The dashboard is a reader, not a model. This page states the frozen definitions, the thresholds, the uncertainty machinery and the known limits, so any figure can be traced back to the research pack.
SupeRR Scout is a read-only surface over a frozen research run. The Rust API loads CSV artefacts produced by the analytical pipeline and serves them without changing a score, a rank or a threshold. The Next.js frontend formats values and draws charts, but it never recomputes a ranking.
The single allowed client-side arithmetic is display aggregation: turning stored high-pressure and baseline counters into readable rates, and mapping values into chart geometry. Those derived values are labelled wherever they appear.
Four refusals hold everywhere in this dashboard.
Pressure is a pre-delivery index in the range 0 to 100. It combines three context components, weighted by innings type. The outcome of the ball is never an input.
pressure_index = 100 * ( w_run_rate * c_run_rate + w_wickets * c_wickets + w_resource * c_resource ) first innings weights 0.40 / 0.35 / 0.25 chase weights 0.45 / 0.30 / 0.25
High pressure is an index of 70 or more. Medium starts at 40. The thresholds are provisional and reviewed in the model config, not in this app.
The ranking converts pressure performance into a score using winsorised effect sizes against each player's own baseline, then applies reliability shrinkage so that a small sample cannot dominate the top of the table.
score = f( z(primary_delta), z(dot_rate_delta), z(boundary_rate_delta), z(dismissal_rate_delta), z(wicket_rate_delta), z(extras_rate_delta) ) * reliability
Only players who clear the eligibility floor enter the pool. IPL currently holds 100 eligible batters and 85 eligible bowlers. T20 international and combined pools are separate and never blended into IPL.
Every ranking row carries a score interval, a rank interval and a top-10 probability from the bootstrap and leave-one-match-out stability run. A wide rank interval means reordering is plausible, not that the player is worse.
The dashboard shows the interval beside the point estimate everywhere it matters. Filtering by sample size is not cosmetic; it is the fastest way to see how much of a rank is carried by a handful of innings.
Moving the high-pressure line is the largest single source of definitional drift. The desk re-ran the model at six lines and compared each top ten with the one at 70, the frozen baseline. Only the baseline matches itself, and no alternative run keeps more than six of ten names in either discipline. That is why the line is reported as provisional rather than settled.
| High line | Batting | Bowling | Reading |
|---|---|---|---|
| 55 | 3 of 10 | 3 of 10 | Many more balls qualify, ordering shifts hard |
| 60 | 4 of 10 | 3 of 10 | More balls qualify, ordering shifts |
| 65 | 5 of 10 | 6 of 10 | Close to the baseline, still not stable |
| 70 | 10 of 10 | 10 of 10 | Frozen baseline used by this desk |
| 75 | 5 of 10 | 5 of 10 | Fewer balls qualify, ordering shifts |
| 80 | 5 of 10 | 5 of 10 | Fewest balls qualify, ordering shifts |
Overlap counts came from the stored sensitivity runs under the research pack, not from this page. The dashboard reads only the 70 run.
Matches without usable ball data are excluded from delivery-level modelling but stay in the match register. Abandoned and incomplete matches are flagged rather than silently dropped. Super overs are modelled separately and never folded into regulation innings totals.
Players below the eligibility floor are kept in the canonical directory but are not ranked. The rankings page only ever shows the eligible pool.
| Model | Identifier |
|---|---|
| metrics | metrics/v1.0.0 |
| pressure | pressure/v1.0.0 |
| ranking | ranking/v1.0.0 |
| scouting | scouting/v1.0.0 |
Every value on this desk can be reproduced from the repository.
# 1. Refresh the frozen artefacts (writes data/rust) python pipeline/run_all.py # 2. Serve the read-only API on 127.0.0.1:8080 cargo run -p scout-api --manifest-path engine/Cargo.toml # 3. Serve the dashboard cd app/web && npm run build && npm start # 4. Verify parity and contracts cargo test -p scout-api cd app/web && npm test
The API also exposes the frozen ranking CSV and the shortlist scorecards for direct download, so a reader can diff the dashboard against the source rows.
Pressure is descriptive, not causal. It says a delivery was high-leverage by the model definition; it does not prove a player "handles pressure". Auction facts for 2027 are unknown, and this desk makes no bid, price, retention or availability claim.
The research pack is ratified by named human sign-off. Missing 2027 auction facts remain unknown and still block any bid recommendation.
Research and educational use only. Not investment, retention or auction advice.