v0.10.0

ModelManager v0.10.0

Diff since v0.9.1

Merged pull requests:

• A refused sbatch submission is not a failed simulation (#51) (@drbergman)
• Drop eight dependencies ModelManager never loads (#55) (@drbergman)
• Restorable means JLD2 can name it; stored QoI values read back as written (#56) (@drbergman)
• Compact the maintainer records to what is true now (#57) (@drbergman)
• Give the variations databases the busy timeout, and clean up a deleted run's batch tags (#60) (@drbergman)
• Fix two discrete-parameter defects: resume, and co-variation joint consistency (#63) (@drbergman)
• An in-flight generation folder is not a completed generation (#64) (@drbergman)
• Publish the paths and defaults a backend was rebuilding by hand (#65) (@drbergman)
• Do not snap a discrete coordinate to the CDF grid (#66) (@drbergman)
• deleteCalibration(delete_subs=true) deletes only the monads that run alone used (#67) (@drbergman)
• Recover a campaign whose driver process died (#68) (@drbergman)
• Make the assertions that could not fail actually able to fail (#69) (@drbergman)
• One QoI contract across the sink, sensitivity analysis and calibration (#70) (@drbergman)
• Fix the remaining calibration review findings (#71) (@drbergman)
• Draw from an ABC-SMC posterior with samplePosterior (#72) (@drbergman)
• Choose which parameters a calibration or GSA plot draws (#73) (@drbergman)
• Draw the Sobol' total-order bars behind the first-order ones (#74) (@drbergman)
• Stop the workers from starting queued specs after run fails fast (#75) (@drbergman)
• Give the default reducer the two-argument form data selects (#76) (@drbergman)

Closed issues:

• Design brief: one QoI contract across the sink, sensitivity analysis and calibration (#52)
• Design brief: recover a campaign after the driver dies; make HPC detection overridable (#53)
• Design brief: the MM/PCMM boundary — interface contract, schema ownership, XML seam, exports (#54)
• deleteCalibration(delete_subs=true) deletes monads other runs still use, unlike deleteSampling (#58)
• Test gaps: assertions whose two arms coincide, and no end-to-end run() on the SLURM path (#59)
• CDF-grid snapping distorts a discrete parameter's prior and can leave levels unsampled (#61)
• Calibration review: the remaining 19 findings, triaged (#62)

v0.9.1

ModelManager v0.9.1

Diff since v0.9.0

Merged pull requests:

• Spread a keyed measurement into one sensitivity analysis per key (#49) (@drbergman)
• Move QoI evaluation into qoi.jl, over one shared reduction (#50) (@drbergman)

Closed issues:

• GSA is the only QoI consumer that won't spread a Dict-valued measurement (#48)

v0.9.0

ModelManager v0.9.0

Diff since v0.8.3

Merged pull requests:

• rm_hpc_safe: try the real removal, stage only the residue, sweep on startup (#26) (@drbergman)
• Auto-detect run_on_hpc during initialization (#27) (@drbergman)
• CompatHelper: bump compat for QuasiMonteCarlo to 0.4, (keep existing compat) (#29) (@github-actions[bot])
• Refuse to migrate when the loaded package version differs from the installed one (#30) (@drbergman)
• Close trial-ID accessor gaps; make trialID(::Vector{Sampling}) a pure lookup (#31) (@drbergman)
• Address a calibration run as coalesced Sampling views, and bring Calibration into the tag subsystem (#32) (@drbergman)
• Collapse runABC's duplicated defaults, make method.toml table-driven, and widen observed_data (#33) (@drbergman)
• Run sensitivity analysis over a CalibrationProblem, and report every unusable parameter at once (#34) (@drbergman)
• Name the two epsilons apart, and record the threshold each generation ran against (#35) (@drbergman)
• Represent discrete variations as DiscreteUniform over value indices (#36) (@drbergman)
• Add the proposal-distance histogram, and persist the distances it needs (#37) (@drbergman)
• Let calibration take discrete parameters (#38) (@drbergman)
• Give each generation its own folder (#39) (@drbergman)
• Make resume's method= patch instead of reset, and pair the entry points (#40) (@drbergman)
• Add StudySpec: build the model and parameters once, use them for either study (#41) (@drbergman)
• Open every database with a busy timeout (#42) (@drbergman)
• Add the QoI seam: one measurement, three consumers (#43) (@drbergman)
• Make run_kwargs and the loose splat one channel (#44) (@drbergman)
• One contract: every measurement function receives a Simulation (#46) (@drbergman)
• Detect SLURM completion from the filesystem; ModelManager runs the command (#47) (@drbergman)

v0.8.3

ModelManager v0.8.3

Diff since v0.8.2

Merged pull requests:

• Add trial tagging and feature-based recovery (#23) (@drbergman)
• Restrict docstring @ref links to public bindings (#24) (@drbergman)
• Docs findability: split post-processing, add Results & Analysis, declare the real public API (#25) (@drbergman)

v0.8.2

ModelManager v0.8.2

Diff since v0.8.1

Merged pull requests:

• Implement evaluation failure handling in ABC-SMC (#22) (@drbergman)

v0.8.1

ModelManager v0.8.1

Diff since v0.8.0

Merged pull requests:

• Fix sort_by handling in simulationsTable/monadsTable (#21) (@drbergman)

v0.8.0

ModelManager v0.8.0

Diff since v0.7.5

Merged pull requests:

• Bump codecov/codecov-action from 6 to 7 (#18) (@dependabot[bot])
• Bump actions/checkout from 6 to 7 (#19) (@dependabot[bot])
• Post-processing hook + QoI sink, monadsTable, batch run(Vector), and calibration budget fix (#20) (@drbergman)

v0.7.5

ModelManager v0.7.5

Diff since v0.7.4

Merged pull requests:

• Update documentation and assets for clarity and consistency (#17) (@drbergman)

v0.7.4

ModelManager v0.7.4

Diff since v0.7.3

Merged pull requests:

• docs: restructure into manual + API reference, add logo (#15) (@drbergman)
• Add plot recipes for GSA sampling results (MOAT, Sobol', RBD) (#16) (@drbergman)

v0.7.3

ModelManager v0.7.3

Diff since v0.7.2

Merged pull requests:

• Add console progress reporting to ABC-SMC calibration (#14) (@drbergman)