Study Programs

The platform's evidence is organized as four study programs, one for each evidence tier — core, speed, quality, and platform. The tiers nest: the core program admits core and public evidence; the speed program adds speed evidence; the quality program adds quality evidence; and the platform program presents the complete public field. Each program scopes exactly what its evidence tier may reveal, so a lower-tier program never surfaces a higher-tier mechanism.

The showcase below reads every roster, benchmark family, comparison cohort, methodology detail, and claim-gate status from the committed programs bundle, so the figures stay current as the platform evolves without hand-edited prose. No comparative claim is published until a study program's claim gate opens; today's published exports carry smoke-development evidence only.

Core program

The core program studies foundational baseline objective performance across the baseline solver field — the comparison that grounds every other program. Its benchmark families are machine scheduling, job shop, and flexible job shop, and it runs the native core solver composition against a metaheuristic comparison cohort. Its methodology combines Wilcoxon signed-rank tests, a Friedman omnibus with Holm post-hoc correction, Nemenyi critical-difference analysis, and Vargha–Delaney A12 and Cliff's delta effect sizes with BCa bootstrap confidence intervals, over at least 30 independent runs per stochastic cell, with a campaign plan that runs pilot, comparative, ablation, and sensitivity studies in turn.

Its comparative claim gate stays closed until a full comparative campaign clears the statistical gates; the current published exports are smoke-development evidence. A companion write-up is published only from an approved, disclosure-filtered evidence bundle after this program's claim gate and the release gates open.

Speed program

The speed program studies runtime and anytime efficiency under matched termination — anytime traces, performance profiles, and a matched-quality-at-smaller-budget gate. Its benchmark families are flow shop and resource-constrained project scheduling, and its comparison cohort is iterated greedy, ALNS, and VNS. It applies the same non-parametric methodology and run floor as the core program.

Its comparative claim gate stays closed until a full comparative campaign clears the statistical gates; the current published exports are smoke-development evidence. A companion write-up is published only from an approved, disclosure-filtered evidence bundle after this program's claim gate and the release gates open.

Quality program

The quality program studies solution quality and multi-objective indicators — objective gaps, Pareto-frontier and critical-difference views, quality-indicator tables, and ablation. Its benchmark families are resource-constrained project scheduling and flexible job shop, and its comparison cohort is memetic search and NSGA-III. It carries the same statistical discipline and run floor as the programs above.

Its comparative claim gate stays closed until a full comparative campaign clears the statistical gates; the current published exports are smoke-development evidence. A companion write-up is published only from an approved, disclosure-filtered evidence bundle after this program's claim gate and the release gates open.

Platform program

The platform program presents the widest field the platform publishes — a curated cross-family roster that seats every solver family in the registry rather than every registered solver, with feature coverage, distribution-distance labels, and reproducibility and tooling evidence. In keeping with the platform inspection view, an unsupported claim routes to limitations rather than overclaiming. Its coverage and reproducibility posture is supported by artifact inspection today, a capability posture rather than a performance claim; its comparative gates stay closed like the others.

A companion write-up is published only from an approved, disclosure-filtered evidence bundle after this program's claim gate and the release gates open.

Adding a study program

A new study program follows the same fail-closed path the showcase above renders for every existing one. The future-candidate row in the showcase is the worked example.

1. Register the tier. Add one tier entry to the tier registry together with its evidence-tier value; the registry contract tests enforce completeness.

2. Declare the evidence track. Add an evidence track to the experiment atlas — its benchmark families, solver roster, comparison cohort, statistics, claim gate, and run-count floor — or start it as a future candidate first.

3. Scaffold the workspace. Create the program's workspace directory with its recipe, plus a pilot and a comparative campaign configuration:

   New-Item -ItemType Directory -Path .\experiments\programs\<id>
   Copy-Item .\experiments\configs\core-pilot.json `
     .\experiments\configs\<id>-pilot.json
   Copy-Item .\experiments\configs\core-comparative.json `
     .\experiments\configs\<id>-comparative.json

4. Run it. Execute the pilot campaign for the new tier, then curate the tiered bundle with the analytica command-line tool:

   uv run python experiments\scripts\regenerate.py `
     --config .\experiments\configs\<id>-pilot.json --tier <id>

5. Publish. Regenerate the portal so the program appears here automatically; the freshness gate fails the build if the regeneration is skipped:

   uv run python tools\build_site_assets.py

6. Document. Add the program's section to this page and mirror it across the locales.

Navigation

• Evidence bundles — the tiered exports each program curates.
• Reproducible campaigns — the workspace that runs the campaigns behind each program.
• Results explorer — the disclosure-filtered run evidence.
• Downloads — the published evidence bundles.
• Citation — how to cite the platform and its evidence companions.
• Platform inspection — the artifact-inspected coverage and reproducibility posture the platform program summarizes.