Skip to contents

Changelog

opal 0.1.1

  • 2026/10/15 (PR #143): Fix fit validation for random-effect models by calculating the marginal Hessian with optimHess() instead of calling RTMB’s unsupported exact Hessian. Converged, biologically valid random-effect fits can now pass validation and proceed to grid sampling. Diagnostic version v3 invalidates old checks and grid checkpoints; rerun opal_check() on existing v5 objects. This diagnostic fix does not change model objectives or fitted estimates.

  • Quantify the existing seasonal CPUE correction in reproducible paired simulations: maximum prediction changes at fixed parameters are 8.24% and 12.28%, and refitted B0 changes are +4.99% and +11.71%, for the specified two- and four-season examples. These are illustrative, not real-stock impact estimates or general correction factors. Full-coverage examples fail under the legacy indexing but evaluate under v5. The single-season control is unchanged, and all bundled stock inputs use one season. See the seasonal CPUE impact article for observation coverage, diagnostics, and instructions for rebuilding and refitting saved contract-v4-or-earlier assessments.

  • 2026/10/13 (e1fbdbd): Add optional SS3 GrowthModel 3 cumulative age-specific K multipliers and plus-group length adjustment, credited to Stock Synthesis (Methot and Wetzel 2013). Existing model configurations retain their exact growth calculation and numerical contract.

opal 0.1.0

This release makes the staged opal_obj workflow the primary way to run an assessment, from configuration and fitting through MCMC, diagnostics, projections and saved results. It also adds posterior and fit-checking tools. Scientific contract v5 changes CPUE predictions when n_season > 1, so saved models from earlier contracts must be rebuilt.

  • 2026/09/29 (0ea5022): Add biological fit and posterior checks with versioned validation records, posterior report summaries, OSA residuals and dataset-level SDNR plots, composition and prior–posterior plots, objective profiles, resumable model grids, balanced posterior selection, and deterministic equilibrium MSY summaries. Include a simulated posterior and an assessment-tools tutorial. OSA covers lognormal indices and multinomial, Dirichlet, and Dirichlet–multinomial length and weight compositions, including latent-state integration. Dirichlet–multinomial OSA uses sequential beta-binomial densities; ordinary fitting and simulation retain the existing density implementation. Fix seasonal CPUE indexing, boundary diagnostics, invalid prior handling, and figure units. Scientific contract v5 changes CPUE predictions when n_season > 1; the single-season Opakapaka reference is numerically unchanged. Earlier saved model contracts require rebuilding under the new contract. Projection behaviour, selectivity time blocks, and close-kin components are outside this change.

  • 2026/09/29 (fba96ea): Make assessment objects the primary documentation workflow, add a runnable projection vignette, organise the function reference, and publish a separately labelled development pkgdown site with internal-link checks. Model numerics and function interfaces are unchanged.

  • 2026/09/29 (3475a2e): Expand tests for projection uncertainty, recruitment forecasts, selectivity, parameter diagnostics, and bundled data access. Tighten optimiser stopping tolerances in the Quickstart and projection examples while retaining the existing fit-check thresholds. Package model calculations are unchanged.

  • 2026/09/29 (17ae7a6): Introduce the staged opal_obj workflow: configuration, two-pass fitting, MCMC, diagnostics, direct plots, projections with source identities, and portable opal_save()/opal_read() persistence. Preserve legacy fit construction and migrate saved fits with objective verification. Configuration changes invalidate dependent results; metadata updates and same-target refits preserve compatible samples. Model numerics and scientific contract v4 are unchanged.

opal 0.0.4

  • 2026/09/22 (2ae7722): Use portable integrity in the bundled-fit examples so they run across R versions while retaining strict model compatibility and objective checks. Build the package website on pull requests to dev as well as main. Model numerics are unchanged.
  • 2026/09/22 (06f0d7c): Constrain and penalise initial seasonal survival before seasonal compounding when fisheries overlap, and protect equilibrium recruitment from becoming negative under excessive initial fishing. The initialisation penalty is included in the model objective and reported as lp_init_penalty. Safely positive equilibria are unchanged. Previously saved fits use the earlier scientific model contract; the current contract is v4.
  • 2026/09/16 (c5dcdf1): Dirichlet-multinomial composition preparation now preserves half-up rounded effective sample sizes with largest-remainder integer counts. Existing saved fits use the previous scientific model contract.
  • 2026/09/16 (bac7a99): Fished equilibrium initialization now uses the same seasonal harvest-fraction survival convention as the population dynamics, eliminating drift for partial selectivity.