Cred-RepoCRED-REPO

How to make a commitment

What is a commitment?

A Cred-Repo commitment is a falsifiable claim about a future outcome, made before the outcome is known, sealed cryptographically with your name, and scored against reality when it resolves.

It is not a hot take. It is not a prediction in the casual sense. It is a public record that cannot be edited, deleted, or quietly forgotten. The permanence is the point — it is what makes the credential meaningful.

Five things every commitment must have:

  1. A specific, falsifiable claim
    Not: "AI will transform healthcare"
    Yes: "FDA will approve an AI-assisted diagnostic tool for radiology by Q4 2026"
  2. A confidence level (0.0000 to 1.0000)
    Your honest estimate of how likely this is.
    0.5000 means you have no view. Be honest about that.
  3. A resolution date
    When will you know if you were right?
    Pick the earliest date a definitive answer could exist.
  4. A resolution source
    The specific public artifact that will prove the outcome.
    "EDGAR filing", "FDA Drugs@FDA record", "FOMC statement" — be exact.
  5. At least one falsifier
    What would prove you wrong?
    If you cannot name it, the claim is not falsifiable. Do not submit it.

Worked examples

Each example is a complete commitment in the same format as the commit form. This is what good looks like. Cred's current domains include NBA, Earnings (EDGAR), and FDA; the principles below apply across domains.

Example 1 — Finance

Title: Apple services revenue beats Q3 2026 consensus

Domain: Finance

Horizon: Medium

Claim: Apple Q3 2026 earnings confirm services revenue exceeded $26B

Confidence: 0.7200

Resolution: 2026-07-31

Source: Apple Q3 2026 earnings release via EDGAR / investor relations

Criteria: Services revenue line item in official earnings release > $26.0B

Reasoning: Services segment has grown 14% YoY for 6 consecutive quarters. Installed base expansion and price increases support continued momentum. Consensus sits at $25.1B — leaving room for beat.

Falsifier: Macro-driven consumer spending contraction reduces App Store revenue meaningfully in Q2, dragging services below $24B.

Example 2 — Macro

Title: Fed delivers at least two cuts before end of cycle

Domain: Macro

Horizon: Medium

Claim: Fed funds target range is 4.00-4.25% or lower by Dec 31 2026

Confidence: 0.6100

Resolution: 2026-12-31

Source: FOMC official statement at federalreserve.gov

Criteria: Target range upper bound at or below 4.25% in final 2026 statement

Reasoning: PCE trajectory and labor market softening provide cut justification by Q3. Market pricing reflects 1.8 cuts — slight underpricing.

Falsifier: CPI re-accelerates above 3.5% in Q3, removing political cover for cuts before year-end.

Example 3 — Policy

Title: EU AI Act high-risk provisions enter into force

Domain: Policy

Horizon: Medium

Claim: EU AI Act Article 6 high-risk system requirements apply to covered entities by Aug 2026 without further delay

Confidence: 0.5800

Resolution: 2026-08-15

Source: Official Journal of the European Union publication record

Criteria: No amending regulation or Council decision delaying enforcement date published in OJ before Aug 1 2026

Reasoning: Current timeline has provisions entering force Aug 2026. Political will appears stable post-election cycle. Industry lobbying has not moved the needle on timeline.

Falsifier: Member state coalition requests 12-month delay; Council votes to approve before June 2026.

Confidence levels

Confidence calibration is usually the most misunderstood part of the format. Use these anchors directly.

0.5000 You genuinely have no view. This is the maximum-uncertainty position. It is valid. Use it honestly.

0.6000 You have a slight lean. You'd bet on it but you're not confident.

0.7000 You think this is more likely than not, with real reasons. This is the most common meaningful starting point.

0.8000 You have strong conviction backed by multiple evidence sources. Most commitments should not be here.

0.9000+ You are highly confident. Ask yourself: what am I missing? If you cannot name a credible falsifier, lower this number.

1.0000 You are certain. This is almost never appropriate for a future event. If you submit 1.0000, you are saying there is no possible way you are wrong. The scoring will penalize this harshly if you are.

What the platform scores

Calibration Does your stated confidence match your actual outcomes? If you say 70% on 10 things, roughly 7 should resolve true.

Participation Are you engaging consistently within your subscribed domains? Cherry-picking easy calls is visible and penalized.

Update discipline When you update a probability, is it reasoned? Oscillating without new evidence degrades your discipline score.

Horizon Longer-horizon commitments contribute more to your profile. Short-horizon calls resolve quickly but prove little.

Not scored Whether you are right or wrong on any single commitment. A well-calibrated wrong call is better than an overconfident right call.

What does not belong here

Retrospective claims "Company X did well in 2025" — outcome already known. Commitment must be made before outcome is known.

Vague claims "AI will disrupt finance" — no falsifiable resolution. If you cannot define exactly what resolves it, it is not valid.

Self-defined outcomes "My strategy will succeed" — you define success. Resolution must be against an objective public source.

No falsifier If you cannot name what would prove you wrong, your confidence should be 0.5000 and you should reconsider whether you have a real view.

Institutional shields "Our team thinks..." — commitments are individual. Your name. Your call. No committee.

Your profile over time

The profile is not a scorecard. It is a diagnostic — a record of how you reason under uncertainty, accumulated over time, owned by you permanently.

A profile with 10 resolved commitments is early. A profile with 50 resolved commitments across three domains starts to show real signal. The value compounds with time and breadth. A single-domain record proves pattern matching. A cross-domain record proves reasoning ability.

Your profile travels with you. It belongs to you, not to this platform. You can export it at any time as a self-verifying bundle — a JSON file that proves every commitment was made before its outcome, signed by you, with the full update chain and resolution record attached.

Start your first commitmentTerms of ServicePrivacy PolicyBack to Cred