Validate UHI_MODEL + THRESHOLD_MODE env vars; warn + fallback on typos

Simplify: reuse shared WBGT+percentile utils, batch DB writes, memoize reads

Surface per-zone thresholds on Vercel frontend

UHILSTCorrector: defer to synthetic UHICorrector for LST-uncovered zones

Zone-specific trigger thresholds behind THRESHOLD_MODE env var

Fix FEATURES_PATH resolution for Docker (COPY to /app) vs local

Add LST-anchored UHI corrector behind UHI_MODEL env var flag

Fix TOA accumulation window: 6h → 1h to match ERA5/GraphCast

pipeline: one-shot reconnect on DB stale connection

graphcast: handle static vars without a time dim (GFS path)

gfs: analytical TOA forcing + pipeline traceback logging

Fix GFS shortName regression: use ecCodes names, not cfgrib names

Add GFS init source as alternative to ERA5T (Phase 2 wiring)

Cut dead neural/LSTM pricing + predictor paths

NASA POWER ingest: drop all-missing days instead of inserting empty rows

predictions ON CONFLICT: overwrite xgb/lstm/confidence/ensemble too

Cut XGBoost/LSTM heat predictor from pipeline and demo

Refresh Neon connection at the start of each pipeline run

heat_indices + healed_readings ON CONFLICT: overwrite all derived fields

Fix real Claude cost bug: token double-count, revert Haiku swap

Pipeline hardening: Haiku, dead code cleanup, consec fix, ERA5-lag init

Prewarm GraphCast at FastAPI startup to avoid cold-start fallback

NASA POWER ingest: heal NULL fields, request precip, respect 2-day lag

Fix stale Neon connection on Space wake

Fix pipeline failures: api.py import, NASA POWER resilience, CPU torch

Clean up merge conflict markers in neural_actuarial.py (reference code)

Replace neural actuarial model with GraphCast triggers + empirical burn analysis

Coverage polish + DPI-style worker pilot with Disbursements drill-down

Make fork-and-adapt flow truly region-agnostic

Add AI trigger explanations behind EXPLANATION_MODE flag

Harden serverless DB checks, CORS, cursor lifecycle, and healing JSON parse

Fix $0.00 cost: pass total_cost_usd to finish_pipeline_run

Disbursements page: M-Pesa payouts replace Alerts

Two-tier neural triggers + funding decomposition (SEWA pilot structure)

Write neural trigger_prob + pricing to predictions table

Neural model now decides triggers AND payouts end-to-end

Neural compound triggers: model confirms risk + current conditions

Compound heat triggers + settlement-specific thresholds, retrained

Chronos-Bolt encoder, retrained pricer, 14-day pipeline, sync DB

Rewrite pipeline + DB to fully sync psycopg2 (matches Weather AI 2)

Skip DB writes in pipeline — psycopg2 blocks async event loop

Add print/flush logging to ingestion for HF debugging

Add flush=True print statements for pipeline logging

Fix pipeline: non-blocking DB startup, fix date.month crash in downscale

Remove blocking reconnect before pipeline run

Switch from asyncpg to psycopg2 for Neon DB (same as Weather AI 2)

Fix _pool references in CRUD functions for poolless connection manager

Replace asyncpg pool with fresh connections per query

Restructure pipeline: rule-based heal, template alerts, one Claude review

Cut ERA5-Land from live pipeline, fix zone count to 15 Dar zones