0002/0011/0013-0016 → `makemigrations` drift.	Collapse B0+B5 into one phase: checkout demand migrations `0002 → 0016` as one contiguous block + models.py + views.py. The chain is unsplittable.
B1	tags/ + business_rules/	🟢	tags 0003-0005 + business_rules 0002 are intra-app (no cross-app deps; `seed_rules` RunPython own-app idempotent).	⚠️ `seed_tags.py` dirty in source (DEBUG-guard) — re-apply after checkout (§6).
B2	api/fsm/users “minor” + seeds	🟡	Not minor: +1178/−936 across 12 files. `seed_notti_structure.py` (apps/api) is unlisted but called by `entrypoint.sh:32` → boot failure. `seed_transition_logs.py` deleted. 3 dirty source seeds/test.	Add `seed_notti_structure.py`; re-apply 3 dirty diffs; decide fate of removed seeds (§7-Q4).
B3	assignments core, hold back `slot_index`	🟡	`slot_index` lives in models.py + serializers.py (4 sites: tuples ln 32/58/82, extra_kwargs 85, `validate_slot_index` 88-94) — `views.py` has ZERO references (no PATCH endpoint; rides DRF `partial_update`). Plan strips the wrong file.	Strip from models.py and serializers.py, ignore views.py — or drop the B3/B7 split entirely (§7-Q2).
B4	weekly_structure full	🔴	`ws/0001` deps `demand/0002`; `ws/0003` deps `demand/0004`; `ws/0004` RunPython replays `OrgLevel` + `OrgUnit.level` + `WeeklyTier` (transient — created mid-chain, dropped by `demand/0009` & `ws/0005`). History-replay: only works if the full demand 0002-0009 chain ran in order first. No squash shortcut.	Gate B4 on the fixed B0 (full demand 0002-0016). Then ws 0001-0012 apply cleanly.
B5	demand 0013-0016	🔴	`0013` deps `0012` deps `0011` deps `0010` — unscheduled before B5.	Redundant once B0 is fixed — fold B5 into B0 (indivisible chain).
B6	calendar_api + “Migration 0001” + pytest	🔴	`calendar_api` has no models.py, no migrations/, no tests/ — 5 files (`__init__, apps, services, urls, views`), a pure external-holiday-API service. Migration + pytest steps are fiction.	Plain checkout of the 5 service files + wire `urls` into `config/urls.py`; drop migration/pytest steps; smoke `/holidays/`.
B7	scheduling + restore slot_index	🟡	`scheduling/0001` deps `api/0001` + `assignments/0004` + `demand/0012` → gated on fixed B0 + B3≥0004. slot_index restore must target models.py + serializers.py, not views.py. Smoke needs B6 wired first.	Correct slot_index targets; sequence B6 before the B7 smoke.
C0	mock-data + docker + entrypoint	🟡	docker-compose `./mock-data:/mock-data:ro` ✅. entrypoint.sh ✅ changes — but new boot calls `seed_anesthesia_rotation` and `seed_notti_structure` (unlisted) and drops 6 legacy seeds. mock-data dirs diverge (Coveris-e: `notti-anesthesia/`+`notti-structure/`; target: `clinic-bienestar/`). `seed_anesthesia_rotation.py` dirty in source.	Add `seed_notti_structure.py` + `mock-data/notti-structure/`; re-apply dirty guard; resolve clinic-bienestar (§7-Q3).
C1	coverage-summary Decimal fix	🟢	2-line `sum(..., Decimal("0"))` in `demand/views.py`; lands with views.py in the fixed B0.	Keep as a verification checkpoint, not a standalone commit.
D0–D9	frontend (deferred)	🟡	Directionally right but materially under-scoped (§5): ~11k LOC net-new, 2 wrong line counts, 1 no-op phase, 2 omitted prerequisite buckets, destructive org-chart reconcile.	Re-scope per §5 before starting; net-new FE areas are 1:1 coupled to specific net-new backend apps.

2. The load-bearing correction — demand is ONE indivisible chain

The plan’s central mistake is treating Django migrations like cherry-pickable patches. Verified dependency declarations:

demand/0003_add_org_level        → deps [('demand','0002_staffing_plan_line')]
demand/0013_position_time_range  → deps [('demand','0012_seed_clinic_extra_hours_cap')]
weekly_structure/0001_initial    → deps [('demand','0002_staffing_plan_line'), AUTH_USER_MODEL]
weekly_structure/0004_materialize_tiers_as_org_units → RunPython reads OrgLevel + OrgUnit.level + apps.get_model('weekly_structure','WeeklyTier')
scheduling/0001_initial          → deps [('api','0001'),('assignments','0004_backfill_hours_source_cargo'),('demand','0012_seed_clinic_extra_hours_cap'), AUTH_USER_MODEL]

Consequences that rewrite the phase plan:

demand 0002→0016 must be one phase (collapse B0 + B5). You cannot bring 0003-0010 without 0002, nor 0013-0016 without 0011-0012.
weekly_structure (B4) can only migrate by replaying demand’s deleted intermediate state — hard-gated on the full demand chain running in order; no squash.
scheduling (B7) is gated on demand≥0012 + assignments≥0004 + api, and its smoke needs calendar_api (B6).
assignments slot_index is a models.py + serializers.py concern (4 sites), never views.py.

3. Tier A (docs) — verified detail

A0 committed statuses (= what a checkout ships): 025 Accepted · 026 Superseded-by-027 · 027 Accepted · 028 Accepted · 029 Accepted · 030 Proposed · 031 Accepted · 032 Proposed · 034 Proposed. Source-dirty edits flip 030/032/034 → Accepted, but the plan checks out committed → ships them Proposed (which the §EXECUTION PLAN “all Proposed” guardrail wants anyway).
A0 README hazard: Coveris-e’s README links the monolithic adr-001/002/003/005/009/012 via plain .md links and has no adr-000/adr.base row → wholesale checkout regresses our vault conventions. Hand-append instead.
A1 API.md ↔ ADR-027 (verified quotes): API.md :1205-1213 still documents /api/v1/org-levels/ GET/POST/PATCH/DELETE + reorder; :1223,1246 keep unit_type “survives one release”; :1245 embeds nested level. ADR-027 :105 “Drop OrgLevel model, OrgUnit.level FK, and the unit_type cache”; :132-134 adds POST/DELETE /api/v1/org-units/:id/manager — grep /org-units/.*/manager in Coveris-e API.md → 0 hits. Shipped code is already on depth. ⟹ org-units API section is unadoptable as-is; rewrite per §EXECUTION PLAN Phase 2.
A1 API.md footer byte-identical to ours (Last Updated: 2026-03-26) despite new endpoints → the plan’s “verify footer matches today” step fails in source; bump by hand.

4. Tier B/C (backend) — verified detail

seed_notti_structure.py (apps/api) is the silent boot-breaker: present in Coveris-e, called by entrypoint.sh:32, in no checkout list → add to B2 + C0.
entrypoint.sh boot (Coveris-e): create_superuser → seed_mock_users → seed_org_units → seed_tags → seed_anesthesia_nomina → seed_therapy_nomina → seed_anesthesia_rotation → seed_notti_structure. Legacy seed_employees/positions/assignments/employee_tags/position_tags/transition_logs no longer booted (seed_transition_logs.py deleted upstream).
audit_log/ and offer/ are byte-identical across branches — plan correctly ignores them. demand/0001_initial identical across branches (0003’s AlterField is valid against target).
The hours_source quartet (demand 0011, tags 0003, assignments 0003, business_rules 0002) is all intra-app — the plan’s split across phases is graph-safe per se; the only danger is the linear-chain gaps inside demand/assignments.

5. Tier D (frontend) — scope reality-check (deferred tier, informational)

Diffstat: ~87 files, +18,965 / −1,841 — most volume is net-new dirs, not edits.
Net-new (add-a-module): features/scheduling/ (6), features/structure/weekly-structure/ (19), features/structure/staffing-plan/ (3), shared/components/staffing/ (7), 8 net-new core/models. The last two buckets are omitted from D0–D9 and are prerequisites for D3/D4/D5.
Wrong line counts: scheduling-month-grid = 1,690L (plan: 1142); WeeklyStructureStore = 834L (plan: 818, and it’s 1 of 19 files / ~4k-LOC dir).
D7 “help system” is a NO-OP — features/help/ is byte-identical between branches.
D0 org-chart is destructive: Coveris-e deletes assign-drawer, candidate-assign-panel, node-detail-drawer, and worker-detail-drawer (not re-added); only manager-badge/manager-dialog are added. Plan phrasing implies additive (§7-Q10).
Hard backend coupling: net-new FE services hit /weekly-structure/, /scheduling/, /staffing-plan/, /employees/{id}/capacity/ (backend apps absent on target) → D3/D4/D5 + roster-capacity can’t precede their Tier-B phases.
frontend/design_handoff_scheduling_redesign/ (8-file JSX/CSS prototype) is already excluded by this plan’s hard rules ✅ — keep excluded (reference-only).

6. The dirty-source trap — files a plain `git checkout` silently reverts

git checkout Coveris-e -- X ships committed 6bfd1bd content. These files are dirty in the source worktree and would be silently reverted (their uncommitted edits lost):

File	Uncommitted edit that would be lost
`backend/apps/tags/management/commands/seed_tags.py`	`raise CommandError` DEBUG-guard
`backend/apps/api/management/commands/seed_anesthesia_nomina.py`	DEBUG-guard
`backend/apps/api/management/commands/seed_therapy_nomina.py`	DEBUG-guard
`backend/apps/scheduling/management/commands/seed_anesthesia_rotation.py`	DEBUG-guard
`backend/apps/api/tests/test_seed_anesthesia_nomina.py`	+7-line test
`backend/tests/test_regression_backlog.py`	expects start_time/end_time/pinned_slot_indexes (030/032) + slot_index (034)
`API.md`	adds the `Assignment.slot_index` contract row + ADR-034 notes
`docs/adr/adr-030/032/034-*.md`	Status `Proposed → Accepted`
`docs/api-conventions.md`	adds frontmatter + wikilink conversion
`docs/adr/README.md`	(dirty)
`frontend/.../manager-dialog.component.ts` · `employee-form.component.ts` · `staffing-plan.service.ts`	`httpResource.value()` → `.hasValue()` guards
`frontend/.../layout/component/app.menu.ts`	nav hover-bold width-stability CSS fix

Mitigation (strongly recommended): commit these edits on Coveris-e before running the migration so every checkout carries the intended content. Otherwise each phase must hand-re-apply its diffs. (§7-Q5.)

7. USER QUESTIONNAIRE — the exit (decisions only you can make)

The dry-run is mechanically complete; the remaining blockers are decisions, not code lookups. Answer these tomorrow and the plan can be rewritten into executable form. Defaults reflect the team’s recommendation. ADR/API decisions are NOT repeated here — they live in the §EXECUTION PLAN “Three quick confirmations” above; this dry-run only confirms they’re mechanically migratable.

Forced by code (confirm you accept the restructure):

demand phasing — Collapse plan B0 + B5 into one indivisible phase feat(demand): checkout migrations 0002→0016 + models.py + views.py? The chain is mechanically unsplittable — there is no alternative that applies cleanly. Default: yes.
assignments slot_index split — Keep the B3-hold-back / B7-restore dance (corrected to strip/restore in models.py + serializers.py, not views.py), or land assignments whole in one phase (simpler; slot_index is the additive chain tail)? Default: land whole — drop the split.

Dataset & seed strategy: 3. clinic-bienestar keep-or-drop — Coveris-e’s entrypoint stops seeding mock-data/clinic-bienestar/ and switches to notti-anesthesia/ + notti-structure/. Keep clinic-bienestar alongside, or fully switch to the Notti dataset? Default: switch to Notti (matches the new entrypoint); archive clinic-bienestar. 4. Removed seed commands — seed_employees/seed_positions/seed_assignments/seed_employee_tags/seed_position_tags/seed_transition_logs are no longer booted (seed_transition_logs.py deleted upstream). Delete from the repo, or keep dormant? Default: drop the upstream-deleted one; keep the rest only if still referenced by tests. 5. Commit dirty source first? — Commit the 6 backend + 4 frontend + API.md/ADR/api-conventions uncommitted edits on Coveris-e before migrating (so plain checkout carries them), or hand-apply per phase? Default: commit-first — removes the §6 trap entirely.

Frontend (Tier D — for when it’s scheduled): 6. org-chart deletions — Confirm the 4 component deletions on target (assign-drawer, candidate-assign-panel, node-detail-drawer, worker-detail-drawer) are intended (D0 is a destructive reconcile, not additive)? 7. Re-scoped D plan — Accept adding a “D-pre: shared-staffing kit + scheduling/capacity models” phase, dropping the no-op D7, and fixing the D3/D4 line counts? Default: yes.

Carry-over (still open from §INTEGRATION FINDINGS-6, restated not re-decided): the ADR human-in-the-loop picks — org-nucleus reshape 025/026/027 (supersede ADR-018, rewrite ADR-001-a), staffing 028 vs 024 (+ amend ADR-001-c violet band), scheduling stack 029→031→030→032→034. Those remain your decisions; this dry-run verified only that the code matching them is mechanically migratable once the phasing above is corrected.

Dry-run by: Agent Team coveris-migration-dryrun (backend-dryrun, docs-dryrun, frontend-dryrun — Opus) + orchestrator corroboration, 2026-05-31. Method: read-only git inspection of both worktrees; zero mutations to code or branches.

Last edited: 2026-05-31 — Opus orchestrator + Agent Team coveris-migration-dryrun: appended §DRY-RUN SIMULATION RESULTS (per-phase verdicts, migration-graph reality, dirty-source trap, consolidated questionnaire). Prior 2026-05-31: Opus appended §EXECUTION PLAN (ADR curation + API.md fix, awaiting approval). Prior 2026-05-31: coveris-adr-review Agent Team — ADR assessment, models/API diff, HITL findings. Prior 2026-05-24: sesión Opus que integró luinloder y armó mock data de anestesia.