This analysis isolates three length-frequency (LF) choices around the reviewed BET 2026 effort-creep model. It uses the existing MFCL executable and changes neither MFCL source code nor executable files.
Every cell retains the exact effort-crept bet.frq archived by Kflow Job 5319.
The tag-group controls and display maps are refreshed from
PacificCommunity/ofp-sam-bet-2026-stepwise branch
experiment/tag-grouping-reg-scaling-2026, commit
26c74dc6f303faa951b1ab331d7de14ea20b7489.
The tag observations come from
PacificCommunity/ofp-sam-2026-BET-YFT-tag-prep main, commit
79733c429b320e84ed5047aa6c932c8f19dab187, file
BET/bet.2026.low.recaps.removed.tag. This file is byte-identical to
experiment/pdh-skip-opr-lengthsel Step 13 13-DataWeighting/model/bet.tag.
Effort creep is not reapplied.
The reference bundle is stored in reference-inputs/job-5319/mfcl-inputs.
| Provenance item | SHA-256 |
|---|---|
| Refreshed 10-file reference bundle | a8e0598d06a1f795bf5cd0ced5c19e4462fa16921fde7412b295e460cacc8dbc |
Retained Job 5319 bet.frq |
d77f97c348409f845f1f0fc801af808d15b6cb119349d1f083308cfc9d4fba8c |
Refreshed bet.ini |
3c9503e0762547762bab20b26997c3a4e627b0965b1d88418d71a1a17f40bb11 |
Latest tag-prep bet.tag |
3f1b836a844ec2ca8e70fc5814d94c5a1ebc37ff4a5571c1dc1f6b83e477dfe8 |
The refreshed ini contains 98 tag-release rows and retains
tag_flags(:,2)=0 for every row. tag_rep_map.R records the updated reporting
rate groups, initial values, targets, penalties, and fishery names from the ini.
The existing Job 5319 doitall.sh is retained so this refresh does not
introduce later OPR controls.
The full factorial is defined in job-config.R.
| Axis | Levels | Scope |
|---|---|---|
| MFCL LF tail compression | 0%, 1%, 3%, 5% | All observed LF samples |
| Observed upper-bin zeroing | None, above 100 cm, above 70 cm | F21, F22, F23 only |
| LF likelihood downweight | 1x, 10x, 100x | F21, F22, F23 only |
| Regional-scaling weight | 50 | Fixed in all cells |
For a cutoff cell, observed LF counts in bins with midpoint above the stated
cutoff are set to zero. The bins remain as categories in the MFCL option-3 LF
likelihood, and MFCL internally renormalizes the retained counts. Counts are not
moved to another bin. If no LF count remains, the LF vector is replaced by one
-1 whole-sample sentinel. Record metadata and weight-frequency data are
unchanged.
Tail compression is the existing MFCL behavior controlled by flags 311 and 313; it pools tail mass rather than discarding it. Downweighting adds only flag-49 overrides for F21/F22/F23, using divisors 20, 200, or 2000. All inherited Job 5319 fishery settings outside those three overrides remain unchanged.
Each model stores the MFCL-ready 20x5 active matrix as bet.reg_scaling and the
complete 292x5 sensitivity source as bet.reg_scaling.full. The active matrix
is exactly full-source rows 53:72; MFCL reads only bet.reg_scaling.
Rscript R/prepare_bet_2026_step_inputs.R
Rscript R/validate_sensitivities.REach of the 36 titled folders contains a complete doitall input set, a concise
README, and input_manifest.csv. Cutoff cells also contain
model/lf_cutoff_audit.csv. Neither script submits Kflow jobs.