docs: source sweep — P.75 already is the Mae Ngat release signal

Verified from a Thai ISP that lsim.rid.go.th is unreachable (not
geo-blocked), then swept for any better-than-daily Mae Ngat source.

Key finding: P.75 sits 3.8 km below the dam, reports hourly, and has
been a model feature since v1 — the model has always read the dam's
actual outflow, hourly and directly. That is the likelier reason the
daily reservoir table adds nothing, beyond its publication lag.

Intraday reservoir feeds do exist and are open (bigdata-api.rid.go.th
SWOC, ThaiWater ridhydro_TUP.16 at the dam) but are snapshot-only with
no archive, so they cannot retrain against past events; the HII
collector accumulates them from 2026-08-11 for a post-monsoon revisit.
Also documents api/dam (whole per-dam range in one request, vs the
~2,900-request per-day loop) and several cross-check mirrors.

docs/FLOOD_FORECASTING.md

@@ -509,10 +509,28 @@ event-peak amplitude (−0.11 → −0.03 m on the Sep 2024 event), and `rise_da
 opt-in via `--dam` on the training CLI, `scripts/backtest_render.py --dam`,
 and the `rise_rain_dam` / `rise_dam` harness variants). The collector keeps
 accruing daily rows; revisit post-monsoon when the 2026 season adds dam-era
-flood events. Removing the publication-lag objection needs an *intraday*
-reservoir source; the obvious candidate, `lsim.rid.go.th`, is unreachable
-even from a Thai consumer ISP (see `docs/DATA_SOURCES.md`), so that remains
-an open search.
+flood events.
+
+**Why the lag is probably not the whole story — P.75 already *is* the dam
+signal.** A 2026-08-13 source sweep put the negative result on firmer
+ground: **P.75 "บ้านช่อแล" sits 3.8 km downstream of the Mae Ngat dam** on
+the Mae Ngat river (nearest other station: P.4A at 10.9 km), it reports
+hourly, and it has been a model input since v1 with a 12 h routed lead into
+P.1. Whatever the reservoir releases flows past P.75 within the hour and the
+model already reads it. The daily reservoir table therefore offers a stale,
+coarser proxy of a signal the features capture hourly and directly — which
+is the more likely reason it adds nothing and costs alarm responsiveness.
+
+That reframes what a future intraday source would have to beat: not "no dam
+information", but "hourly observed dam *outflow*". Genuine intraday
+reservoir-state feeds do exist and are open (`bigdata-api.rid.go.th` SWOC
+telemetry, and ThaiWater station `ridhydro_TUP.16` *at the dam*), but both
+are **snapshot-only — no archive** (verified: the history endpoint returns
+empty grids for them at every era, including the current one). They can only
+be accumulated forward, so they cannot retrain against 2024/2025 events.
+The HII collector already captures both hourly as of 2026-08-11; revisit
+after the 2026 monsoon, when a season of true intraday reservoir state
+exists alongside its flood events.
 
 **Shipped from the same work:** the HII gap-fill merge in the data loader
 (`fill_from_hii`, +9,341 h at P.81, +682 h at P.92, +810 h at P.20) is