# Weekend picks — supervised work queue

Date: 2026-05-16

**Hard filters:**
- Meets all EXECUTION_READY criteria EXCEPT they're LOW_REVERSIBILITY
- supervised_execution_recommended == true

These are NOT in the autonomous Saturday queue. They ARE the highest-impact work available — they just need James actively reviewing. **Pair on these in a Sunday or post-Saturday session.**

---

## Supervised Pick 1 — Catalog refactor: cut 5 surfaces over to /system/strategies.json

- **Task id:** catalog-refactor-shared-json
- **Property:** foundation (INFRA / CROSS-CUTTING)
- **Time horizon:** near_term
- **Estimated effort:** L
- **Tier:** 1 compounding (highest-leverage task on the board, period)
- **Reversibility:** LOW (touches 5 surfaces; cutover-shaped)
- **Leverage justification:** Compounding across 5 active properties (scorecards, strategy_library, live_classifier, param_surface, playbooks). Direct unlock_count=4. Multiple drift findings point at this as the root cause — `split-td-vwap-from-vwap-pull-in-in-classifier` task explicitly documents that `_scientific_overlay.js` and `gen_scorecard.py classify_entry()` have already diverged. Every future strategy change touches 5 files today; this makes it touch 1.
- **Why supervised:** Touches 5 production-rendered surfaces. If `/system/strategies.json` schema differs subtly from the inlined copies, all 5 surfaces misrender simultaneously on the same commit. Per-surface visual parity check is the safety net.
- **Done when (with verification + rollback per LOW rules):**
  - `/system/strategies.json` is the canonical (post-5/15 migration; confirmed)
  - `_scientific_overlay.js` fetches `/system/strategies.json` instead of inlining
  - `gen_scorecard.py classify_entry()` reads `/system/strategies.json` via Python helper
  - `classify/*.json` files reference the same canonical IDs
  - `/strategies/` page renders from `/system/strategies.json`
  - **Per-surface visual parity check:** before/after screenshots match for /strategies/, scorecard pie, classify HUD, /playbooks/ scanner, /param-surface/
  - **5-trading-day observation window** with no downstream regressions
- **Rollback:** Each surface can fall back to its prior inline catalog via `git revert <commit>`. The canonical JSON itself stays (it's already in use by /strategies/).
- **Files likely touched:** `_scientific_overlay.js`, `gen_scorecard.py`, `classify_v2.py`, `playbooks.json`, `gen_param_surface.py`, `/strategies/index.html`

**This is the headline pick.** If we ship one supervised piece this weekend, it's this one.

---

## Supervised Pick 2 — Add role + is_approved columns to main.users

- **Task id:** add-user-role-approval
- **Property:** foundation (INFRA / CROSS-CUTTING)
- **Time horizon:** near_term
- **Estimated effort:** S
- **Tier:** 2 direct (unlock_count=4)
- **Reversibility:** LOW (ALTER TABLE with downstream consumers)
- **Leverage justification:** Unlocks 4 downstream tasks: `strategy-upload-ui`, `roberto-classifier-overrides-persist`, `phase-4-trader-selector`, `monthly-per-trader-view`. The "schema half" of the identity story — without role+is_approved, none of the multi-trader features can ship.
- **Why supervised:** Schema migration with downstream consumers. James pairs on the migration timing + downgrade plan + which consumers begin reading the columns first.
- **Done when:**
  - `main.users` has `role VARCHAR(20) DEFAULT 'trader'` and `is_approved BOOLEAN DEFAULT false`
  - James's row updated to `role='admin', is_approved=true`
  - gen_scorecard.py + accountability-v2 + strategy-upload-ui can read role from a single join
  - A test trader row with `role='trader', is_approved=false` renders correctly on each consumer
  - **5-trading-day observation window** after deployment
- **Rollback:** `ALTER TABLE main.users DROP COLUMN role, DROP COLUMN is_approved`. Any data set on those columns is lost — accept that this is metadata only, not load-bearing data yet.
- **Files likely touched:** `Z:\setup_daily_reporting\reports\migrations\scientific_layer\<NNN>_add_user_role_approval.sql`, possibly README

---

## Conditional Pick — Daily summary view (becomes #1 supervised if migrations run)

- **Task id:** p0-daily-summary-view
- **Property:** scorecards (per todos.json) — should arguably be `foundation`
- **Time horizon:** immediate
- **Estimated effort:** M
- **Tier:** 1 compounding (3 active properties)
- **Reversibility:** LOW (SQL view consumed by 3 surfaces)
- **Status:** currently STILL_PENDING_REAL because blocked on user-owned `run-scientific-layer-migrations` (P0, 24 days old)

**If James runs the scientific_layer migrations 001-009 this weekend**, this task becomes the #1 supervised pick:
- Resolves the two P0 acct-v2 bugs (risk-bump + ticket-meter) as side effects
- Makes scorecards + accountability + monthly read identical P&L by construction
- Eliminates the entire class of divergent-arithmetic bugs

**The migration is the blocker.** Once user runs `set_db.sh start` + `psql -f 001...009`, this task moves to supervised pick 1 ahead of catalog-refactor.

---

## Conditional Pick — Strategy_id column on indicator_alerts + main.journal

- **Task id:** add-strategy-id-column-to-indicator-alerts-main-jo
- **Property:** scorecards
- **Time horizon:** immediate (P0)
- **Estimated effort:** M
- **Tier:** 2 direct (unlocks live grade column on HUD)
- **Reversibility:** LOW (ALTER TABLE + backfill via classifier)
- **Status:** STILL_PENDING_REAL because depends on `catalog-refactor-shared-json` (Supervised Pick 1)

After Supervised Pick 1 ships, this is the natural sequence — give the classifier output a stable column to write to. **Don't ship before the catalog cutover** or the backfill will use the wrong source of truth.

---

## Other-owner / out-of-scope (not in queue, listed for awareness)

- **mgp-cron-skipped-5-14** (P0, owner=James) — scheduler diagnostic, needs schtasks /query on the Windows machine.
- **mgp-historical-pages-cleanup** (P0, owner=james) — compliance follow-up; James must approve the deletion list.
- **acct-v2 bugs** (2× P0, owner=jimbo) — jimbo's lane.
- **journal-user-steps** (P0, owner=user) — the 3-step journal persistence finish.
- **run-scientific-layer-migrations** (P0, owner=user) — gates p0-daily-summary-view + 4 other views.
- **p0-indicator-alerts-ddl** (P0, owner=user) — gates the webhook enrichment cluster.

---

## Summary of supervised queue

| Pick | id | tier | rev | est | downstream |
|---|---|---|---|---|---|
| 1 | catalog-refactor-shared-json | 1 compounding | LOW | L | unlocks 4 + retires drift across 5 surfaces |
| 2 | add-user-role-approval | 2 direct | LOW | S | unlocks 4 multi-trader tasks |
| (cond) | p0-daily-summary-view | 1 compounding | LOW | M | resolves 2 P0 acct-v2 bugs; unlocks 3 surfaces' identical P&L |

**Pick 1 is the supervised headline.** The two-queue split exists precisely because compounding-infra tasks like this one belong in a paired session, not a hands-off Saturday — they're too high-stakes for autonomous execution, but they're THE highest-leverage work available.

If James clears the user-owned migration backlog this weekend (`run-scientific-layer-migrations` + `p0-indicator-alerts-ddl`), the supervised queue depth doubles overnight.
