examples: emergency_readiness_portfolio — the weakest-link (min) objective showcase - #137
Merged
Conversation
…e showcase The first bundled example whose signature objective is min-aggregated: portfolio Readiness is the weakest hazard line, not a total, so funding an asset can lower it. - 96 response assets, 20-24 funded, 4 objectives (Readiness min / SurgeCapacity sum / Cost sum / ActivationTime avg), 45 coupled constraints, 2 scenarios. - Flagship proof-layer receipt, promoted to the README headline: the same NSGA search rerun under another seed tops out at a weakest link of 38.7 while the exact audit proves 41.8 attainable — on a min objective a heuristic's best value is a lower bound on what's attainable; the audit probe confirms the true ceiling (generator CLAIM-14). - The exact MILP gate declines the shape by design (min is nonlinear, avg is fractional); the proof layer is explore audit, whose feasibility solves reason through the constraint set alone. - Governance: a ten-conjunct coverage guarantee holds across every feasible portfolio, the tightened variant is refuted with a witness, and an emergent $25.5M minimum spend falls out of the floors. - Bottleneck: a mandated group floor caps the weakest link at that group's k-th best asset, so no legal portfolio beats 41.8 at any budget down to $27M/yr — proven by a pair of HiGHS feasibility probes and attained by the baked frontier. - Bundle re-baked on the post-merge-train engine (adds the scenario base_constraints_fingerprint / base_fingerprint fields; solutions unchanged). Upstream kit + ask-literal guard in tests/test_upstream_kits.py; index row and audit paragraph in examples/README.md.
cafzal
force-pushed
the
claude/readiness-example
branch
from
August 3, 2026 22:27
29d5dd7 to
79f74cc
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the emergency_readiness_portfolio bundled example: 96 readiness assets, 2^96 portfolios, and the portfolio's first min-aggregated signature objective (weakest hazard line), with subregion floors/caps, crew-sharing exclusions, and cache dependencies.
Showcase. The proof layer earns its place: the same NSGA search rerun under another seed tops out at a weakest link of 38.7 while the exact audit proves 41.8 attainable (a 100-seed survey during prep found 53 misses, 38.7 modal; the repo-asserted claim is the single seed-2 rerun). On a min objective a heuristic best is a lower bound — the audit probe confirms the true ceiling, pinned by a single asset (Delta hazmat), and shows budget is not the lever.
Discipline. Seeded generator asserts all 14 README claims (including the seed-2 miss); bundle baked on current main — byte-identical re-bake post #132/#134/#135/#136. Upstream kit guarded by tests/test_upstream_kits.py (+52 lines). Index row + governance-paragraph mention in examples/README.md.
Verification. Full suite 1042 passed / 2 skipped on the branch; live pipeline-tested through the MCP tools (three-layer bar: worth the agent / the metaheuristic / the proof layer — verdict publish); provenance/secrets sweep over origin/main...HEAD clean.