3 commits

Author	SHA1	Message	Date
Graham McIntire	2fd88a94ea	fix(propagation): 7 pipeline bugs + validation harnesses + model improvements ... Phase A — 7 pipeline bugs: - retain_scores_window: fix timeline self-deletion (NOTIFY payload run_time|valid_time) - Rust/Elixir weight divergence: Rust loads band_weights.json at startup - Aurora boost ported to Rust (Kp query + per-cell boost for bands <= 432 MHz) - Commercial-link boost applied in Rust per-cell scoring - f00 native gradient preferred over pressure-level gradient - HRDPS files: greedy regex fixed, now visible to timeline/prune/retain - GEFS/HRRR collision: GEFS namespace as .gefs.prop, merge at read Phase B — Validation harness: - scripts/validate_algo.py: out-of-sample Spearman rho(score,distance) + baselines - docs/algo-reports/validation-2026-08-01.{json,md} Phase C — Forecast-skill evaluation: - scripts/validate_forecast.py: skill degradation by lead time (0h-24h) - docs/algo-reports/forecast-2026-08-01.{json,md} Phase D — Calibration + model improvements: - recalibrate.py: validation gate before weight deploy - Recalibrator: Nx.max(0.0) replaces Nx.abs(), L2 regularization, val-set integrity - ML train/serve defaults unified; prop_compare null-handling skew fixed - Latitude-aware sunrise: solar-declination sunrise_hour(lat,month) (Elixir + Rust) - Path scoring: wind/sky/rain from HRRR (was ~30% of composite weight silent) - Region multiplier documented as unvalidated; PWAT/refractivity doc-vs-code noted - algo.md: synced scoring sections, marked retired features, D5/D6 changelog - Credo: path_compute cyclomatic complexity bumped (6->10 fields) — informational only	2026-08-01 19:28:41 -05:00
Graham McIntire	c66af5bf64	fix(recalibrate): correct stale "documentation only" report text, map 9cm band ... Two defects surfaced by the 2026-08-01 production run. The generated report's "Applying these weights" section claimed the JSON was inert until someone wired it into BandConfig. That stopped being true when BandWeights landed: BandConfig.weights/1 consults BandWeights.lookup/1 ahead of the in-source :weights override and the global @weights default, so band_weights.json is a live scoring input. Anyone reading the report would have concluded a recalibration run was a no-op. Replace the paragraph and document the operational half that was missing entirely -- BandWeights caches into :persistent_term on first read, so a running node serves the previous weights until it is redeployed or BandWeights.reset/0 is called. PSKR_BAND_TO_MHZ was missing 9cm, so every run logged "skipping unknown PSKR band" and dropped those rows from the corpus accounting. Weight derivation is unaffected -- 3400 MHz is not in PSKR_BANDS, so it still routes to the contacts fit (source=contacts, n=334, unchanged). The dict already carries other non-routed bands (23cm, 13cm, 3cm); 9cm was simply absent.	2026-08-01 16:34:56 -05:00
Graham McIntire	daafa5a02a	perf(algo): unify recalibration into single Python pipeline with JSON output ... Replace the two-script Python pipeline (analysis report + Elixir-source emitter) with a single `scripts/recalibrate.py` that fits per-band weights from PSKR spot density (VHF/UHF) and contacts↔HRRR correlations (microwave), writing `priv/algo/band_weights.json` as a machine-readable artifact. A new Elixir BandWeights module loads this JSON once via `:persistent_term` cache; BandConfig.weights/1 consults it before falling back to in-source overrides or global defaults. The script never touches Elixir source — recalibration is now `python3 scripts/recalibrate.py` followed by an app restart.	2026-05-25 14:45:55 -05:00