What (One-liner)

Fork an upstream reaction-function knob into named variants (mode A / mode B), re-run the whole network once per variant, and compare the divergent endgames side by side — the runnable kernel behind a report §7 forward-scenario figure. Scenario, not prediction.

How to use

US President (agent_v1) → FOMC → Rate Transmission → [12-node US Treasury fragility network]
                                                          │ cascade / spectral / regime / portfolio
                                                          ▼
                                            Scenario Fork (mode A hawkish hold | mode B cave-to-cut)
                                            → run network ×2 → compare ρ_A vs ρ_B, endgame A vs B

Wire the endgame nodes (cascade_backstop / spectral / regime / portfolio) INTO the fork — those become the compareTargets. The fork's variants declare which upstream knobs to vary (the President stance + the decided rate path). Running the graph runs each variant and shows the A/B compare.

Core formulas

See frontmatter. executeAsGraphlet detects the fork node → runScenarioFork re-runs the (fork-stripped) network per variant with overrides applied → extractOutcome pulls each compareTarget's endgame field → computeDelta reports the A vs B difference.

Assumptions & applicability

Assumptions: ≥2 variants, compareTargets emit known fields, each variant framing sourced + observable, N× cost acceptable.

Applies to: forward/counterfactual scenario analysis on a fragility network, report §7 mode-A/B figures, stress-testing under switchable postures.

Does not apply to: point prediction, psychological prediction of an individual, live decisions, single-operator numeric ground truth.

Known limitations

See frontmatter. Five items: orchestration-not-measurement / N× cost / partly hand-authored framing / scenario-not-prediction / office-holder caveat.

§5 discipline (political/policy figures)

Variants are sourced framings of the office's reaction function — observable policy postures ("defend Fed independence" vs "pressure for cuts" are both documented public postures), never psychological prediction. Each variant carries a sourced framing (value + sources + observable:true), the panel shows a fixed "scenario, not prediction" label, and the footer carries the strongest disclaimer + methodology byline.

References

• Pangura Society Decision-Process Design (2026-06-22)
• Pangura Sourced Actor Model Design (2026-06-21)

Golden Test

Orchestration tier — no numeric golden of its own; endgame correctness is inherited from the compareTarget operators' goldens (cascade_backstop_v1 / spectral_fragility_v1 / cross_asset_regime_v1). The fork's pure logic is unit-tested in scenarioFork.test.ts (11 cases).

Changelog

• 1.1.0 (2026-06-25) — M2b (Phase 3a): engine='desk' operator-grounded Fed + Treasury desks → emergent rescue mode vs the rule; baked dual-desk demo + renderer desk column
• 1.0.0 (2026-06-23) — First cut (Society §7 forward; reaction-function mode A/B fork, orchestration tier)