[mhovd]

Points PMcore at pharmsol main (e684464, ahead of 0.29.4) to measure the migration cost and the performance impact. Because this is a git dependency it can't be published as-is; it's meant to be merged once pharmsol cuts a release.

Size of the change

24 files, +133 / −147. Most of it is mechanical.

Library (src/, ~9 files)

• simulate_subject_dense → simulate; bind_error_models → validate_error_models; EquationTypes no longer re-exported.
• AssayErrorModels is now a BTreeMap<OutputLabel, _> (pharmsol#392):
  • error_optim: gamma_delta is indexed by enumeration position instead of the output index.
  • cycles.csv: closures take &OutputLabel.
  • validate_nonparametric_error_models: finds the error model for each observed output by resolving its label, instead of going through the removed bound/dense models. The outeq_<N> fallback (resolve_output_index) is deleted to match pharmsol.
• Prediction::outeq() returns &OutputLabel:
  • NPPredictionRow::outeq() now returns &OutputLabel (public API change).
  • BestDose turns the label back into an index, the same way it already handles observations.

Examples / benches / tests (~15 files)
pharmsol#383 removed the 1 ↔ outeq_1 / input_1 aliasing; labels now have to match exactly. Every model changes outeq_N/input_N → N, e.g. outputs: [1], infusion[1], y[1], "1" in AssayErrorModels::add. Downstream users with prefixed names will hit this at runtime, not at compile time.

Behaviour changes worth noting

1. Error-model factor optimisation now runs for label-added models. Before, AssayErrorModels::add("outeq_1", …) stored the model where iter() couldn't see it, so NPAG/NPOD skipped gamma/lambda optimisation and cycles.csv had no gamlam.* column. Both now work. (The pharmsol#392 commit message describes this bug.)
2. pred.csv outeq column now holds the data label (1) instead of the dense 0-based slot (0).
3. cycles.csv header is gamlam.<label> (e.g. gamlam.1).

Numerical check (bimodal_ke example, NPAG)

	main	this PR
Cycles to convergence	287	320
Final −2LL	−429.20383	−429.20387
Trajectory	—	identical to ~1e-12 through cycle 49
Final support points / weights	—	identical (weights agree to ~1e-13)

The extra cycles come from the error-model optimisation that now runs. It does two extra Ψ evaluations per cycle, and with lambda = 0 those are no-ops.

Performance (cargo bench --bench bimodal_ke, Criterion, same machine, baseline = main)

Bench	main	this PR	change
bimodal_ke_npag	3.093 s	2.919 s	−5.6 %
bimodal_ke_npod	1.376 s	0.993 s	−27.8 %
bimodal_ke_postprob	621 ms	456 ms	−26.6 %

NPAG improves less because it now does more work: ~11 % more cycles plus the extra Ψ evaluations. NPOD and postprob are closer to the raw simulator gain, about 27 %.

Testing

• cargo test --all-features: 84 passed, 0 failed.
• cargo clippy --all-targets --all-features: no new warnings (the 2 that remain are on main too).
• Smoke-ran every example. meta, drusano (solver stiffness) and vanco (label mismatch) fail identically on main, so they aren't regressions. iov/vanco_sde were capped at 90 s with no errors.

[github-actions]

Bencher Report

Project	PMcore
Branch	chore/pharmsol-main
Testbed	mhovd-pgx

Click to view all benchmark results

Benchmark	Latency	Benchmark Result milliseconds (ms) (Result Δ%)	Upper Boundary milliseconds (ms) (Limit %)
bimodal_ke_npag	📈 view plot 🚷 view threshold	3,539.70 ms (-18.54%) Baseline: 4,345.58 ms	6,448.85 ms (54.89%)
bimodal_ke_npod	📈 view plot 🚷 view threshold	802.24 ms (-38.79%) Baseline: 1,310.54 ms	2,051.75 ms (39.10%)
bimodal_ke_postprob	📈 view plot 🚷 view threshold	343.06 ms (-12.01%) Baseline: 389.89 ms	836.87 ms (40.99%)

🐰 View full continuous benchmarking report in Bencher

[Copilot]

🟡 Changes recommended

The upgrade makes residual error models label-keyed, but ParametricBuilder still stores them under dense indices, leaving constructed problems without models under their actual output labels.

1 open issue

• Use output labels when adding residual error models · New

What changed in this PR

Migrates PMcore to pharmsol main to assess API migration costs and performance ahead of a published release.

Changes:

• Updates simulation calls and error-model validation for the new APIs.
• Adapts error-factor optimization, prediction output, and cycle logs to label-keyed models.
• Replaces prefixed input/output aliases with exact numeric labels in examples, benchmarks, and tests.

File	Description
tests/​bestdose_tests.rs	Migrates test models to exact numeric labels.
src/​lib.rs	Removes the EquationTypes re-export.
src/​iov/​optimizer.rs	Uses the renamed simulation API.
src/​estimation/​problem.rs	Validates assay models using exact output labels.
src/​estimation/​nonparametric/​predictions.rs	Preserves output labels in prediction rows.
src/​estimation/​nonparametric/​cycles.rs	Adapts cycle logging to label-keyed models.
src/​bestdose/​cost.rs	Updates simulation calls and converts prediction labels to numeric keys.
src/​algorithms/​nonparametric/​error_optim.rs	Indexes optimization steps by model enumeration position.
src/​algorithms/​mod.rs	Updates simulation calls and label debug output.
examples/​vanco.rs	Uses the renamed simulation API.
examples/​vanco_sde/​main.rs	Migrates routes, outputs, and assay labels.
examples/​two_eq_lag/​main.rs	Migrates input, lag, output, and assay labels.
examples/​theophylline/​main.rs	Migrates route, output, and assay labels.
examples/​new_iov/​main.rs	Migrates route, output, and assay labels.
examples/​neely/​main.rs	Migrates multiple outputs and assay labels.
examples/​meta/​main.rs	Migrates infusion, output, and assay labels.
examples/​iov/​main.rs	Migrates route, output, and assay labels.
examples/​drusano/​main.rs	Migrates multiple inputs, outputs, and assay labels.
examples/​chain_algorithms.rs	Migrates the chained-estimation example’s labels.
examples/​bimodal_ke/​main.rs	Migrates the benchmark example’s labels.
examples/​bimodal_ke_backend_compare.rs	Aligns labels across DSL, legacy, and macro models.
examples/​bestdose/​main.rs	Updates DSL labels and their explanatory comment.
Cargo.toml	Points pharmsol at its Git main branch.
benches/​bimodal_ke.rs	Migrates model and assay labels for all benchmark cases.

_{Review effort: Balanced}