MashupAbility//OpenSource

ILLUSTRATIVE POLICY MODEL · OPENNASH 1.1

Inspect the model.
Then test the policy.

OpenNash is a transparent 2×2 policy stress test. It publishes the payoff matrix, every policy transform, all equilibria, uncertainty results, and a durable run receipt. The model is an argument you can inspect. It is not a forecast.

Open the workbench ↓Read the method ↓All projects ↑

THE CLAIM BOUNDARY

An equilibrium is conditional evidence, not a forecast.

The solver answers one narrow question: given this matrix and these policy transforms, which strategy profiles satisfy the stated equilibrium conditions?

Change the assumptions and the answer can change. OpenNash makes that dependence part of the output instead of hiding it behind a verdict.

MODEL WORKBENCH

Change the assumptions in public.

Select a worldview panel and apply three direct one-shot payoff transforms. OpenNash reports the complete equilibrium set and tests whether the submitted matrix is a stag hunt. It does not choose a preferred policy.

Scenario
ai-pact@1.1
Solver
opennash-2x2@2.5.0
Payoffs
Grade C · authorial
Model class
Simultaneous · one-shot

SET INPUTS

Policy package

0.0 / 8model units
Worldview panel

Each panel is an authorial hypothesis. Compare results across both panels.

Capacity investment

Add the selected amount to one actor whenever that actor chooses its cooperative strategy.

uᵢ′(s) = uᵢ(s) + investment, if actor i chooses its cooperative strategy
Market-access condition

Subtract a direct expected payoff effect when the selected actor chooses its adversarial strategy.

uᵢ′(s) = uᵢ(s) − condition, if actor i chooses its adversarial strategy
Transition transfer

Move payoff from one actor to the other only in the mutual-cooperation cell. The transform is zero-sum.

uᵣ′ = uᵣ + transfer; uₚ′ = uₚ − transfer, only when both cooperate

Intervention envelopecapacity investment + market-access condition + transition transfer ≤ 8

INSPECT MODEL

Published payoff matrix

Growth and interdependenceThis authorial panel gives substantial value to mutual capacity growth and verified exchange while retaining a short-run temptation to exploit diffusion.

GRADE C
First value: United States. Second value: China.
United States ↓ / China →Trade industrial access for verificationCommoditize intelligence + use chokepoints
Build energy + diffuse intelligence
(8, 8)

US6.5–9.5Grade C

CHINA6.5–9.5Grade C

Rationale + provenance

The United States expands energy-backed adoption and reduces its reliance on closed-model rents. China keeps industrial access credible and accepts verification costs. Both asset bases grow through complementarity instead of mutual depreciation.

United States: Authorial estimate of US gains from capacity growth and verified exchange.

China: Authorial estimate of Chinese gains from dependable industrial access and verified exchange.

(3, 9)

US1.5–4.5Grade C

CHINA7.5–10.5Grade C

Rationale + provenance

The United States pays to expand energy and diffuse intelligence. China captures more of the adoption gain through cheap weights and industrial leverage while US knowledge rents fall faster than physical capacity adjusts.

United States: Authorial estimate of US transition losses when diffusion outpaces physical adjustment.

China: Authorial estimate of Chinese gains from combining low-cost intelligence with industrial leverage.

Protect IP + constrain industry
(7, 3)

US5.5–8.5Grade C

CHINA1.5–4.5Grade C

Rationale + provenance

China offers industrial access and greater verification, but US containment preserves knowledge rents and limits the market benefit China expected from reciprocity.

United States: Authorial estimate of US gains from protected knowledge rents under partial reciprocity.

China: Authorial estimate of Chinese losses when verification does not produce expected market access.

(4, 6)strict NE

US2.5–5.5Grade C

CHINA4.5–7.5Grade C

Rationale + provenance

The United States protects proprietary intelligence and constrains industrial inputs. China commoditizes software value and uses manufacturing chokepoints. Each side damages the asset base where the other is more concentrated, and both lose wider adoption and investment.

United States: Authorial estimate of US security gains net of lost adoption and retaliation costs.

China: Authorial estimate of Chinese leverage gains net of constrained inputs and markets.

TRANSFORM LEDGER

0 payoff edits in the submitted run

No policy transform is active. The submitted run uses the published panel without edits.

COMPLETE RESULTCURRENT · RECEIPT READY

Both sides attack the other’s balance sheet.

US controls target manufacturing capacity. Chinese proliferation targets knowledge rents. Under this matrix, neither actor gains from a unilateral move away from the adversarial profile.

The complete equilibrium set contains exactly one pure-strategy Nash equilibrium, so OpenNash reports it as the unique reference outcome.
Pure equilibria
1
Mixed analysis
none
Coordination lens
not-stag-hunt
Cross-panel common
1
Uncertainty draws
2,000

STAG-HUNT COORDINATION LENS

Separate the best joint outcome from the safer response.

Not a stag hunt

The submitted matrix is not a stag hunt. China can gain or tie by leaving mutual cooperation alone.

Mutual cooperation
United States: Build energy + diffuse intelligence / China: Trade industrial access for verificationNot a strict equilibrium
Mutual defense
United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepointsStrict equilibrium
Payoff-dominant
United States: Build energy + diffuse intelligence / China: Trade industrial access for verification
Risk-dominant
Not applicable

COOPERATION-CONFIDENCE THRESHOLDS

United States50%

United States prefers cooperation when confidence that China cooperates is above this threshold.

ChinaNot applicable

The submitted payoffs do not create a stag-hunt threshold for China.

OpenNash reports this selection lens without using it to choose an outcome. Risk dominance is a property of the submitted payoff matrix. It is not a forecast of state behavior.

ALL EQUILIBRIA

Nothing is hidden by a verdict.

strict pure NE(4, 6)
United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints
Single-entry tie margin
1 payoff units
Local joint L∞ tie radius
0.5 payoff units
Binding constraints
United States: United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints → United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepoints · slack 1

NEAREST PURE-PROFILE BOUNDARY

0.5 payoff units

Exact payoff-entry L∞ distance to the nearest strict, weak, or non-equilibrium classification boundary.

  • United States: Build energy + diffuse intelligence / China: Trade industrial access for verificationChina: United States: Build energy + diffuse intelligence / China: Trade industrial access for verification → United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepoints · slack -1 · deviation gain 1
    not-equilibrium → not-equilibrium-to-weak
  • United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepointsUnited States: United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepoints → United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints · slack -1 · deviation gain 1
    not-equilibrium → not-equilibrium-to-weak
  • United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepointsUnited States: United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints → United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepoints · slack 1 · deviation gain -1
    strict → strict-to-weak
Mixed-strategy report · none

No additional interior mixed equilibrium exists for this matrix.

PAYOFF UNCERTAINTY

Does the result survive its own ranges?

≈80%

of seeded draws keep United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints as a pure equilibrium.

Sampling
Independent uniform draws across every published range
Seed
2152165525
Most common structure
contain|depreciate · ≈65%
United States: Build energy + diffuse intelligence / China: Trade industrial access for verification≈20%
United States: Build energy + diffuse intelligence / China: Commoditize intelligence + use chokepoints≈15%
United States: Protect IP + constrain industry / China: Trade industrial access for verification≈0%
United States: Protect IP + constrain industry / China: Commoditize intelligence + use chokepoints≈80%
Global sensitivity screen

Rank inputs by absolute Pearson correlation between the jointly sampled payoff and the sample-level sensitivity metric. This is a global screening statistic, not a Saltelli or Sobol variance decomposition.

Metric: Indicator that the point-game reference profile contain|depreciate remains a pure-strategy Nash equilibrium.

  1. contain|depreciate · US|r| ≈0.5
  2. rebalance|depreciate · US|r| ≈0.5
  3. contain|reciprocal · US|r| ≈0.0
  4. rebalance|reciprocal · CHINA|r| ≈0.0

ADVERSARIAL WORLDVIEW CHECK

Apply the same policy to both published panels.

Growth and interdependencecontain|depreciate
Security firstcontain|depreciate
Common pure equilibriacontain|depreciate

DURABLE RUN RECEIPT

Create the artifact that goes in the memo.

The server re-runs the model, hashes the canonical inputs and outputs, and stores a content-addressed, tamper-evident record. Identical runs resolve to the same receipt.

Read the calculation definitions
Pure equilibrium
No actor gains by one unilateral strategy change. A strict equilibrium has positive slack in every such comparison. A weak equilibrium contains at least one tie.
Single-entry tie margin
The binding incentive slack for one reported pure equilibrium. An equal one-payoff edit creates indifference. A larger edit makes the named deviation profitable.
Local joint L∞ tie radius
Half the binding slack for one reported pure equilibrium. This is not the nearest boundary elsewhere in the game.
Nearest pure boundary
The smallest payoff-entry L∞ radius to a strict, weak, or non-equilibrium classification boundary across all four pure profiles.
Stag-hunt lens
Mutual cooperation and mutual defense must both be strict equilibria, and mutual cooperation must strictly improve both actors' payoffs. Risk dominance compares the products of the unilateral-deviation losses at the two equilibria.
Uncertainty screen
Sample every published payoff independently from its uniform [low, high] range. Vary all payoff inputs together in each seeded draw. Rank inputs by absolute Pearson correlation between the jointly sampled payoff and the sample-level sensitivity metric. This is a global screening statistic, not a Saltelli or Sobol variance decomposition.

METHOD SPECIFICATION

Small enough to check. Complete enough to challenge.

The 2×2 model stays visible. Robustness is an added layer, not a substitute for the matrix, the rationale, or the model-class limits.

One model class

This release is a two-player, one-shot, simultaneous 2×2 normal-form game. It does not model repeated enforcement, hidden types, firms, allies, or changing capability stocks.

Three direct transforms

Capacity investment changes a cooperative payoff. A market-access condition changes an adversarial payoff. A transition transfer changes only the mutual-cooperation cell. Each changed number is shown.

Coordination without silent selection

The solver tests whether the matrix is a stag hunt. If it is, OpenNash reports payoff dominance, risk dominance, and cooperation-confidence thresholds. The lens does not choose an outcome or predict behavior.

Uncertainty is part of the result

Every authorial payoff has a published range. A deterministic simulation tests whether each equilibrium survives across those ranges and records the seed in the run receipt.

Scenario JSON ↗Solver source ↗Method note ↗

RELEASE LEDGER

What is complete. What remains open.

These states separate implemented checks from open validation work. They do not certify the model or its policy conclusions.

  1. SourcePublished

    Versioned matrices, payoff rationales, policy transforms, scenario data, and solver source for inspection. A separate code license is not stated.

  2. InternalPassed

    Canonical 2×2 games, transforms, input validation, deterministic uncertainty, and receipt hashing have executable tests.

  3. RobustnessImplemented

    Stag-hunt classification, risk dominance, cooperation thresholds, incentive margins, seeded payoff uncertainty, and cross-panel survival.

  4. ExternalOpen

    Independent solver cross-validation, expert elicitation, postdiction, named review, and a separate repeated-game model remain undone.

LIMITATIONS + INTERESTS

Read before citing a run.

  • All payoff values and ranges are authorial examples. They are provenance grade C, not calibrated estimates.
  • The result is conditional on the selected worldview panel, policy transforms, and uncertainty ranges.
  • The model does not forecast state behavior or recommend a policy.
  • The release has not completed expert elicitation, historical postdiction, or named external methodology review.
  • The one-shot model contains no detection-probability or commitment control. Those concepts require different model classes.
Conflict-of-interest statement

Social Protocol Labs operates MashupAbility//OpenSource and supports open-weight AI. The scenario includes open-weight diffusion. Treat its framing, payoff values, ranges, and rationales as authorial assumptions that can reflect that position.