Bounded parameter scans

CompiledDocument::scan and pharmflux scan MODEL REQUEST.json evaluate an explicit ordered list of 1–25 parameter points. The shared run defines regimen, solver, initial-state overrides, covariates and base parameter values. Each point has a unique 1–128 byte ID and a parameter quantity map that overrides the base map. Defaults still supply unspecified model parameters. There is no implicit grid expansion. The fixture conformance/requests/synthetic-scan.json shows three clearance values for the owned one-compartment model.

Requests use pharmflux.scan/v0.1; results use pharmflux.scan-result/v0.1. Every point returns its ID, merged explicit parameter map and full simulation result with execution identity. The parameter map preserves supplied units; normalized bound values are represented by the simulation identity. The scan hash binds the full ordered request and scan algorithm. Points with equal scientific inputs may have equal simulation identities while retaining distinct point IDs. Point order is preserved.

All point parameter/initial-state bindings are validated before any simulation. Run-level protocol and execution validation still occurs in the ordinary run path. Each point receives the smaller of the run budget and its equal share of the total scan budget, independently for callbacks, events and numeric output values. This conservative reservation bounds aggregate work and retained numeric results. Unused budget from one point is not transferred to another. Execution is sequential within one scan; independent scans can run concurrently.

Any point failure aborts the operation without returning partial results. The error expression identifies its point. CLI files are never modified and retain the one-million-byte request limit (JSON model documents permit four million bytes). Generated request/result schemas describe structure; runtime checks enforce point limits, units, domains and budgets.

Qualification covers independent analytic trajectories, mixed-unit overrides, ordered results, distinct bound identities, malformed points, excess/empty point lists, aggregate-budget exhaustion, recovery, CLI parity and independent schema validation. This scan operation does not itself estimate global sensitivity indices.

Python execution

CompiledModel.scan(request) returns a dict; scan_json(request) returns JSON. Both accept the versioned envelope, release the GIL during native execution, and permit sharing a compiled model across independent scans. See the Python and R user guides for executable synthetic examples.