2026 product definition

Truth console · process is not world truth

Plug any model in as the brain. KELLY forces: ask → missing facts → forecast/answer → proof receipt → check-in → settlement → calibration → only then a governed memory proposal.

  1. 1Ask
  2. 2Facts
  3. 3Receipt
  4. 4Reality
  5. 5Score
  6. 6Earned memory

Sector labs are lenses. JobProof is paused. Brains are engines — not the product. Ops ceilings (Neon opt-in, soft-cap) are not densify thrash — after Railway ship: npm run accountability:post-deploy-hold.

Truth console · public scoreboard

Accountability passport

Live public Brier volume mapped to the loop. Process is not world truth. Memory remains gated.

  1. Ask
  2. Facts
  3. Receipt
  4. Reality
  5. Score
  6. Earned mem

Settled

0

Weather / Sports

0 / 0warming

Markets

0

Scale remaining

1000

SHA Posture vs_market cal Full scoreboard →
KELLY live pulserefreshed agolearninglifetime heartbeat 0browser pulses warmingcells 0replay-ready 0engine checkinglatest cell watchingevidence checkinglive memory laneloop: checking live activity
activity feed
Proof control room|public writes blocked|profession: proof auditor|kelly-anatomy-protocol-v1
Interactive lens

Proof auditor · Answered directly by KELLY · public writes blocked

Glass Lab · proof lens into KELLY

Proof console · why KELLY did this

Glass Lab is where all proof is accessible — intuitively layered and deep enough to show why KELLY did what it did. Scoreboard = settled world. Gen wall / debt = process. Answers are never self-truth.

Inspect live proof, truth debt, generation wall, and organ health. This lens does not write ground truth — KELLY remains the only write authority.

glass lab = proof consoleprocess ≠ world truthorgans empower abilityUIs = interactive lensesproof fidelity · not fake certainty

Every organ (Ask, Superforecasting, Hypothesis, Science, Memory) empowers a KELLY ability. Users can inspect and share forecasts with full proof records. That is complete trail fidelity — not a guarantee that predictions are always right. Reality still settles; Brier still scores; memory only after court.

Paid desk first-win moat (proof console)

Weather + sports desks toward 100 settled · process is not truth · total settled

Full scoreboard

Weather first-win

0/100
5 stable30 proof100 moat

Open 0 · to 100: 100 · Brier waiting

Open hazard desk

Sports first-win

0/100
5 stable30 proof100 moat

Open 0 · to 100: 100 · Brier waiting

Open sports desk

Engine → scoreboard bridge

Waiting for first weather feed settle tick (or rebridge)

No bridge run yet · Neon optional

Scoreboard

Feed ticks settle forecasts; bridge mirrors them into durable precommits for the public scoreboard. No prime memory write.

Glass Lab · why-proof console

Why KELLY did what it did

All proof lives here: intuitively layered, deep enough to inspect the full trail — claim, evidence, self-inquiry, precommit, settlement, organs, memory gate, share.

proof console

    Reality accountability scoreboard · paid product

    Loading scoreboard…

    Not bigger model — accountable to reality: predict, prove, settle, score, improve only when reality confirms.

    Plan gaps · honest densify · not densify thrash

    process_is_not_truth · not a launch claim · JobProof private MVP measured separately

    • Memory Court: status unavailable · Glass Lab memory
    • Via negativa: status unavailable · eliminations
    • L9 / OGI measurement: status unavailable · no public AGI marketing
    Sports deskHazard desk

    Weather first-win

    0/100
    5 stable30 proof100 moat

    Open 0 · to 100: 100 · Brier waiting

    Open hazard desk

    Sports first-win

    0/100
    5 stable30 proof100 moat

    Open 0 · to 100: 100 · Brier waiting

    Open sports desk
    Settled

    0

    Pending

    0

    Mean Brier

    waiting

    Hit rate

    waiting

    vs coin-flip

    waiting

    Proof lane: 0 / 30 settled for public calibration proof · Scale target: 1000 open+settled forecasts

    vs market

    pending link

    Market-implied baseline attaches when settlements carry it.

    vs expert / human

    pending dual-lane

    Dual forecasts settle both lanes separately.

    over time

    building

    Rolling windows after timestamped settlements densify.

    No settled trail yet. The engine room fills when forecasts settle against reality — not fixtures. Start with the forecasting journal.

    Organ share · full proof records

    Empower the organ. Share what KELLY forecasted — with proof.

    Each app is a KELLY OS organ with an interactive lens. You can share knowledge and forecasts with complete proof records. That is proof fidelity — not a guarantee of 100% predictive accuracy.

    ScoreboardGlass Lab bind

    Self-inquiry · answers are not truth

    Break the answer into new questions for reality

    Good loop: answer → question self → reality check → only proven lessons become memory. Bad loop: answer → believe self → reinforce.

    • · I answered this, but what did I assume?
    • · What part of my answer could be wrong?
    • · If I am right, what should happen next?
    • · What evidence would prove or disprove this?
    • · What follow-up question should I ask reality?

    Live from KELLY · memory provenance

    Generation wall

    Root cells, gen-1 revision children, and ancestry edges — inspected through Causalea lenses, owned by KELLY.

    live · neon
    Generation 0

    3,247

    Generation 1

    13

    Ancestry edges

    13

    Truth debt · process only

    Loading process debt…

    Truth debt tracks unfinished proof / product legs — not settled world outcomes.