Introduction Data & base model Sensitivities Grid Projections

More

Getting started GT skipping Review B10+ and TRO OMMP16 Model notes ADMB vs SBT Data weighting Length frequency Packages sbt vignettes sbtdata vignettes

Stock Assessment for the 31st Meeting of the Extended Scientific Committee (ESC31)

This directory contains the ESC31 assessment workflow split into ordered documents. Run the documents in numeric order so each step can read the saved outputs from the previous step.

The HTML links open rendered reports. The QMD links open the source files.

Southern bluefin tuna painted by Joanne Webber.

Published-results checklist

✓ 2 Base
Accepted ✓ 3 Sensitivities
Accepted* ✓ 4 Grid products
Accepted and complete ✓ 5 Projections
Accepted and complete ✓ 6 GT programme
Comparison complete

*NoPOPHSP is accepted under its explicit two-divergence exception. The balanced posterior, direct-get_M 108-fit MLE grid, its 2,000-draw resample, and all four 2,000-draw production projection arms are accepted. The retained 100-draw caches are accepted diagnostic products; the older format-5 200-draw caches are legacy diagnostic evidence. Neither is the published result. The completed four-arm long-horizon GT comparison supports retaining annual gene tagging.

Step 2

Data & Base Model

Review the accepted one-block LL4 base MLE, MCMC, posterior MSY, parameter tables, model-fit plots, and residual diagnostics.

Open HTML Open QMD

Output: runs/esc31_base.rds

Step 3

Sensitivities

Using the selected separate standard LL3 and one-block standard LL4 base treatment with LL3-matched selectivity hyperparameters and the global A10 preventive wall, run and compare all 13 sensitivity models, including NoPOPHSP, then diagnose the base and sensitivity MCMCs.

Open HTML Open QMD

Outputs: one complete sbt_fit per model—runs/esc31_base.rds and 13 runs/sens/esc31_sens_*.sbt.rds files. Each canonical object keeps its MLE, MCMC, diagnostics, provenance, and posterior plot summaries together. NoPOPHSP retains its original registry slot and seed.

Step 4

Grid

Review the accepted nine-cell grid, balanced 2,000-draw posterior, and direct-get_M 108-fit MLE grid and resample. All cell-level and downstream grid products pass their numerical, sampler, provenance, and biological-state gates.

Open HTML Open QMD

Outputs: runs/grid_mcmc/ runs/projections/esc31_projection_mcmc_2000.rda runs/grid_mle/m0_m10_posterior_quantile_mle_grid_108.rds

Step 5

Projections

Review the accepted 2,000-draw production projection report. It covers both CTP allocation scenarios, the NoUAM comparison, and the sampled direct-M MLE grid; all four arms pass the report and CTP optimisation gates.

Open HTML Open QMD

Outputs: runs/projections/esc31_projection_run_2000.rds runs/projections/esc31_projection_run_scenario_2_2000.rds runs/projections/esc31_projection_run_no_uam_2000.rds runs/projections/esc31_projection_run_mle_grid_2000.rds

Step 6

GT year skipping

Review the accepted alternate-release-cohort MLE and posterior, the paired 100-draw CTP diagnostic, and the completed common-posterior four-arm, 2,000-draw long-horizon programme comparison. The comparison supports retaining annual gene tagging.

Open HTML Open QMD

Outputs: runs/gtskip/esc31_gtskip_odd_release_cohorts.sbt.rds runs/gtskip/esc31_gtskip_odd_release_cohorts_comparison_100.rds runs/gtskip/esc31_gtprogramme_comparison_2000_through_2055.rds runs/gtskip/esc31_gtprogramme_annual_full_2000_through_2055.rds runs/gtskip/esc31_gtprogramme_annual_reduced_2000_through_2055.rds runs/gtskip/esc31_gtprogramme_deliberate_skips_2000_through_2055.rds runs/gtskip/esc31_gtprogramme_skip_plus_failure_2000_through_2055.rds

Shared Inputs

The input helper loads the checksum-verified 2026 release from the private sbtdata package. The original files remain under ESC31/csv_2026/ for audit. Model fits, grid results, and projection caches are stored under runs/.

Run Notes

4_grid.qmd defaults to reporting existing grid outputs. To launch a new grid run, render with ESC31_RUN_GRID_MCMC=true and a stable ESC31_GRID_RUN_ID. Grid cells must pass the production diagnostics before they are combined. Decision esc31_2026_h_psi_dense_grid_v8_base_only_reuse_base_cell5_sparse_nuts_150_750 authorizes the recorded grid after the current base posterior passes its machine-validated numeric and biological gates. Decision esc31_2026_h_psi_grid_failed_cells_sparse_nuts_150_1500_v1 authorized the longer reruns now used in the accepted mixed nine-cell grid. The accepted base posterior is the mid cell grid element. The balanced 2,000-draw posterior, direct-get_M 108-fit MLE grid, and its 2,000-draw resample have been built and independently validated. The former supplies all 2,000 draws to the production projections; sensitivity MCMCs remain independent comparison runs. All base and sensitivity models retain preventive-wall decision esc31_2026_preventive_harvest_wall_a10_v1. 5_proj.qmd defaults to the accepted four-arm 2,000-draw production result. Set ESC31_PROJECTION_N_ITER=100 for an abbreviated smoke review or ESC31_PROJECTION_N_ITER=200 for the extended diagnostic. A normal render is cache-only. Explicitly set ESC31_RUN_PROJECTIONS=true, ESC31_RUN_NO_UAM_PROJECTIONS=true, and ESC31_RUN_MLE_GRID_PROJECTIONS=true only when the corresponding missing or incompatible caches should be computed.

Every current base and sensitivity posterior is embedded in its primary portable sbt_fit; separate sensitivity-posterior files are archival rather than canonical.