docs: scrub dangling references to removed paid-API tooling

Fix doc/comment/message references left after the SportMonks removal so the
public repo has zero broken or mangled file paths:
- relevance.py / OPPONENT_ADJUSTMENT.md / experiments_log.json: drop deleted
  file paths (LOO harness + corpus loader live in git history), keep the writeup
- group_scenarios.py + DAILY.md: point at the pre-fetched data/standings/ snapshots
- DATA_STRATEGY / NEW_QUESTION_TYPES / PARSIMONY: genericize provider name + paths
- untrack the orphaned paid-API enrichment runbook (kept locally, gitignored)

No code paths change; 186 tests pass, validators 104/48/52 green.

Author: DataAthleteChamp

.gitignore

@@ -31,3 +31,6 @@ predictions/baselines/**/_fire_map.json
 
 # raw screenshot dump (curated subset lives in docs/results/)
 docs/results-screnshots/
+
+# Orphaned paid-API enrichment runbook (kept locally, not part of the public repo)
+workflows/DATA_ENRICHMENT.md

bot/data/relevance.py

@@ -23,8 +23,9 @@
 Opponent strength is a *corpus-internal* rating (mean goal difference per team), so it covers every
 team in the history corpus without an external Elo table and is internally consistent.
 
-This module is pure/estimation-only; the LOO validation harness lives in
-``scripts/relevance_backtest.py`` and the eventual wiring point is ``StatsJsonProvider``.
+This module is pure/estimation-only; the leave-one-out validation harness and corpus
+loader were removed from the public repo (kept in git history), and the eventual wiring
+point is ``StatsJsonProvider``.
 """
 
 from __future__ import annotations
@@ -120,7 +121,7 @@ def team_strengths(matches: list[Match], stat: str = "goals") -> dict[str, float
 
     A simple, leakage-stable team attribute used only as the *opponent-similarity variable* (which
     past opponents resemble the current one). ``stat`` selects the differential: "goals" (history,
-    where goal-difference is available) or "shots_on_target" (the stats-api corpus has no goals, so
+    where goal-difference is available) or "shots_on_target" (this corpus has no goals, so
     SoT-differential is the available dominance proxy). Z-scoring puts sigma on a ~unit scale.
     """
     diff: dict[str, list[float]] = defaultdict(list)

research/DATA_STRATEGY.md

@@ -446,17 +446,17 @@ WC/Euro 2024 support with `--season` flag]
 
 🟡 [INFERRED from general knowledge and product page structure; not directly verified]
 
-#### 4.2.2 the stats API Football API
+#### 4.2.2 A paid football-stats API
 
 | Property | Detail |
 |---|---|
-| **URL** | `stats-api.com/football-api/` |
+| **URL** | (paid provider) |
 | **What it provides** | "Statistics from professional football clubs, media agencies, and gaming companies... goals, average minute teams/players need to score, average goals per minute per team, trends" |
 | **Free tier** | 14-day trial or limited free plan |
 | **Reachability** | Landing page accessible |
 | **Useful for** | Broader league coverage; international stats including WC qualifiers |
 
-🔵 [stats-api.com/football-api — confirmed accessible; feature list directly verified]
+🔵 [paid provider — confirmed accessible; feature list directly verified]
 
 #### 4.2.3 Opta / StatsPerform
 

research/NEW_QUESTION_TYPES.md

@@ -19,7 +19,7 @@ model RBP +1020.8 unchanged, 187 tests pass — the handlers fire ONLY on the no
 `player A > player B SoT` and single-player SoT props are only as sharp as our player rates. The
 `BaseRateProvider` returns a GENERIC default (sot90≈0.60) for players without specific data, so e.g.
 Yamal and Messi both get λ≈0.47 → P(Yamal>Messi)≈coin-minus-ties≈0.28. **To price these sharply we
-need per-player SoT/90 rates** (fetch via `scripts/fetch_stats-api_player_sot.py` into `data/`), else
+need per-player SoT/90 rates** (fetch per-player SoT/90 rates into `data/`), else
 apply a manual volume tilt at pricing time (Yamal is a higher-volume shooter than Messi's deep role).
 
 ## "Odds can mislead" (apply at pricing)
@@ -41,7 +41,7 @@ the model where it has a real data basis and the market is thin.
 H2H and brace (Poisson tail). 2026-07-18.*
 
 ## Update — player-SoT data fetched & wired (2026-07-18)
-Ran `scripts/fetch_stats-api_player_sot.py --write` → `data/players/stats-api_sot.json` (157 players,
+Fetched per-player SoT/90 → `data/players/player_sot_rates.json` (157 players,
 leak-free, WC-2026 season excluded). This is **auto-wired** via `bot/data/teamstats.py` (blends
 `sot_per90` into each player rate). Finalists now carry REAL rates: **Messi 1.66, Mbappé 2.03, Kane
 1.88, Oyarzabal 2.42, Olmo 1.51, Álvarez 0.96** (vs the 0.60 generic default). Backtest-gated:

research/OPPONENT_ADJUSTMENT.md

@@ -17,7 +17,8 @@ OTHER matches — weighting each past match by how relevant it is to *this* matc
 - `w_blowout` down-weight ≥5-goal blowouts.
 - `k`       pseudo-matches of shrinkage toward the league prior μ0.
 
-Code: `bot/data/relevance.py` (estimator) · `scripts/relevance_backtest.py` (leave-one-out harness).
+Code: `bot/data/relevance.py` (estimator). The leave-one-out harness and corpus loader were
+removed from the public repo (kept in git history).
 
 ## Validation (history LOO, 526 team-matches, 2,104 territorial outcomes)
 
@@ -32,9 +33,8 @@ Scored with **Brier** (binary markets) + **RPS** (3-way match outcome; Constanti
 
 ## Cross-validation on the WC2026 teams' OWN data (7,088 territorial outcomes)
 
-Built a per-match corpus for all 48 WC2026 teams from **cached** stats-api internationals
-(`data/stats-api/between/*.json`, 2024-25 qualifiers/friendlies) — `bot/data/match_corpus.py`.
-Coverage: **all 48 teams**, median 21 matches/team (min 9, max 35). The method generalises:
+Built a per-match corpus for all 48 WC2026 teams from **cached** internationals
+(2024-25 qualifiers/friendlies). Coverage: **all 48 teams**, median 21 matches/team (min 9, max 35). The method generalises:
 
 | Config | terr Brier (WC corpus) |
 |---|---|

research/PARSIMONY_AND_DATASETS.md

@@ -23,7 +23,7 @@ highest-ROI signals make it significantly better (less overfitting)?
 |---|---|---|---|
 | 1 | **Sharp market odds** | ★★★★★ (dominates liquid markets) | anchored (_ODDS_WEIGHT=4) |
 | 2 | **Elo / SPI / pi-ratings** | ★★★★☆ | have (elo + margin-elo) |
-| 3 | **xG / shot quality** | ★★★☆☆ (counts; can add noise on sparse intl) | have (stats-api xG) |
+| 3 | **xG / shot quality** | ★★★☆☆ (counts; can add noise on sparse intl) | have (paid-API xG) |
 | 4 | **Shots-on-target / territorial** | ★★★☆☆ (dir-acc 0.80) | have |
 | 5 | **Corners** | ★★★☆☆ (0.79) | have |
 | – | Offsides totals (ref/base-rate) | ★★☆☆☆ | have |

results/experiments_log.json

@@ -344,7 +344,7 @@
     {
       "name": "exp/relevance-weighted-rates (E0-E5)",
       "date": "2026-06-26",
-      "change": "per-fixture opponent-aligned rate estimator: opp-strength Gaussian kernel + Gamma-Poisson shrinkage + (optional) time-decay/competition/blowout weights. Validated via scripts/relevance_backtest.py LOO on data/history (526 team-matches, 2104 territorial outcomes).",
+      "change": "per-fixture opponent-aligned rate estimator: opp-strength Gaussian kernel + Gamma-Poisson shrinkage + (optional) time-decay/competition/blowout weights. Validated via leave-one-out on data/history (526 team-matches, 2104 territorial outcomes).",
       "metric": "territorial_Brier + 3-way RPS (history LOO, NOT the WC2026 gate yet)",
       "results": {
         "flat_baseline": {
@@ -364,7 +364,7 @@
       "findings": "k (shrinkage) is the dominant lever: k=0 is WORSE than flat (+0.0074), k=5 best (-0.0120). Opp-strength kernel helps modestly on territorial and more on RPS (sigma~0.8-1.2). For COUNT stats time-decay HURTS (xi=0 best) and blowout-downweight HURTS (blowout games carry real territorial signal) - opposite of the goals/results literature. Competition weighting ~no-op on the all-major-tournament history corpus (kept as the live friendly/qualifier hook).",
       "verdict": "VALIDATED on history LOO; NOT yet deployed - needs per-match data for WC2026 teams to wire into StatsJsonProvider, then the WC2026 gate.",
       "cross_validation_wc_corpus": {
-        "corpus": "bot/data/match_corpus.py (cached stats-api internationals, all 48 WC teams, median 21 matches/team, 7088 territorial outcomes)",
+        "corpus": "cached internationals corpus (all 48 WC teams, all 48 WC teams, median 21 matches/team, 7088 territorial outcomes)",
         "flat_terr_brier": 0.2025,
         "best_terr_brier": 0.19,
         "best_config": "sigma=1.0,k=0",
@@ -434,8 +434,8 @@
       "verdict": "PASS-NEUTRAL (handlers fire only on novel types; 0 regression on 621-question backtest; 187 tests pass). Forward-test on final/3rd-place to calibrate."
     },
     {
-      "name": "data/stats-api-player-sot-refresh",
-      "change": "Fetched leak-free per-player SoT/90 for 157 players via fetch_stats-api_player_sot.py --write (WC-2026 season excluded). Auto-wired through bot/data/teamstats.py (blends stats-api sot_per90 into player rates). Finalists now carry real rates (Messi 1.66, Mbappe 2.03, Kane 1.88, Oyarzabal 2.42, Alvarez 0.96) vs the 0.60 generic default.",
+      "name": "player-sot-rate-refresh",
+      "change": "Fetched leak-free per-player SoT/90 for 157 players from a paid stats API (WC-2026 season excluded), later removed from the public repo. Blended sot_per90 into player rates via bot/data/teamstats.py. Finalists carried real rates (Messi 1.66, Mbappe 2.03, Kane 1.88, Oyarzabal 2.42, Alvarez 0.96) vs the 0.60 generic default.",
       "category": "player_prop / SoT data",
       "rbp_before": 1020.84,
       "rbp_after": 1039.08,

scripts/group_scenarios.py

@@ -29,7 +29,7 @@
 def _latest_snapshot() -> Path:
     files = sorted(glob.glob(str(REPO / "data" / "standings" / "*.json")))
     if not files:
-        raise SystemExit("no standings snapshot; run scripts/fetch_standings.py first")
+        raise SystemExit("no standings snapshot found in data/standings/")
     # prefer a -live file for today if present
     return Path(files[-1])
 

workflows/DAILY.md

@@ -57,9 +57,9 @@ When the user reports a settled result, log it: write `results/games/<NNN>.json`
 3. It will **not** submit unless you tell it to.
 
 ## Group standings & motivation (Matchday 3+)
-From MD3, group games kick off in pairs. Pull the table and qualification scenarios:
+From MD3, group games kick off in pairs. Pull the qualification scenarios from the
+pre-fetched standings snapshots in `data/standings/`:
 ```bash
-python scripts/fetch_standings.py            # data/standings/<date>.json (add --live mid-matchday)
 python scripts/group_scenarios.py            # per-team clinched / must_win / eliminated / alive
 ```
 **Use it as a DECISION-SUPPORT aid, NOT a model tilt.** Empirical test (526-match history) found the