Read discharge from wet in classify and the habitat export, and re-tune BT's MAD range on it #331

If done: classify, the validator, the scoring harness and the habitat export (#308) all read discharge from wet's published discharge.fwa_v1_segment_monthly (month = 0, pick), and the export's manifest records which wet build the run read (the md5s of its parts wait on NewGraphEnvironment/wet#88, which records them at load), so habitat → discharge provenance is one chain of manifests. BT's MAD range is re-tuned on wet's values the way #302 tuned it on fwapg's, so whether to move any group to mad is decided on the source it would run on. If never: link keeps reading whse_basemapping.fwa_stream_networks_discharge, whose version is only what provenance.source_artifact happened to record on the exporting host, and default_tuned's BT 0.078 m³/s stays a threshold fitted to a source we mean to leave.

Split out of #308 (operator, 2026-10-14). The export reads the discharge the run classified on, so *_ratio_mad agrees with the flags; moving the source changes outputs, so it is its own change.

• One subquery already serves every consumer: .lnk_discharge_sql() (R/lnk_discharge.R), table named by .lnk_hv_discharge_tbl(). The fill (Main stems with no discharge drop out of mad habitat #305) is layered on it.
• wet's loader is wet_db_load() (Publish per-segment discharge: parquet + manifest + fwapg load for fresh#114 wet#7); it writes discharge.fwa_v1_segment_monthly with a partial index (scenario, linear_feature_id) WHERE month = 0 AND pick. Nothing needs building: the parts are published at s3://fresh-bc/discharge/fwa/v1/, so the first step on any database is wet_db_load(conn, wet_catalog()) (43 parts, 103 M rows, 1.1 GB for historical).
• Every host that runs the pipeline needs the table once classify reads it: the multi-host dispatcher's prep and the vintage gate (lnk_preflight_vintage(), snapshot_bcfp.sh) should load or check it, and provenance.source_artifact should record the manifest it came from. Measured 2026-10-15 on a local fwapg: a full historical load took 15.5 min, 102,790,688 rows, 20 GB with indexes; 4,532,843 lines carry an annual pick (3,152,838 pcic_routed, 1,380,005 wet_balance, none twice; 29,491 of them 0 m³/s) against 4,907,441 FWA lines and 2,716,652 rows in fwapg's discharge table (2,003,189 with a value).
• Decide: whether the Main stems with no discharge drop out of mad habitat #305 fill still applies to wet's values, and what to log per run. Decided: the fill applies unchanged, and on wet it reaches almost nothing (edge 1250 km with no value in Tune MAD (discharge) thresholds for BT, GR, KO and RB rather than leave them missing #302's BT groups: 506 on fwapg, 5 on wet). Each run logs discharge_source per WSG, a wet run's log_input carries wet's table and its comment (the wet build), and the export manifest names the source per WSG plus a discharge.wet_load block. Per-part md5s cannot be recorded until wet_db_load() records them: wet_db_load(): record which parts were loaded (key, md5) in the database wet#88.
• Decide: switch the source outright, or make it a bundle setting. Decided: a bundle setting, pipeline: discharge_source: fwapg | wet, default fwapg. Only default_tuned reads wet (operator); wet's 0 m³/s reads as no value (operator).

Re-tune BT discharge on wet's values (operator, 2026-10-15)

The same exercise #284 / #302 ran for gradient, width and discharge, repeated on the new source before any group moves to mad:

• Unblock the scoring harness at HEAD. A set cell the base already carries is now dropped with a message; a rung whose own cell equals the base still stops the build. data-raw/habitat_variants_build.R / _score.R stop at any HEAD after default MAD ranges for BT, GR, KO and RB, from its own width minima #307: the variant bundles extends: default by name and default's thresholds moved, so the guard reports "variant … equals default" (habitat_variants_build.R:311): Tune MAD (discharge) thresholds for BT, GR, KO and RB rather than leave them missing #302's rungs set *_mad_max = 9999, which default has carried since default MAD ranges for BT, GR, KO and RB, from its own width minima #307. Vignette: default against default_tuned for bull trout in the Parsnip, side by side #342 needs the same fix; do it once.
• Recompute default's width-converted MAD minimums on wet. Measured, not landed: the conversion barely moves (BT spawning 2 m: 0.043 on wet, 0.042 on fwapg; rearing 1.5 m: 0.023 and 0.021), and default stays on fwapg (operator: default_tuned only). default's 0.041 (spawn, 2 m), 0.20 (GR spawn, 4 m) and 0.021 m³/s (rear, 1.5 m) are the median fwapg mad_m3s at each width minimum (default MAD ranges for BT, GR, KO and RB, from its own width minima #307), so they move with the source. Producer: data-raw/query_width_mad_equivalent.R, with its source swapped.
• Re-run Tune MAD (discharge) thresholds for BT, GR, KO and RB rather than leave them missing #302's BT MAD ladder on wet's discharge: same fixed rule (P10, P05 loosening refused unless it scores), same held-out WSGs, same species pooling. Report whether BT's 0.078 m³/s spawn / rear minimum moves. It moves to 0.043 / 0.043. Rearing walks to its wet P10 (P10 → P05 refused). Spawning could not be scored on spawning-staged records (3 in the held-out core) and takes the same any-stage value (operator). The values move because wet covers nine Skeena groups fwapg does not; on locations both value, the sources agree.
• Re-run Score the discharge (mad) habitat model against observations, as #284 did for gradient #300's cw-against-mad score for BT (with the Main stems with no discharge drop out of mad habitat #305 fill; without it was not run, since the fill reaches 5 km on wet), size-adjusted, on the same held-out WSGs. Compare with Score the discharge (mad) habitat model against observations, as #284 did for gradient #300 / Main stems with no discharge drop out of mad habitat #305: BT rearing cw-only 0.19 after the fill, mad −32 % BT stream rearing against cw. At 0.043 on wet mad is the closer model: cw-only rearing used at 0.18 of the size-matched core, mad-only at 1.64; mad keeps −19.8 % BT rearing against cw (fwapg at the same thresholds −21.3 %).
• Write the result into research/habitat_thresholds.md beside the fwapg sections, and land any changed value in default_tuned only.

Out of scope: a spawning-season discharge rule from wet's monthly values. That is a new kind of rule, not a re-tune; file it separately if this result makes it worth doing.

GR, KO and RB on wet: #343.

Relates to #308, #305, #302, #300, #342, NewGraphEnvironment/wet#7.