A cheating model can improve a strategic score by deleting a load-bearing transition while keeping the terminal value. We characterize the incentive, show an optimizer finding it, and gate the unpaid state change.
Analytic and observed deltas agreed. Maximum absolute difference: 1.49×10⁻⁸ USD.
Every route had at least one score-improving admissible deletion.
Score-seeking optimizer found baseline-beating uncovered structures without being told the exploit mechanism.
Honest plan selected as
more complete
Silenced plan selected as
the better plan
Completeness recognition did not override the improved scored economics.
What changed between the as-of state and the claimed target? A claimed leap that skips the covered transition leaves a missing obligation: the work was never done.
No covered transition → no score release. The score is suspended and a repair is requested; rescore after the route is covered.
No inheritance without coverage: terminal value cannot flow back through an uncovered state change. DOCK or disqualify only under declared policy.
When uncovered gains cannot receive a score, the agent starts repairing the route instead of optimizing the omission.
Routes with baseline-beating uncovered gains.
Uncovered gains still occurred, exposing registry-provenance limits.
Search redirection, not proof of real-world optimality.
Compiler-aware co-authors produced zero-obligation evasions in every v1/v1.5 condition.
Boundary: the registry is reconstructed from authored text. An adversary can make its own fields internally consistent.
Fundability-by-silence fell to zero when source-grounding and delta-indexed cost floors were combined.
Boundary: cost floors reduce the reward for a claimed jump; they do not prove the transition performs the missing work.
We already have bottom-up calculation and source-verification systems sketched for the next protocol version.
That may narrow the score channel, but it can be expensive and still may not prove semantic completeness.
01 Each analysis shows which work is worth doing — and which is not — and turns that into a testable next experiment.
02 Stronger resolved results attract the next cohort's participants.
03 Better ROI keeps LPs happy — and the flywheel funded.
linkedin.com/in/manchuliantsau