run-modeling-project

Run Modeling Project

Plan each question's saved result, falsification check and figure purpose before solving. Keep the selected venue separate from the process profile; a failed check returns to the earliest affected stage and invalidates dependent figures and prose.

Coordinate the existing Skills around a strong final paper. The project record supports reproducibility, but it is not a second product or a substitute for modeling.

Workflow

1. Read references/workflow-profiles.md and choose rapid, competition (the default), or audit before creating deliverables.
2. Read references/distilled-competition-playbook.md, then record the problem, data sources, deliverables and venue constraints. Apply CUMCM-only layout rules only when that venue requires them.
3. Invoke analyze-model-data where data is used and select-model for decomposition, dependency analysis and structure opportunities. Classify each needed input as observed, externally supplied, model-derived, or unavailable. Before blocking on a missing derived input, decide whether the problem expects it to be estimated from information available at the decision time; if so, make that estimation a leakage-safe upstream subproblem. Invoke research-model-literature when external facts, parameters, datasets, or citations affect a claim.
4. Invoke solve-model for equations, algorithms, baselines, constraints, sensitivity and independent checks.
5. Invoke make-model-figures only after identifying the conclusions each figure must establish. Use the shared algorithms.figures registry through get_spec, list_figures and get_figure. Select a dedicated renderer with fields matching the saved model results; use mode="formal". Inspect the live registry and generated capability matrix for current structured formal routes. Preview uses the same validators with explicit mode="preview", simulated=True. Calling successfully does not establish paper suitability; verify the actual output and its evidence. English keys and registered Chinese names identify the same renderer. Pass provenance, units, independent unit and actual interval bounds into the figure handoff; never synthesize missing statistics. Carry the user's visual preferences with the figure handoff. For a new end-to-end project or a cross-stage revision, read references/research-evidence-handoffs.md; reuse existing records rather than creating parallel ledgers.
6. Invoke write-model-paper to turn verified results into a contest paper, then review-model-paper for reader-visible and technical review.
7. In competition, produce a formal PDF by default (PDF or DOCX is sufficient) and a passing submission preflight. Provide a support archive and clean-directory reproduction when the selected venue requires them. In audit, add both document formats, independent review, HMAC and complete hash/producer evidence.