downstream docs: user-ready sidecar + fallback_rate key + precompute note

downstream/README.md

@@ -9,7 +9,7 @@ downstream/<method>.meta.json       # display + diagnostic sidecar
 ```
 
 See `SCHEMA.md` for the exact column / field schema (including the
-`overall_fallback_rate` diagnostic), and `bootstrap/` for the per-draw CI reference.
+`fallback_rate` diagnostic), and `bootstrap/` for the per-draw CI reference.
 
 Unlike Tracks 2/3 (per-user MAE), Track-1's headline metrics — binary AUPRC, ordinal
 Spearman, regression Pearson — are cohort-level ranking / correlation metrics that do
@@ -21,10 +21,11 @@ server-side against the `linear` baseline.
 
 The substrate parquets are the canonical inputs for:
 
-- The OpenMHC HF Space (`MyHeartCounts/OpenMHC`) — it downloads these parquets + the
-  sidecars and runs the downstream reducers
-  (`src/downstream_evaluation/evaluation/bootstrap_skill_rank.py`) to produce the live
-  leaderboard table: skill / fair-skill / mean-rank vs the `linear` baseline.
+- The OpenMHC HF Space (`MyHeartCounts/OpenMHC`) — the live leaderboard table (skill /
+  fair-skill / mean-rank vs the `linear` baseline). Track-1's headline scores are
+  paired-bootstrap means, too heavy to reduce on each page load, so the maintainers reduce
+  these substrates offline and publish the per-method rows as `downstream/leaderboard_rows.json`,
+  which the Space reads directly.
 - Independent re-aggregation (the reducers in `scripts/paper_results/downstream/`).
 - The cluster-bootstrap reference at `downstream/bootstrap/` (per-draw CIs) is reduced
   from these substrates, so any change here must be matched by a bootstrap refresh.
@@ -43,14 +44,14 @@ parquet = hf_hub_download(
 df = pd.read_parquet(parquet)
 print(df.shape, df.columns.tolist())
 
-# Display + diagnostic sidecar (incl. overall_fallback_rate)
+# Display + diagnostic sidecar (incl. fallback_rate)
 meta_p = hf_hub_download(
     "MyHeartCounts/OpenMHC-leaderboard-data",
     "downstream/xgboost.meta.json",
     repo_type="dataset",
 )
 print(json.loads(open(meta_p).read()))
-# -> {"display_name": "XGBoost", "type": "Statistical", ..., "overall_fallback_rate": 0.0}
+# -> {"display_name": "XGBoost", "type": "Statistical", ..., "fallback_rate": 0.0}
 ```
 
 ## Pooled substrate
@@ -66,9 +67,9 @@ pooled = pd.concat(
 )
 ```
 
-## `overall_fallback_rate`
+## `fallback_rate`
 
-Each sidecar carries `overall_fallback_rate` (issue #39) — the fraction of the method's
+Each sidecar carries `fallback_rate` — the fraction of the method's
 test predictions the harness left non-finite and substituted with the `linear` baseline
 before scoring. `wbm` is the only non-zero method (it embeds only participants with a
 full weekly window); the rest are `0.0`. A high rate means the headline scores partly