MAD (discharge) ranges for BT, GR, KO and RB in default_tuned, scored on held-out groups (#302)

[NewGraphEnvironment]

Summary

• default_tuned gains mean-annual-discharge ranges for BT, GR, KO and RB, the species default leaves without one. Any group moved to mad used to lose all of their stream habitat. The values (m³/s, every maximum open) are BT 0.078 spawn and rear, GR 0.96 spawn and 0.97 rear, RB 0.011 spawn and 0.019 rear, and KO 0.57 spawn, unscored. They change no output while every group is on cw.

• Each value was calibrated on fish observations at the 46 fresh_default groups that have discharge, using the new data-raw/query_habitat_thresholds_mad.R with lnk_habitat_validate() as the instrument. Each was then scored on held-out groups by a rule fixed before any build. Every P10 → P05 loosening was refused except RB spawning, so the walked values land.

• Cost before moving any group to mad: on the held-out groups, BT keeps at least 33 % less stream rearing than under cw, GR at least 82 % less, and RB 4 % less. Stream size and sampling effort are not yet separated, so a size-stratified core is the next tuning. Details: research/habitat_thresholds.md, "MAD (discharge) ranges for BT, GR, KO and RB (Tune MAD (discharge) thresholds for BT, GR, KO and RB rather than leave them missing #302)".

• The Research: calibrate CH and BT gradient and channel-width thresholds from observations #284 variants harness now scores MAD ladders.

  • New model and set columns.
  • An anchor rung steps from the cw base to the first mad rung.
  • _min columns loosen downward.
  • A MAD ladder's core is its rungs only.
  • New flags --floor=found|expected and --working-prefix.
  • Guards for Research: calibrate CH and BT gradient and channel-width thresholds from observations #284's state.

  Re-scoring Research: calibrate CH and BT gradient and channel-width thresholds from observations #284's build reproduces all ten committed outputs.

Related Issues

• Closes Tune MAD (discharge) thresholds for BT, GR, KO and RB rather than leave them missing #302
• The harness was built here rather than after Score the discharge (mad) habitat model against observations, as #284 did for gradient #300; Score the discharge (mad) habitat model against observations, as #284 did for gradient #300 can now be mostly a run on it
• Relates to NewGraphEnvironment/sred#24

Test plan

• devtools::test(): 2,378 pass, 0 fail, 16 warnings, the same as the existing baseline. test-lnk_config.R pins the new default_tuned cells.
• lnk_config_verify(lnk_config("default_tuned")): no drift.
• devtools::check(): 0 errors. The 3 warnings and 2 notes are all in R/ and man/ files this branch does not touch.
• Research: calibrate CH and BT gradient and channel-width thresholds from observations #284 regression: the modified score script on Research: calibrate CH and BT gradient and channel-width thresholds from observations #284's build reproduces verdict, bands, bands_pooled, totals, habitat_change, taper, elevation, elevation_adjusted, bridge_band and summary.
• /code-check with three rounds each on the driver, the harness and the landing, plus a plan review. The harness and landing rounds ended by enumeration, of 83 derivation sites and 58 prose claims.

Notes

• Decisions you made mid-run:
  • a thin spawn cell falls back to any stage;
  • the expected-count floor is the floor of record;
  • an underpowered walk lands P05, unscored.
• Not filed: a fresh issue. fresh's main rear predicate ignores a wetland rule's wetland_ha_min. The draft is in the session for your review.
• NEWS and the version bump are left for the release commit (/gh-pr-merge).
• Runs: build 145 min, 20-group closure, schemas score302_* / working_score302_* in local docker fwapg.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WJE2szSCkeV9R3qKhuSwyx