Eastern Bering Sea pollock stock assessment

Assessment record, model bridges, and current development

The central orientation page for operational EBS pollock assessments and the ADMB, RTMB, Rceattle, and SPoRC model-development program.

This site is the authoritative orientation point for the annual Eastern Bering Sea walleye pollock assessment. It connects the accepted assessment record with reproducible model runs, review decisions, ADMB bridge results, RTMB development, the close Rceattle comparison, the SPoRC development comparison, and work toward the next data update.

Open the 2026 assessment workspace September 2025 research report

Important

This site brings together assessment results, supporting research, source files, and model history. The operational model and each alternate implementation remain separately versioned so that a documentation merge does not obscure differences in code, data, or model configuration.

Current status

Current assessment cycle

2026 workspace active

The assessment is anchored to the accepted 2024 model and methodological work completed during 2025. New observations enter only after delivery and validation.

Assessment overview →

Validated model bridge

ADMB ↔︎ RTMB reproduced

The restored September 2025 fixed-parameter bridge matches key ADMB outputs to less than 0.0005%, with an absolute total-NLL difference of 0.00410.

Bridge and update overview →

Comparative development

Rceattle close bridge documented

The fitted NonParametricPM implementation closely reproduces the modified m23_rceattle_full ADMB reference. It is a close bridge, not a bit-for-bit translation.

Rceattle handoff results →

Ecosystem research

Early survival pathways active

The ESP organizes what is known by life stage, causal hypotheses, cohort-aligned indicators, and candidate PCMCI+ and DSEM analyses for recruitment processes.

Explore early survival pathways →

TipBiological patterns in fishery and survey data

Otolith and acoustic-survey data provide a separate view of pollock biology. Current summaries examine sex ratio by age, length, season, year, and area, along with sex-specific length-at-age patterns. These results can help identify questions for assessment interpretation and future sensitivities; they are not currently part of the model-bridge acceptance tests.

Explore fishery-age and sex-ratio patterns →

How the pieces fit together

Part of the assessment Scientific question Where to find it
Assessment record What data, model, review decisions, and advice define each assessment year? This repository and its annual tracking tables
Research narrative Why was the ADMB-to-RTMB transition undertaken, and what broader assessment research accompanied it? September 2025 report
Exact bridge Does RTMB reproduce the paired for_rtmb ADMB model at matched parameter values? ADMB–RTMB summary
Close bridge Does fitted Rceattle reproduce the modified m23_rceattle_full ADMB assessment trajectory? ADMB–Rceattle summary
Near-exact bridge Does SPoRC reproduce its paired ADMB reference closely enough for a controlled September review comparison? SPoRC development assessment
Biological patterns in observed data How do sex ratio and length-at-age vary by age, length, season, year, and area in fishery and acoustic-survey samples? Fishery-age and sex-ratio report
Ecosystem and recruitment research Which environmental and biological pathways may affect survival from eggs and larvae through recruitment? EBS pollock early survival pathways
Updated-data comparison How do matched ADMB, RTMB, Rceattle, and SPoRC configurations behave with the next common data update? Future annual comparison page on this site
Uncertainty and sensitivities What do profiles, retrospectives, selectivity alternatives, and MCMC show after the deterministic model is accepted? Linked method-specific reports

The exact RTMB, close Rceattle, and near-exact SPoRC comparisons use separately paired ADMB reference configurations. Their bridge statistics establish fidelity within each pair; they should not be interpreted as a direct ranking of software. The SPoRC bridge script and exact paired reference still need to be recorded in the review package.

How the models and results fit together

%%{init: {"theme": "base", "flowchart": {"curve": "basis", "nodeSpacing": 25, "rankSpacing": 55, "htmlLabels": true}}}%%
flowchart TD
  accTitle: Model lineage and evidence flow
  accDescr: The accepted 2024 Model 23 assessment leads to separate exact ADMB to RTMB, close fitted ADMB to Rceattle, and near-exact ADMB to SPoRC bridges. Those baselines and an accepted annual data inventory feed a future matched four-model update, common results, diagnostics, and accepted uncertainty analyses.
  A["Accepted 2024 assessment<br/><code>m23 / pm.tpl</code><br/>data · configuration · outputs"]

  subgraph B["2024 model comparisons"]
    direction LR
    subgraph T["Exact ADMB–RTMB bridge"]
      direction TB
      T1["ADMB reference<br/><code>for_rtmb/pm.tpl</code><br/><code>pm.par · pm.rep · pm.std</code>"]
      T2["RTMB translation<br/><code>R/Rpm.R</code><br/>matched values and map"]
      T3["Exact regression gate<br/>ΔNLL 0.00410<br/>outputs &lt; 0.0005%"]
      T1 -->|"same parameter point"| T2
      T2 --> T3
    end

    subgraph R["Close ADMB–Rceattle bridge"]
      direction TB
      R1["Modified ADMB reference<br/><code>m23_rceattle_full/pm.tpl</code>"]
      R2["Prepared workbook<br/><code>2024_EBS_pollock_<br/>m23_rceattle_full.xlsx</code>"]
      R3["Rceattle <code>NonParametricPM</code><br/>two-stage fit"]
      R4["Close fitted bridge<br/>2024 SSB −0.2%<br/>recruitment −0.9%"]
      R1 --> R3
      R2 --> R3
      R3 --> R4
    end

    subgraph S["Near-exact ADMB–SPoRC bridge"]
      direction TB
      S1["Paired ADMB reference<br/>exact configuration to record"]
      S2["SPoRC implementation<br/>bridge script pending"]
      S3["Near-exact bridge package<br/>due August 25"]
      S1 --> S2
      S2 --> S3
    end
  end

  A -->|"bridge configuration"| T1
  A -->|"modified configuration"| R1
  A -.->|"assessment inputs"| R2
  A -->|"paired configuration"| S1

  U["Accepted annual data inventory<br/>observations · covariance · weights<br/>maturity · catch · compositions"]

  subgraph C["Matched annual model update"]
    direction LR
    C1["ADMB<br/>reference run"]
    C2["RTMB<br/>matched update"]
    C3["Rceattle<br/>matched update"]
    C4["SPoRC<br/>matched update"]
  end

  T3 -.->|"validated baseline"| C2
  R4 -.->|"documented baseline"| C3
  S3 -.->|"bridge baseline"| C4
  A -.-> C1
  U --> C1
  U --> C2
  U --> C3
  U --> C4

  C1 --> C5["Common results<br/>likelihoods · SSB · recruitment · F<br/>selectivity · predictions"]
  C2 --> C5
  C3 --> C5
  C4 --> C5

  D["Acceptance and diagnostics<br/>input hashes → fixed-parameter bridge → optimization<br/>residuals · retrospectives · profiles · uncertainty"]
  C5 --> D
  D --> M["Accepted uncertainty and sensitivities<br/>MCMC and environment-linked analyses<br/>only after deterministic checks pass"]

  classDef source fill:#f5c242,stroke:#805f00,color:#172b3a,stroke-width:2px;
  classDef admb fill:#e8f2f8,stroke:#005ea8,color:#17324d,stroke-width:2px;
  classDef rtmb fill:#dff1e4,stroke:#27864b,color:#17324d,stroke-width:2px;
  classDef rceattle fill:#eee9f8,stroke:#6d5cae,color:#17324d,stroke-width:2px;
  classDef esp fill:#e2f3f0,stroke:#087f72,color:#17324d,stroke-width:2px;
  classDef evidence fill:#ffffff,stroke:#063b66,color:#17324d,stroke-width:2px;
  classDef future fill:#f4f9fc,stroke:#55778f,color:#17324d,stroke-width:2px,stroke-dasharray: 5 4;

  class A,U source;
  class T1,R1,C1 admb;
  class T2,T3,C2 rtmb;
  class R2,R3,R4,C3 rceattle;
  class S1,S2,S3,C4 future;
  class C5,D evidence;
  class M future;

  style B fill:#f7fafc,stroke:#8ba9bd,stroke-width:1px
  style T fill:#eef7f1,stroke:#27864b,stroke-width:2px
  style R fill:#f4f0fa,stroke:#6d5cae,stroke-width:2px
  style C fill:#f4f9fc,stroke:#55778f,stroke-width:2px
Figure 1: Model lineage from the accepted 2024 PM through the exact ADMB–RTMB, close ADMB–Rceattle, and near-exact ADMB–SPoRC bridge lanes to a future matched-data comparison. Solid arrows show implementation or fitting steps; dashed arrows carry validated baselines or shared inputs forward.

The diagram deliberately keeps the RTMB, Rceattle, and SPoRC branches separate until the common-data experiment. The exact RTMB gate validates a translation at a matched parameter point. The Rceattle result validates the behavior of a separately fitted model against its modified ADMB reference. The near-exact SPoRC bridge will be fully documented when its reproducible script and paired ADMB configuration are added. Only a controlled annual update can support a matched multi-model comparison. The ESP research workflow is shown separately in Figure 2 so it does not compress the model-implementation diagram.

The 2026 assessment overview maps this model-development plan to the named operational runs and current data, review, and delivery records. The software-comparison track is supporting evidence; it does not replace the operational assessment run sequence.

Early survival pathways

The EBS pollock early survival pathways site is the assessment’s ecosystem and recruitment research companion. It brings together hypotheses about survival from eggs and larvae through age-0, age-1, and age-2 fish reaching age 3. Its current development tracks include:

  • a citation-linked review of what is known at each life stage;
  • alternative directed acyclic graphs for competing causal pathways;
  • a cohort-aligned table linking environmental indicators to recruitment;
  • PCMCI+ guidance for time-series causal discovery;
  • DSEM development for multivariate pathway evaluation; and
  • technical and public-facing infographics.

The ESP does not participate in the fixed-parameter software bridge. Instead, it documents why an environment-linked recruitment hypothesis might be tested, which indicators and lags are defensible, and how results should be interpreted. Candidate covariates enter RTMB or Rceattle sensitivities only after the base deterministic configuration passes its bridge and optimization checks.

%%{init: {"theme": "base", "flowchart": {"curve": "basis", "nodeSpacing": 30, "rankSpacing": 55, "htmlLabels": true}}}%%
flowchart TD
  Q["Assessment recruitment estimates<br/>and cohort assignments"]
  E0["Environmental and biological indicators<br/>temperature · cold pool · prey · predators"]

  Q --> E1["Cohort information table<br/>eggs and larvae → age 0<br/>→ age 1/2 → age 3"]
  E0 --> E1
  E1 --> E2["Review of what is known by life stage<br/>causal DAGs and alternatives"]
  E2 --> E3["Candidate analyses<br/>PCMCI+ · DSEM<br/>hypothesis evaluation"]
  E2 --> I["Recruitment interpretation<br/>ecosystem context and indicators"]
  E3 --> I

  G["Deterministic model gate<br/>bridge · optimization · diagnostics"]
  I -.-> S["Candidate environment-linked sensitivity<br/>clearly named and compared with the base"]
  G -->|"must pass first"| S

  classDef model fill:#ffffff,stroke:#063b66,color:#17324d,stroke-width:2px;
  classDef source fill:#f5c242,stroke:#805f00,color:#172b3a,stroke-width:2px;
  classDef esp fill:#e2f3f0,stroke:#087f72,color:#17324d,stroke-width:2px;
  classDef future fill:#f4f9fc,stroke:#55778f,color:#17324d,stroke-width:2px,stroke-dasharray: 5 4;

  class Q,G model;
  class E0 source;
  class E1,E2,E3,I esp;
  class S future;

  click E2 "https://afsc-assessments.github.io/ebswp_esp/" "Open the EBS pollock early survival pathways site"
Figure 2: How the early survival pathways program connects assessment recruitment estimates and ecosystem indicators to a review of what is known, candidate analyses, interpretation, and model sensitivities.

Open the ESP overview Review causal DAGs Review DSEM development

Assessment timeline

2024 · accepted assessment anchor

The committed assessment supplies the operational model, annual data update, diagnostics, projections, and review record used as the current accepted baseline.

2025 · review and model translation

There was no operational assessment. Work focused on the CIE response, ADMB-to-RTMB translation, alternate model implementations, diagnostics, and preparation for the next assessment cycle.

2026 · integrated assessment workspace

The active workspace records data delivery, model-run roles, validation, review decisions, and final deliverables. Matched model comparisons should be added only after common inputs and acceptance gates are documented.

Model bridge program

Exact ADMB–RTMB bridge

The restored fixed-parameter comparison is the deterministic regression gate for RTMB development:

Metric Result
ADMB total NLL 7940.720000000
RTMB total NLL 7940.715902252
Absolute total-NLL difference 0.004097748
Maximum key-output difference 0.000492728%
Maximum absolute gradient 6.3653498e-06

Open the RTMB development repository

Close ADMB–Rceattle bridge

Grant Adams’s handoff pairs Rceattle with the modified ADMB/m23_rceattle_full model. Terminal 2024 differences for the fitted NonParametricPM model were -0.2% for spawning biomass, -0.9% for recruitment, and -0.2% for total biomass. Early-series differences are larger and the comparison is appropriately labeled a close fitted bridge.

Note

The detailed Rceattle handoff and separate SparseNUTS reports are staged for publication with the RTMB model-development documentation. Public links will be added here after that documentation branch is deployed.

Next common-data comparison

The next controlled experiment should compare separately named ADMB, RTMB, and Rceattle configurations using an identical accepted data inventory. It should gate inputs, fixed-parameter likelihoods, optimization, state variables, diagnostics, and uncertainty in that order. No updated-data equivalence is claimed until those checks are complete.

Annual assessment framework

The annual workspace makes the following relationships explicit:

  • data delivery and validation;
  • model run ancestry and scientific role;
  • bridge, base, and sensitivity distinctions;
  • assessment-to-projection traceability;
  • Plan Team and SSC review decisions; and
  • final deliverables and release status.

2024 framework example

Run Role Change from parent Status
m23.ly Bridge Prior Model 23 with prior-year data Comparison
m23 Updated base Adds new AVO, BTS, ATS, catch, and fishery-age data Author-recommended
m1 Data withholding Excludes new BTS age and biomass inputs Comparison
m2 Data withholding Additionally excludes new ATS inputs Comparison
m3 Data withholding Additionally excludes new AVO input Comparison

The sequential survey-data exclusions produced relatively minor spawning-biomass changes. The principal change from the previous assessment was associated with updated 2023 fishery catch-at-age information.

Documentation principles

  • The annual assessment record lives here.
  • Executable model projects retain their own Git histories and tests.
  • Exact and close bridges are labeled distinctly.
  • Development reports link to precise configurations and artifacts.
  • Missing source information remains visibly marked rather than inferred.
  • Published advice must trace back to reviewed model and projection outputs.

This structure lets the site act as the motherhood page without turning it into a monolithic model repository.