session-h-deploy

# Session H Deployment & Backfill ## Prerequisites - Claude Code has confirmed Session H code is complete and committed - `scripts/backfill_v31_rolling.py`, `scripts/backfill_v32_zscores.py`, and `scripts/migrations/drop_bdl_stat_id.py` all exist - Railway workspace token is loaded (run `. ./scripts/load-env.ps1 -Force` if you see "Unauthorized") ## Workflow — Execute in EXACT Order ### Step 1: Pre-Deploy Validation ```bash # Syntax check modified files for f in $(git diff --name-only HEAD | grep '\.py$'); do python -m py_compile "$f" || { echo "SYNTAX ERROR in $f"; exit 1; } done # Run critical test subset python -m pytest tests/test_waiver_integration.py tests/test_mlb_analysis.py -q --tb=short ``` - If tests fail → STOP, report output to Claude Code ### Step 2: Deploy ```bash railway up ``` - Wait for build to complete (~2-5 minutes) ### Step 3: Verify deployment health ```bash curl -s --max-time 20 https://fantasy-app-production-5079.up.railway.app/health ``` - Expected: `{"status":"healthy","database":"connected","scheduler":"running"}` - If not healthy → run `railway logs | head -50`, report errors to Claude Code ### Step 4: Drop bdl_stat_id column ```bash railway run python scripts/migrations/drop_bdl_stat_id.py ``` - Expected output: `bdl_stat_id dropped (or did not exist).` ### Step 5: V31 rolling backfill (allow up to 10 minutes) ```bash railway run python scripts/backfill_v31_rolling.py --execute ``` - Watch for progress logs every 500 rows - Report final "N rows updated" count ### Step 6: V32 z-score backfill (run ONLY after Step 5 completes) ```bash railway run python scripts/backfill_v32_zscores.py --execute ``` - Same pattern. Report final count. ### Step 7: Re-trigger valuation cache ```bash curl -s -X POST --max-time 30 https://fantasy-app-production-5079.up.railway.app/admin/refresh-valuation-cache ``` - Report the response body ### Step 8: Spot-check queries Run via `@postgres` MCP or `railway run python -c` with psycopg2: ```sql SELECT COUNT(*) AS w_runs_populated FROM player_rolling_stats WHERE w_runs IS NOT NULL; SELECT COUNT(*) AS z_r_populated FROM player_scores WHERE z_r IS NOT NULL; SELECT scarcity_rank, COUNT(*) AS cnt FROM position_eligibility GROUP BY scarcity_rank ORDER BY scarcity_rank; SELECT quality_score IS NULL AS is_null, COUNT(*) FROM probable_pitchers GROUP BY quality_score IS NULL; SELECT COUNT(*) AS rows FROM information_schema.columns WHERE table_name='mlb_player_stats' AND column_name='bdl_stat_id'; ``` **Expected results:** - `w_runs_populated` > 60,000 - `z_r_populated` > 60,000 - `scarcity_rank` has non-null values (1-12) - `quality_score IS NULL` = false has count > 0 (no nulls) - `bdl_stat_id` column count = 0 (column dropped) ## Rules - **NEVER** skip Step 1 (pre-deploy validation) - **NEVER** run Step 6 before Step 5 completes (V32 depends on V31 data) - If ANY step fails → STOP, capture exact error output, and report to Claude Code - Do NOT retry failed steps without Claude Code approval - Do NOT modify scripts or code — only run what Claude Code prepared ## Report Format For each step, report: 1. Step number and name 2. Result (PASS / FAIL) 3. Key output or row counts 4. Any errors (verbatim)