# Wave 1 Integration Layer M1 Ship Summary ## Commit Ladder | Step | Commit | Primary artifact | | --- | --- | --- | | Wave 1 Step 1 | `f76c9de` | Triple-confirmed signals data layer and `/analytics/triple-confirmed` | | Wave 1 Step 2 | `0c56daf` | Consensus disagreement split view at `/betting/consensus?view=split` | | Wave 1 Step 3 | `0ab5368` | Savant-style player percentile strips | | Wave 1 Step 4 | `7384ac4` | Projection percentile bands on season and projection pages | | Wave 1 Step 5 | This commit | Verification and this ship summary | ## Executive Summary The audit recommended four Wave 1 integrations: triple-confirmed signals, a disagreement-first consensus view, Savant-style percentile strips, and BP-style projection bands. All four shipped. The key product change is that Mithrandir now intersects its four data layers rather than displaying them as separate islands: projections, live state, regression candidates, and three-lane betting consensus can now meet on a single player row. ## Feature 1: Triple-Confirmed Signals Surface: `/analytics/triple-confirmed` Data sources: - `data/derived/live/deltas_player_hitter_.parquet` - `data/derived/live/regression_candidates_.parquet` - HR consensus artifacts under `outputs/betting/home_run_edges//` - TB consensus artifacts when three-lane coverage exists Computation: - Buy/sell rows require projection pace delta, surface-vs-underlying regression gap, and a three-lane aligned consensus signal. - Hitter buy/sell shipped first; pitcher version is deferred until K Valinor and pitcher consensus are production-ready. Sample output from latest full live snapshot (`2026-05-07`): | Player | Direction | Pace delta | Regression gap | Consensus | | --- | --- | ---: | ---: | --- | | Suarez, Eugenio | sell | `+16.8%` | `xwOBA-wOBA -0.109` | HR aligned below market, lane average `15.8%` vs market `22.0%` | Rendering description: - The page renders tiered buy/sell sections, stack chips for the three confirming signals, and honest empty states. - The May 12 forced task wrote a clean empty artifact because May 12 HR consensus outputs are not present on disk yet; this is expected until the full daily betting build advances. ## Feature 2: Consensus Disagreement View Surface: `/betting/consensus?view=split` Data sources: - Existing Phase 2 consensus loader and lane-contract columns from HR/K/TB/Hits/Outs/Futures outputs. Computation: - Adds `lane_dispersion = max(lane probabilities) - min(lane probabilities)` when all three lanes are present. - Adds high-lane and low-lane labels for row-level explanation. - Split mode sorts by highest dispersion first. Sample row: | Entity | Market | High lane | Low lane | Dispersion | Market fair | | --- | --- | --- | --- | ---: | ---: | | Lane Thomas | HR | Valinor `19.6%` | Palantir `4.1%` | `15.5 pts` | `12.6%` | Rendering description: - The default consensus view remains aligned picks. - Split mode explains the disagreement in one templated sentence: Valinor sees `19.6%`; Palantir sees `4.1%`; gap `15.5 pts`; market fair `12.6%`. ## Feature 3: Player Percentile Strips Surfaces: - `/players/` - `/projections/hitter/` - `/projections/pitcher/` Data sources: - `data/historical/statcast_hitter_seasonal/hitter_seasonal_2026.parquet` - `data/historical/statcast_pitcher_seasonal/pitcher_seasonal_2026.parquet` - `data/derived/projections/hitter_season_2026.parquet` - `data/derived/projections/pitcher_season_2026.parquet` Computation: - Actual percentile ranks are computed against the current-season population. - Projection ticks show where the preseason projection would rank inside the current population when a matching projection metric exists. - Negative metrics such as hitter K% and pitcher BB% are inverted so `100th = best in MLB`. Sample output: | Player | Metric | Actual | Percentile | Projection tick | | --- | --- | ---: | ---: | ---: | | Aaron Judge | Hard-hit% | `56.2%` | `97.0` | `97.6` | | Aaron Judge | Barrel% | `14.5%` | `99.5` | `98.4` | | Trent Grisham | Hard-hit% | `53.4%` | `94.6` | `64.4` | | Andrew Abbott | K% | `16.4%` | `16.3` | `40.6` | Public sanity check: - Aaron Judge's strip is directionally consistent with public Statcast-style references. [Baseball Scope](https://baseballscope.com/players/aaron-judge-592450.html) lists Judge around `57.6%` hard-hit, `94.2 mph` average exit velocity, and 96th-97th percentile ranges; Mithrandir's current pull shows `56.2%`, `94.1 mph`, and `97.0` hard-hit percentile. Rendering description: - The strip uses the Signal-to-Warn design ramp, never red/green. - Tooltip text should be expanded in a later polish pass to state `100th = best in MLB`, especially for inverted metrics. ## Feature 4: Projection Percentile Bands Surfaces: - `/projections/hitter-season` - `/projections/pitcher-season` - `/projections/team-season` - `/projections/standings` - `/projections/leaders` Data sources: - Existing p10/p50/p90 player intervals in projection parquets. - `data/derived/projections/team_wins_sim_2026.parquet` for empirical team projected-win quantiles. - Season projection JSON outputs for display rows that exist outside the richer parquet overlap. Computation: - New sidecar: `data/derived/live/projection_quantile_bands_.parquet`. - Existing p10/p50/p90 intervals are reused where available; p25/p75 are midpoint interpolation. - Team standings bands use empirical p10/p25/p50/p75/p90 from `team_wins_sim_2026`. - HR and K/9 display rows use conservative parametric display bands when richer source quantiles are unavailable. Sample output (`2026-05-12`): | Entity | Metric | p10 | p25 | p50 | p75 | p90 | | --- | --- | ---: | ---: | ---: | ---: | ---: | | Aaron Judge | HR | `35.4` | `40.4` | `46.0` | `51.6` | `56.6` | | Shohei Ohtani | HR | `36.2` | `41.3` | `47.0` | `52.7` | `57.8` | | Andrew McCutchen | HR | `7.0` | `8.4` | `10.0` | `11.6` | `13.0` | | LAD | Team-season wins | `86.4` | `90.8` | `95.6` | `100.4` | `104.8` | Rendering description: - Each row shows p10/p25/p50/p75/p90 in mono numerals with a faint CI-band style track beneath. - Tied source rows remain tied in the display. Collapsed bands are transparency, not a bug. ## Scheduler Tasks | Task | Cadence | Health signal | | --- | --- | --- | | `compute_triple_confirmed_signals` | Daily `05:40` | File exists; empty is valid because the query is intentionally high precision | | `compute_player_percentiles` | Daily `05:45` | Non-empty percentile output | | `compute_projection_quantile_bands` | Daily `05:50` | Non-empty quantile-band output | Forced targeted scheduler cycle on May 12: | Task | Status | Metric | | --- | --- | ---: | | `compute_triple_confirmed_signals` | success | `1.0` health metric, `0` current rows | | `compute_player_percentiles` | success | `4,725` rows | | `compute_projection_quantile_bands` | success | `5,548` rows | ## Verification Route checks: | Route | Result | | --- | --- | | `/analytics/triple-confirmed` | `200` | | `/betting/consensus` | `200` | | `/betting/consensus?view=split` | `200` | | `/players/592450` | `200` | | `/players/663757` | `200` | | `/projections/pitcher/671096` | `200` | | `/projections/hitter-season` | `200`, bands rendered | | `/projections/pitcher-season` | `200`, bands rendered | | `/projections/team-season` | `200`, bands rendered | | `/projections/standings` | `200`, bands rendered | | `/projections/leaders` | `200`, bands rendered | Tests: - `python -m unittest discover tests`: `57` tests passed. Chalk discipline: - Template grep still finds chalk only in `src/pitcher_card_engine/web/templates/daily.html`. Data sanity: - Top latest triple-confirmed row genuinely satisfies all three conditions: Suarez is above projected pace, has a negative xwOBA-wOBA gap, and has an HR three-lane aligned below-market signal. - Top split consensus row matches the explainer text: Lane Thomas has Valinor `19.6%`, Palantir `4.1%`, dispersion `15.5 pts`, market fair `12.6%`. - Judge percentile strip aligns with public Statcast-style values within normal refresh-window differences. - Judge/Ohtani/McCutchen HR bands and LAD team-season wins bands are monotone and plausible. ## Honest Caveats - WAR bands are not surfaced because there is no validated per-player WAR quantile source in the season projection parquets. - xwOBA is used where the prompt asked for wOBA because xwOBA is the validated hitter rate metric in the current projection stack. - Some projection leader rows have tied p10/p50/p90 from source; Wave 1 preserves those collapsed bands rather than inventing fake variance. - TB-side triple-confirmed is currently empty because TB does not have three-lane coverage yet. It will fill as Phase 3 expands lanes. - May 12 targeted scheduler outputs do not advance the global live `latest.json` pointer. Public routes still use the latest full live-state snapshot until the full live cycle advances. ## Carried Debts - Pitcher version of triple-confirmed is deferred to Wave 4 because it depends on K Valinor and pitcher consensus coverage. - Wave 2 calibration drift moat has not started. - Wave 4 K Valinor plus bat-tracking and ABS integrations remain blocked on K Valinor resumption. - Methods page write-ups remain stubs per user direction. - Mobile polish pass remains deferred per the integration audit recommendation. - Percentile-strip hover copy should explicitly say `100th = best in MLB`. ## Explicit Non-Goals - No new ML training. - No K-specific signals. - No calibration drift banner. - No methodology page content. ## Audit Delivery Check | Audit recommendation | Shipped? | | --- | --- | | Triple-confirmed signals | Yes | | Consensus disagreement view | Yes | | Player percentile strips | Yes | | Projection percentile bands | Yes | The audit identified triple-confirmed signals as the highest impact-to-effort Wave 1 feature. It is now live, with the first real signal class represented by the Suarez sell row.