feat(rid): screen one-hour spikes; refetch and redo-from to repair the store

RID's report sometimes carries a wrong value for one hour and corrects it
later. P.1 read 10.6 m at 2026-10-25 00:00 between 1.04 and 1.07 (RID now
says 1.06); the store kept it, because a report is re-fetched only while it
has holes, and the model issued a 10.8 m forecast for that hour.

- rid::judge / rid::screen: a reading that leaves the previous one (within
  3 h) by more than 1 m (2 m on the flashy tributaries P.76, P.77, P.81,
  P.82, P.84, P.85, P.87) while the next reading comes back near the
  previous level is a spike: not stored, a stored copy is deleted. Limbs
  and steps the next reading confirms are kept; the newest reading waits
  for the next one when it jumps more than twice the limit. On 2018-2026
  history: 16 spikes found (slipped decimals, sign flips, -5/-6/-9 m
  placeholders), no genuine change rejected.
- Every store goes through it (rid::store: run_cycle, fill-gaps, refetch).
- `ping-collector refetch FROM [TO]`: re-fetch reports regardless of holes
  (corrects revised readings, fills the discharge of level-only evening
  hours, removes stored spikes) and print the catch-up command when
  readings changed.
- `catch-up-forecasts --redo-from T`: delete the model forecasts issued
  from T on, then recompute them.
- Tests: judge/screen unit tests; a DB test of refetch, the live hold and
  the forecast deletion. Checked end to end against RID on a scratch DB.
- Docs: DATA_SOURCES (spikes, level-only hours), FLOOD_FORECASTING,
  rust/README.

docs/DATA_SOURCES.md

@@ -23,10 +23,29 @@ The only source persisted to the database and used by the ML pipeline.
 | Params | `DW[UtokID]=1`, `DW[BasinID]=6` (Ping), `DW[TimeCurrent]=<Buddhist-calendar date>`, `rows=100` |
 | Variables | Water level (m, gauge datum), discharge (m³/s, `'***'` = malformed), discharge % of channel capacity |
 | Stations | 16 P-series gauges (P.1 anchor at Nawarat Bridge; see `src/data/stations.json`) |
-| Client | `rust/collector/src/rid.rs` (`ping-collector`; `fill-gaps` re-fetches old reports) |
+| Client | `rust/collector/src/rid.rs` (`ping-collector`; `fill-gaps` re-fetches old reports with holes, `refetch FROM [TO]` any reports) |
 
 Human-facing page: <https://hyd-app-db.rid.go.th/hydro1h.html>
 
+**Spikes.** The report sometimes carries a wrong value for one hour and corrects
+it later: a slipped decimal point (P.1 10.6 m between 1.04 and 1.07 on 2026-10-25
+00:00, later 1.06; P.85 0.101 for 1.1), a flipped sign (P.4A ±1.18, P.67 ±0.6) or
+a placeholder (−5, −6, −9, 50 m). The collector re-fetches a report only while it
+has holes, so a later correction never arrived. Since 2026-10-26 every batch is
+screened before it is stored (`rid::judge`, `rid::screen`): a reading that leaves
+the previous one (within 3 h) by more than 1 m (2 m on the flashy tributaries
+P.76, P.77, P.81, P.82, P.84, P.85, P.87) while the next reading comes back near
+the previous level is not stored, and a stored copy is deleted. A rising or falling
+limb, or a step the next reading confirms, is kept. The newest reading of a gauge
+waits for the next one when it jumps more than twice the limit. On the 2018–2026
+history the rule finds 16 spikes and would have rejected no genuine change (the
+largest genuine 1-h changes: 1.1 m on the mainstem, 3.1 m at P.76). Two wrong hours
+in a row are not caught.
+
+Evening hours often arrive level-only (no discharge) and gain the discharge
+later; the API's history fills a missing discharge with 20 × level (an old
+placeholder, kept for compatibility). `refetch` brings in RID's discharge.
+
 ---
 
 ## 2. ThaiWater / HII ecosystem

docs/FLOOD_FORECASTING.md

@@ -643,7 +643,10 @@ Open-Meteo's estimate of rain that has since fallen, as in training, where
 the forward rain is observed rain. Such rows are written only where no model
 row exists; `created_at` tells them apart (more than an hour after `as_of`).
 `ping-collector catch-up-forecasts --hours N` does the same further back (up
-to 60 days).
+to 60 days); with `--redo-from T` it first deletes the model forecasts issued
+from T on, to recompute them after `ping-collector refetch` corrected the
+readings they came from (a bad reading reaches the features for 72 h through
+`roll_max_72`).
 
 A model row in `forecast_history` is never replaced by a heuristic one. While
 RID is silent overnight each hourly run re-issues the 21:00 reading, and once