Dynamic Resonance

Win by Silence

Search for shorter paths to value — without rewarding missing work.

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.

ROUTE SEARCH COVERAGE GATE SCORE route A · covered route B · hidden work claimed Δ ↓ obligation ↓ admissible? score released route A ✓ score held — repair requested route B · uncovered Δ search redirects → substantiate resubmit covered
Dynamic Resonance

Can AI search for shorter paths to value without being rewarded for skipping the work that makes those paths possible?

Dynamic Resonance

Strategic errors become expensive before they become visible.

Omitted: Prototype Repeated demand Scalable growth the plan reads clean — the work is gone Honest: Prototype Proof of demand Repeated demand Scalable growth the load-bearing transition — costly, uncertain, necessary
Dynamic Resonance

Win-by-Silence is your Anti-Goodhart remedy.

Δk = (∏i<k pi) [ ck + (1 − pk) Rk+1 ]
Silenced record Prototype Repeated demand Scalable growth Proof of demand — deleted cost ck and risk pk vanish value still credited Covered record Prototype Proof of demand Repeated demand Scalable growth transition paid ✓
Removed-cost channel: the missing transition stops paying ck.
Retained-value channel: terminal value still flows through Rk+1.
Dynamic Resonance

Mechanism exists. Score-seeking search finds it.

Deletion identity
57/57

Analytic = observed

Analytic and observed deltas agreed. Maximum absolute difference: 1.49×10⁻⁸ USD.

Improving deletion
26/26

Exploit always available

Every route had at least one score-improving admissible deletion.

Optimizer discovery
21/26

Search finds it unprompted

Score-seeking optimizer found baseline-beating uncovered structures without being told the exploit mechanism.

One frozen 26-route cohort. EXP A establishes the mechanism under the mutation operator; EXP B establishes behavioral relevance — not an independent prevalence replication.
Dynamic Resonance

The judge saw the omission — and still preferred it.

26/26

Honest plan selected as
more complete

26/26

Silenced plan selected as
the better plan

Completeness recognition did not override the improved scored economics.

Dynamic Resonance

Protocol: no state change receives value without coverage.

Core invariant
claimed delta → obligation → admissibility decision
DELTA

Typed state delta

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.

GATE

Score release is blocked

No covered transition → no score release. The score is suspended and a repair is requested; rescore after the route is covered.

DOCK

Value credit is disconnected

No inheritance without coverage: terminal value cannot flow back through an uncovered state change. DOCK or disqualify only under declared policy.

Win by Silence protocol: when shortcuts stop paying, the AI starts repairing the route
Dynamic Resonance

Make silence unprofitable.

When uncovered gains cannot receive a score, the agent starts repairing the route instead of optimizing the omission.

1/26 → 13/26
Strict covered improvement · score-only → score + GATE
87%
of post-refusal revisions repaired to a covered structure (47/54)
Score-only search
21/26

Routes with baseline-beating uncovered gains.

Score + GATE
17/26

Uncovered gains still occurred, exposing registry-provenance limits.

Search redirection, not proof of real-world optimality.

Dynamic Resonance

We narrowed the economic reward. We did not solve semantic completeness.

Obligation channel
6/6

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.

Score channel
5/6 → 0/6

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.

Mitigation direction

We already have bottom-up calculation and source-verification systems sketched for the next protocol version.

  • derive transition cost from required work packages;
  • bind evidence to source packets and timestamps;
  • verify provenance before score release;
  • freeze prediction windows and resolution predicates.

That may narrow the score channel, but it can be expensive and still may not prove semantic completeness.

Dynamic Resonance

The first product is an instrumented decision cohort.

decision learning loop Cohort decisions real stakes, real follow-up Routes compile typed states + obligations Experiments resolve kill-or-confirm · 30/90/180 d Reference class improves the protocol gets sharper New cohort joins drawn by better evidence decisions → routes cheapest decisive test resolved outcomes sharper priors more decisions in
Direction

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.

Dynamic Resonance

Three load-bearing partners.

1 cohort sponsor
10+ strategic decisions and 30/90/180-day follow-up
1 R&D or agent team
A costly branch-selection problem with delayed feedback
2 working collaborators
Evaluation, formal methods, decision science, or engineering
Aleh Manchuliantsau
Founder · Dynamic Resonance
650.713.1091
QR code: linkedin.com/in/manchuliantsau

linkedin.com/in/manchuliantsau