refactor(worker): scheduler jobs split into vrp/surface/ops/market_data families (Wave 7a)

CHANGELOG.md

@@ -10,6 +10,7 @@ version in lockstep (enforced by `scripts/release/version_sync_check.py`).
 ### Changed
 
 - **The data gap healer's 1,154-line `worker/jobs/data_gap_adapters.py` splits by role (Wave 7b).** `data_gap_heal_context.py` holds `RequestBudget`, `HealContext`, `HealSpec` and `_beat`; `data_gap_heal_runners.py` holds the `_run_*` writers, `_eventlog_heal` and `_replay_run_single_stock`; `data_gap_adapters.py` keeps only `HEAL_SPECS` and `run_refresh_adapters`; `data_gap_executor.py` holds `_verify_covered`, the dispatchers, `_verify_and_mark` and `execute_run`. Imports run one way (context ← runners ← adapters ← executor) and there is no re-export shim, so every importer now names the new home. The two test monkeypatches follow the module that looks the name up: `_verify_covered` on `data_gap_executor`, `_replay_run_single_stock` on `data_gap_heal_runners`. Function bodies are unchanged. No runtime behaviour changes.
+- The worker scheduler's last 38 inline `add_job` sites (42 jobs) moved out of `worker/scheduler.py` into four job families under `worker/schedule/`: `vrp.py` (11), `surface.py` (8), `ops.py` (5) and `market_data.py` (14: 13 jobs plus the loop that registers the 5 UW historical-alpha captures). Each block keeps its role guard, and the scheduler-jobs golden is byte-identical for every role. `scheduler.py` goes from 1115 to 191 lines; `main()` now holds only settings, the schema gate, the failure listener, the heartbeat and start. No job, trigger or schedule changed. (Wave 7a)
 
 ## [0.13.25] — 2026-10-05
 

CLAUDE.md

@@ -199,7 +199,7 @@ Worker roles: `ai-deepseek` (provider-pinned, what prod runs); legacy `ai` (clai
 | API contract models                                                                         | `src/uw_scan/models/` (`__init__.py` is export-only; implementations live in domain modules)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |
 | Persistence — aggregate shim                                                                | `src/uw_scan/storage/repository.py` (thin `Repository` assembly; exports only `Repository`, so import other names from their domain module; do not add query methods here)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                         |
 | Persistence — domain modules                                                                | `src/uw_scan/storage/{audit,cockpit,external_api,fetchers,flow,gex,gold,gold_etf,health,jobs,market_data,matrix_state,options,rates_repository,scan_outputs,scan_results,scan_runs,trade_insights_ai,volatility_raw,volatility_v2,watchlist,ws_consumer_state}.py`. New domains go in their own module/mixin and are added to `repository.py` only for assembly/re-export compatibility                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
-| Scheduled jobs                                                                              | `src/uw_scan/worker/scheduler.py` (`main()`: settings, schema gate, listener, the jobs not yet in a family, start) + `worker/schedule/{gold,macro,regime,fundamentals,scan_core,ai}.py` (one `register(sched, settings[, *, ticker_filter])` per job family, each block keeping its role guard) + `worker/schedule/roles.py` (`_pinned`, `_owns_global_daily_jobs`, `_worker_groups`, `_is_primary_worker`, ticker sharding) + `worker/db.py` (`repo_session`, `external_api_recorder`, `uw_client`, `research_budget_ok`). Family modules never import `scheduler`. Golden of every role's jobs: `tests/unit/worker/test_scheduler_jobs_golden.py` |
+| Scheduled jobs                                                                              | `src/uw_scan/worker/scheduler.py` (`main()`: settings, schema gate, listener, heartbeat, start) + `worker/schedule/{gold,macro,regime,fundamentals,scan_core,ai,vrp,surface,ops,market_data}.py` (one `register(sched, settings[, *, ticker_filter])` per job family, each block keeping its role guard) + `worker/schedule/roles.py` (`_pinned`, `_owns_global_daily_jobs`, `_worker_groups`, `_is_primary_worker`, ticker sharding) + `worker/db.py` (`repo_session`, `external_api_recorder`, `uw_client`, `research_budget_ok`). Family modules never import `scheduler`. Golden of every role's jobs: `tests/unit/worker/test_scheduler_jobs_golden.py` |
 | Advisory locks (single-flight jobs) | `src/uw_scan/storage/advisory_locks.py` — the one registry of lock keys (import asserts no two names resolve to the same key) + `single_flight()` (session-scoped `pg_try_advisory_lock`; a held lock means skip). Add a new key here, never a magic number in a job |
 | Worker schema gate (deploy ordering) | `src/uw_scan/worker/schema_gate.py` + migration `158_schema_version.sql` — `migrate_runner` writes the last applied file into `uw_scan.schema_version` after a full apply; `scheduler.main()` waits (INFO, no job failures) until that is at least the newest migration in its own image, because Watchtower ignores `depends_on` |
 | UW daily-budget governor (live/research pools, hot fast lane)                               | `src/uw_scan/sources/uw_budget.py` (pool split + `may_spend` + `read_snapshot`) + `worker/schedule/scan_core.py` `_live_max_tickers` + `worker/db.py` `research_budget_ok` + `worker/jobs/full_scan_hot.py` + config `uw_*_ceiling`/`full_scan_hot_*`/`gex_scan_*_interval_minutes` + migration `096` (`watchlist.hot`) + `web/components/watchlist/HotButton.tsx`; plan `docs/superpowers/plans/2026-07-04-uw-budget-scaling.md`                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                       |

src/uw_scan/worker/CLAUDE.md

@@ -3,7 +3,7 @@
 ## Files
 
 - `scheduler.py` — `BlockingScheduler` entrypoint (`python -m uw_scan.worker.scheduler`). **The authoritative job wiring**, together with `schedule/*.py` — `jobs/` has ~73 modules; read `scheduler.py` plus the family modules for what actually runs, when, and on which worker role.
-- `schedule/{gold,macro,regime,fundamentals,scan_core,ai}.py` — one job family each, registered from `scheduler.main()` via `register(sched, settings[, *, ticker_filter])`; each block keeps its role guard. `schedule/roles.py` holds the role/shard predicates and `db.py` the shared `repo_session` / `external_api_recorder` / `uw_client` / `research_budget_ok`. A family module never imports `scheduler` (circular import); the golden `tests/unit/worker/test_scheduler_jobs_golden.py` pins every role's job list.
+- `schedule/{gold,macro,regime,fundamentals,scan_core,ai,vrp,surface,ops,market_data}.py` — one job family each, registered from `scheduler.main()` via `register(sched, settings[, *, ticker_filter])`; each block keeps its role guard. `schedule/roles.py` holds the role/shard predicates and `db.py` the shared `repo_session` / `external_api_recorder` / `uw_client` / `research_budget_ok`. A family module never imports `scheduler` (circular import); the golden `tests/unit/worker/test_scheduler_jobs_golden.py` pins every role's job list.
 - `jobs/full_scan.py` / `jobs/ohlc_pull.py` / `jobs/rescan_loop.py` — the core scan/OHLC/rescan trio; the rest of `jobs/` is per-feature (gold, rates, regime, skew, `vrp_*`, `option_surface_*`, data_freshness/gap, `trade_insights_ai*`, …)
 - `volatility_jobs.py` — `daily_spy_ohlc_refresh`, `nightly_vol_analytics_rollup` (Volatility tab v2)
 - `massive_ws_consumer.py` + `ws_tick_buffer.py` + `ws_db_writer.py` — the standalone spot WS consumer process (see below)