Continuation checkpoint: weighted facility assignment review

9 October 2026. Previous turn: progress through selected MUB source-data review and public planning fit, published and verified at commit b0036b295980493e298be5d1b3b12ade9e7a8772. This turn: progress through a separate weighted/capacitated facility module, exhaustive tiny validation, eleven generated exports and updated product decisions. The perpetual goal remains active; elapsed time is not a completion condition.

Product finding and implemented scope

Build first: an Optimization Assurance Workbench for facility-planning reviews by logistics/operations consultancies. The new local module checks weighted rectangular costs, independent client/site IDs, required openings, exactly-k/at-most-k selection and whole-client capacity assignments. Original data are preserved; unsupported business constraints are refused. A conventional optional OR-Tools adapter proposes plans that receive separate exact checks. No family-125 theorem is attached to this broader model.

The capacity example rejects overloaded nearest-site assignment, checks submitted cost 12 and an alternative cost 10. Exact relaxed lower bound 4 means suboptimality is at most 6, not that the optimum differs by exactly 6. A generated 100-client/10-site case has submitted cost 5472, checked cost 2373 and bound 2260. These are synthetic units. Solver OPTIMAL stays distinct from independent optimality; solver INFEASIBLE is not an independently accepted certificate.

The website hosts static examples. No customer upload endpoint, accounts or hosted optimization jobs exist. Buyer demand, useful review improvements, actual studies and profitable delivery remain unvalidated. Local browser intake and a permissioned-study evaluation are the next product steps. No outreach, purchases or paid compute occurred.

Current artifacts and verification

• Module guide and eleven examples, current utility, saved solver comparison.
• 7,926 main controls, including 7,776 models and 93,312 submitted plans against a separate direct oracle; 26 display/intake controls. Total added: 7,952.
• Artifact manifest: 301 files, six inert tool copies, ten inherited family-125 source hashes. The display edition preserves original computed JSON/raw inputs and records its renderer separately. No archived tool was executed.
• First-product plan v4, dossier 011 v5, pilot v15.
• 38 utilities v23, priorities v23, feasibility v21, 41-dossier catalog v26, 372-family directory v21, coverage v26.
• Integrity v25: 29,805 passing component checks in 60 reports, 1,564 local links, 143 current Python syntax checks, zero issues. All 22 assignment export manifests, 196 contained files, 22 computation-tool records and eleven renderer-tool records checked. This checkpoint was written afterward; static-site validation covers its links.
• Reading ledger v15 unchanged: 42 selected-section records and 105 source hashes among 719 manuscripts. No new manuscript reading, source execution, Lean kernel acceptance or full proof badge was added.

Publication and continuation

Normal GitHub push and automatic Dokploy application 6yFD3JCmfp0uZcCuJGex2 remain authorized, followed by HTTPS byte/hash verification and a dated publication record. Preserve all historical files and outputs. Do not delete, reset or force-push.

Pinned upstream remains fd4aeeb2ee4fc729c18d98444fed42fd0529eeeb. The latest remote check at 01:55:11 UTC found no change. Do not repeat before 02:55:11 UTC without a new signal. The existing hourly heartbeat remains active while the app/host are available.

Independent next work includes local browser intake, stronger exact capacity/weighted bounds, realistic evaluation materials and further unreviewed manuscripts. Source proof acceptance and practical implementation remain separate from finite controls. Continue meaningful research and corrections rather than polling unchanged state.