Skip to contentCOUNCIL ENGINE⁺WhyRouteEvidence UltraCouncilRuntime ProofInvokeexecutable · 116fd5f0Decision court as a serviceCouncil + UltraA full think tank outside the work lane

Deep thought.
Zero drag.
One verdict.

Research, branch, precommit, reveal, attack, repair, judge. The whole Council runs in isolated contexts while coding and messaging continue—then returns one compact, receipted decision package.

Watch the court Read the contractFrozen questionscope · cutoff · budgetFact packetPX · live reads · GoalieAdaptive Ultrawaves · branches · falsifiersIndependent seatsprivate precommitTargeted repairmaterial issues onlySeparate judgeevidence, not votesCouncilfull-monty engineStandalone sidecar harnessRich inside · compact at the boundaryOperator remains final01 / The product thesis

Keep the brilliance.
Move the cost.

Manual Councils created exceptional ideas—but occupied the same live sessions needed to build them.

The goal is not a smaller Council. It is a full Council that no longer consumes the people and agents doing the work.01

Deliberation is isolated.

Research, branching, cross-examination, simulations, and judging run outside live coding and messaging contexts.

02

Reversible work keeps moving.

Only predeclared irreversible or high-risk gates wait for the verdict. Everything else continues in parallel.

03

The boundary is compact.

Live lanes receive status and one decision package—not the Council transcript. Deep branches hydrate only on demand.

04

Messaging is an adapter.

Foxtalk may carry a pointer or status. It is not the reasoning substrate, seat scheduler, or authority engine.

05

The operator still decides.

Evidence and argument earn recommendations. No model, score, judge, or manifest grants operational authority.

02 / Optional depth

Not every question needs a parliament.

Every top-level intelligence chooses the smallest honest route. The caller may override it.

D

Direct

Answer from sufficient current evidence.

fastU

Ultra

One mind builds an adaptive frontier.

deepC

Council

Independent minds attack and repair.

adversarialX

Decline / defer

Repair evidence, tools, scope, or authority.

honestreasoning-decision-service / routeC

Grounded Independent Council

High impact, unresolved competing options, or explicit Council request. Five isolated seats plus a separate judge.

Highestdeliberation costRequiredcontext independenceVerdictcompact boundary03 / Phase −1

Independent minds.
One shared reality.

“Blind” never means knowledge-starved. Seats share a checksummed fact packet while keeping hypotheses, probes, recommendations, and precommits private.

The non-voting evidence compiler

ScopePXLiveGoalieFreeze01 / QUESTION

Resolve exact scope

Project, question, cutoff, budget, forbidden assumptions.

02 / REPO

PX current truth

Docs, call paths, schemas, tests, freshness, contradictions.

03 / RUNTIME

Approved live reads

Current state when source and stale docs cannot answer.

04 / WORLD

Goalie research

Primary-source leads, prior art, counterexamples, projects.

05 / CUSTODY

Checksum packet

Facts, unknowns, boundaries, evidence horizon, pointers.

Scope boundary

The compiler resolves the exact question and evidence cutoff. It does not recommend a solution or score candidates.

04 / Phase 0

One mind.
An adaptive forest.

Ultra is deep single-agent reasoning—not six canned thoughts and not a hidden swarm.

Industry standardRepresent the boring path honestly.InversionReverse the assumption or objective.Destructive probeFind the counterexample that kills it.Constraint relaxationChange the boundary, not the answer.Analogy / prior artSearch for adjacent solved structures.Simulation / OOTWExperiment, then earn novelty.Five branches are a floor, not a quota. Ten, twenty, or fifty are valid while each adds material information. A failure must end in repair, substitute, experiment, kill, or explicit irrelevance—never critique-only.two+ autonomous waves · unequal depth · lineage requiredFrozen questionWhat decision survives evidence, inversion, and attack?Industry standardBaseline implementation and known tradeoffs.DestructiveCounterexample kills the favored path.Prior artAdjacent systems reveal a missing constraint.OOTW challengeNovel candidate must earn research and 30/40.SimulationDecisive experiment before irreversible work.InversionWhat if the object itself does not exist?Missed frontierWhat was avoided because it was hard?RepairFailure becomes a stronger branch—not a dead memo.W1W2W3stop on diminishing information05 / The executable court

A Council—or it says it is not.

The conductor dispatches the roster. It never impersonates the originator, five experts, Null Defender, repairers, skeptic, or judge in its own context.

−1 Evidence0 Ultra1 Precommit2 Reveal3 Repair4 JudgePHASE −1

Freeze shared facts.

PX, approved live reads, and Goalie compile one solution-neutral packet. Fact disagreements return here instead of becoming design theater.

Question-shaped evidence, not a repository dump.Explicit facts, contradictions, unknowns, and forbidden assumptions.Recomputed checksum before any seat consumes the packet.Evidence researcherBuilder / operator SkepticOOTW challenger Domain expertNull Defender · non-votingJudgeseparate context06 / Fail-closed identity

No more one mind
wearing five hats.

A good answer from self-play is evidence that Ultra works. It is not evidence that a Council happened.

Valid Council shape

Independent contexts, shared custody.

• Five distinct seat targets plus a separate judge
• Private precommit before candidate reveal
• Every seat authors its own probes and falsifiers
• Material issues receive targeted repair and skeptic retest
• Exact model/build/runtime receipts determine independence grade

Refuse or downgrade

Ultra is honorable. Fake Council is not.

• One execution context assigned to multiple seats
• Judge reuses an expert or originator context
• A SELF_ONLY seat spawns nested agents
• Provider changes exact model identity after pinning
• Fewer than five available targets → ULTRA or DECLINE/DEFER

Full independenceUnique contexts, exact receipts, two+ voting families.Degraded CouncilIsolated contexts, but family/build/instance proof is incomplete.UltraOne strong mind with adaptive adversarial lenses.Decline / deferMissing tools, evidence, budget, scope, or authority.07 / Provider-neutral execution

Real models.
Real fallbacks. Real names.

OpenAI-compatible and command adapters bind target IDs to actual executors. Silence, transport failure, or malformed first output rotates before pinning. Once valid, identity drift fails closed.

ORLive proven

OpenRouter

Structured JSON execution with returned exact model identity.

openai/gpt-5.6-solGQLive proven

Groq

Independent provider lane with system fingerprint receipt.

llama-3.3-70b-versatileQRuntime blocked

Local Qwen

Endpoint is visible; inference currently needs its GPU client reconnected.

Qwen3.6-35B-A3B · :8430>_Adapter ready

Command executor

Request over stdin, response over stdout, shell disabled, env allowlisted.

Claude · Codex · local toolsBEFORE PINtarget A→ silencefallback B→ malformedfallback C→ valid receiptPINNED08 / Compact boundary

The work lane gets the decision—not the transcript.

Rich reasoning remains inside immutable artifacts. Operational agents hydrate only what they need.

One immutable reasoning manifest with evidence, branch, issue, repair, and supersession lineage.One compact verdict with reasons, changed assumptions, dissent, risks, decisive test, and receipts.One Hippo pointer with authority=none; no self-promotion to Proven truth.Post-verdict self-evaluation is calibration evidence only. It cannot rewrite the ruling.council_result / immutableCOUNCIL VERDICTDEGRADED_COUNCILRulingProceed only through the bounded reversible pilot.WhyCross-examination repaired three material implementation defects.ChangedOriginator confidence 88% → final judge 82% after repair.DissentMinority risk survives: exact runtime-instance diversity remains unproven.Decisive testRun the same frozen case with independently proven model runtimes.OOTWRejected honestly; novelty did not clear the strict gate.ReceiptsEvidence SHA · model identities · cost/time · manifest digest · artifact pointer.09 / What is actually proven

Green for machinery.
Amber for identity limits.

The page carries the same rule as the harness: code, tests, packaging, provider execution, and local runtime availability are different receipts.

AllProvenDegradedBlocked Published · 116fd5f0Cold courts passed

Self-play refusal

Duplicate targets, reused contexts, judge reuse, and nested delegation are rejected or downgraded.

Cold courts passed

Adaptive Ultra

Autonomous waves, frontier lineage, negative-branch repair, unequal depth, and honest stopping are enforced.

Live provider smoke

Cross-model Council

OpenRouter GPT plus Groq Llama completed precommit, reveal, repair, skeptic retest, and separate judging.

Package proven

Immutable CLI

Evidence digest, validation, executor configuration, create-once output, package build, and command help passed.

Honest limitation

Provider identity

The live run lacked complete exact build/runtime-instance proof, so it returned DEGRADED_COUNCIL.

External runtime

Local Qwen inference

Health and model discovery respond; generation returns 500 until the GPU client reconnects.

72package tests passed43legacy courts passed5+1minimum seats + judge104slive cross-model run0fake Council seats10 / Use it

Ask normally.
Override when you mean it.

Council and Ultra are shared agent skills backed by the same executable package. The syntax changes; the custody rules do not.

Codex

Native registered skills

$council Should we…? Copy$ultra Think deeply about… CopyClaude Code

Slash commands

/council Should we…? Copy/ultra Think deeply about… CopyExecutable package

corepack pnpm --filter @foxflow/reasoning-harness dev council evidence-sha --input evidence.json
corepack pnpm --filter @foxflow/reasoning-harness dev council validate --input case.json
corepack pnpm --filter @foxflow/reasoning-harness dev council run \
  --input case.json --executors executors.json --out immutable-result.json

The operating principle

A full Council.
Nobody stops working.

Independent thought without factual starvation. Adversarial repair without memo theater. Deep reasoning without context sludge. One human-visible decision, with receipts.

Open the canonical contract Open the package guide Back to COUNCIL² V2Council Engine / standalone reasoning harness Source + tests + live provider smoke proven Recommendation only · operator final

PROBAGI / SOURCE PROOF

The page keeps its evidence attached.

source-proven

Exact decision-court source

The published article remains bound to the exact source HTML digest.

git:af6ba1fcbbbaac3714c37b41f9b6641649ecf00acourt-proven

Responsive publication court

The release gate covers structure, privacy, links, accessibility, and responsive rendering.

probagi:council-ultra:publication-court