teranodeLIVE · 159 deliberations on file · 2 in the last 24h

Step inside the data.

The Chairman is the gravitational center. Four reasoner roles orbit it. Their pinned models are moons. Every eval run is an asteroid in the belt — brighter when it's recent, fading when it's deep history. Three perspectives — same system, different angles — and a time scrubber that lets you replay how the graph compounded to today.

How the Council works — in 5 simple steps

  1. 1

    You ask a question.

    The Council reads your question — just like you'd write a text to a smart friend.

  2. 2

    Five thinkers see it at the same time.

    The question goes to 5 different AI models all at once. They don't talk to each other.

  3. 3

    Each thinker does their job.

    Risk thinker portrait — a warmly-lit cosmic orb with a contemplative eye
    Risk
    asks "what could go wrong?"
    Builder thinker portrait — a slate-blue cosmic orb with a structured surface
    Builder
    writes the plan
    Strategist thinker portrait — an emerald cosmic orb gazing into the distance
    Strategist
    sees the long game
    Contrarian thinker portrait — a burgundy cosmic orb rotating against the prevailing direction
    Contrarian
    argues the other side
  4. 4

    The Chairman puts it all together.

    Chairman thinker portrait — a luminous golden cosmic star with a copper rim

    A fifth thinker, the Chairman, reads what the four said and writes one decision. Where the models disagree, the disagreement is preserved in the record.

  5. 5

    You get a timestamped, tamper-evident decision record.

    One artifact you can save, share, or file. With every thinker's voice on it — including the one who disagreed. Cryptographically timestamped so the record can't be quietly rewritten later.

Composing the system — frontier models, eval history, scenario field…

Live activity · what the system is doing

Anonymous structural telemetry from every deliberation — tier, domain, duration, whether the cross-examination round fired, whether the FINRA-examiner found a gap. No scenario text, no client data, no outputs. The reliability graph grows with every run. How we use this.

2
deliberations · last 24h · 7-day trend
0.7%
cross-exam fire rate · share of contested runs the chairman flagged for cross-exam
100.0%
decision-revised rate · when cross-exam fires, share where chairman changed the call
0.0%
FINRA-examiner gap rate · share of artifacts where the examiner flagged a regulatory weakness
Tier distribution · 111 total
  • Quick00.0%
  • Standard8072%
  • FINRA Battle-Tested3128%

The graph records every decision and every outcome.

Every Decision an external system asks the Council for — and every Outcome that comes back — lands in the same graph. Connected domains are anonymized pre-design-partner go-live (client-confidentiality posture); what crosses the wire is structure, not product detail. Reliability patterns will transfer between domains once enough outcomes accumulate to compute them.

9Decisions logged · structured records of every cross-domain run
3Outcomes captured · subject decisions, edits, post-hoc reviews
1 / 5Live / reserved domain slots · live = contributing in last 30 days · reserved = schema-provisioned, pre-go-live
Live9 decisions · 3 outcomesn=3 · accept-rate suppressed (need ≥10)2w ago
ReservedSchema-provisioned · pre-go-live
ReservedSchema-provisioned · pre-go-live
ReservedSchema-provisioned · pre-go-live
ReservedSchema-provisioned · pre-go-live

How to read this map

The bright center

The Chairman — the thinker that makes the final call.

The four orbiting planets

The four other thinkers — Risk, Builder, Strategist, Contrarian. Each runs on a different AI model.

The little moons

The actual AI models each thinker uses. Different brand for every role — no all-eggs-in-one-basket.

The asteroid belt

Every real Council run we've evaluated. Brighter dots are recent; fading dots are older.

Distant stars

Real questions the Council has answered. Click one to watch the Council think in real time.

Constellations

Groups of AI brands (Anthropic, OpenAI, Google, DeepSeek). Shows we don't lean on any one company.

Outer satellites

Each connected domain the Council serves. Active ones pulse copper; queued ones hold dim. Anonymized by design — the graph is the substrate, not a list of products.

Try a real question through the Council

Hypothetical examples for illustration of the Council's reasoning process. Not investment advice, not recommendations, and not based on actual client communications. Teranode does not provide investment advice. Past results of the Council's analysis do not predict future outcomes.

Why we put it in orbit.

A 2D graph reads as “Obsidian-default” — generic mind-map syntax. The Council's reliability story is not generic; it's the architecture itself, and its motion is the moat.

Every eval run is a real head-to-head we ran in production and judged on the record. The asteroid belt is what we have today. Drag the time scrubber back and you're replaying the sequence we lived through to get here. Drag it forward and the next dot is the next thing we ship.

The three perspectives are not just camera positions — they're lenses. Architect sees the design. Analyst sees the validation activity. Witness sits inside one role and looks outward. The Witness lens carries one line worth keeping: what you see depends on where you stand. The data is the same.

policy_version · 2026-05-09 · earliest eval ·

159deliberations · all-time
23deliberations · last 7 days
48A/B evals in the belt
5scenario stars
5 · 5roles · vendor clusters