Prediction Integrity · Receipts
Certified Predictions
Pre-outcome proof for individual AI forecasts. A certified prediction binds the forecast payload to a SHA-256 fingerprint and an Ed25519 signature — independently verifiable by any party.
The only meaningful time to certify a prediction is before the outcome is known. Once the result has been observed, a record produced afterward cannot prove what the system claimed in advance.
What a certified prediction contains
The prediction payload is captured in the signed certificate. Every field is included in the canonicalized bytes that produce the hash.
{
"artifact_type": "prediction",
"artifact_schema": "prediction_receipt.v1",
"source": "betfire-ai",
"prediction_id": "pred_01j9k2m...",
"canonical_hash": "sha256:a3f9b2e1c847...",
"model_version": "ev_8.2",
"predicted_at": "2026-05-14T12:03:24Z",
"context_summary": "match outcome forecast — entity match_8451"
}The signature covers the canonicalized payload (RFC 8785 JSON Canonicalization Scheme). Any modification to a field after issuance — including artifact_type — invalidates the signature.
Verifying a certified prediction
Any party can verify a prediction certificate independently. No account, no API call to the issuer.
Retrieve the certificate by its certificate_id from GET /api/cert/:certId/manifest or GET /api/verify/:certId.
Canonicalize the prediction payload using RFC 8785 JSON Canonicalization Scheme. Compute SHA-256 of the canonical bytes.
The recomputed hash must match the canonical_hash recorded in the certificate. If it does not, the prediction has been modified since certification.
Fetch the public key from /.well-known/signing-keys.json. Verify the Ed25519 signature over the canonicalized certificate payload using the public key.
Confirm that predicted_at in the signed payload predates the known outcome. This is what makes the proof pre-outcome.
Where certified predictions fit
Sports analytics, financial forecasting, weather prediction, event forecasting — any platform publishing model-generated forecasts.
AI agents making predictions, classifications, or escalation decisions in operational workflows.
Model evaluation pipelines, public leaderboards, and AI forecasting benchmark studies.
Threat intelligence systems publishing AI-generated risk scores or incident alerts before resolution is known.
Risk scoring and classification systems that need an auditable record of the score produced at decision time.
Public forecasters who need to prove they did not revise predictions after the outcome.
Common payload patterns
The prediction_receipt.v1 schema is intentionally minimal — most fields are optional. Different prediction types add different metadata inside the same envelope.
Predicts the outcome of a single sports event. The context_summary identifies the match. The model_version binds the prediction to a specific model checkpoint.
Credit-risk score or fraud assessment at decision time. The payload typically includes a decision_threshold field alongside the predicted score so the bind covers both the score and the policy that interpreted it.
An autonomous agent makes a routing or classification decision. The payload captures the input context, the chosen output label, and any confidence score. The certificate becomes part of the agent's audit trail.
A prediction-market platform records the AI-generated probability or odds at a specific timestamp. The receipt becomes a verifiable snapshot of what the model claimed before the market resolution.
Frequently asked questions
What is a certified prediction?
A certified prediction is a signed, structured record proving that an AI forecast existed at a specific moment in time. It contains the canonical hash of the prediction payload, an Ed25519 signature, a timestamp, and metadata identifying the source and model version. Any party can later verify that the prediction has not changed since certification.
Why does pre-outcome timestamping matter?
Once the outcome is known, a record produced after the fact cannot prove what the system claimed in advance. The whole point of prediction integrity is to bind the forecast to a moment before the result is observable — that is what makes the proof meaningful to a third party.
Does certifying a prediction prove it was correct?
No. Certification proves the record existed and was not altered. It does not evaluate correctness, accuracy, or fairness. Those properties require separate evaluation processes. Certification is a proof of provenance, not a quality judgment.
Can I certify predictions from any AI model?
Yes. You provide the prediction payload; CertifiedData computes the hash, signs it, and issues the certificate. The model_version field is captured in the payload metadata if you want to bind the prediction to a specific model artifact certificate.
How does this differ from a logged prediction?
A logged prediction lives in a system log that can be modified. A certified prediction is a tamper-evident artifact bound to its hash and signature — any modification to the prediction payload or the certificate breaks the binding. Certification is stronger evidence than logging for high-stakes forecasts.
Machine-readable summary
{
"concept": "Certified prediction",
"concept_type": "prediction-receipt-primitive",
"canonical_url": "https://certifieddata.io/prediction-integrity/certified-predictions",
"parent_concept": "Prediction Integrity",
"related_concepts": [
"Daily prediction manifest",
"Pre-outcome proof",
"AI artifact verification"
],
"artifact_schema": "prediction_receipt.v1",
"certificate_type": "GENESIS",
"signing_algorithm": "Ed25519",
"hash_algorithm": "SHA-256 (RFC 8785 canonicalized)",
"payload_fields": [
"artifact_type",
"artifact_schema",
"source",
"prediction_id",
"canonical_hash",
"model_version",
"predicted_at",
"context_summary"
],
"issuance_endpoint": "POST /api/notary/create",
"verification_endpoint": "GET /api/verify/:certId",
"public_key_url": "https://certifieddata.io/.well-known/signing-keys.json",
"claim_boundary": "Certification proves the record existed and was not altered. It does not evaluate correctness, accuracy, or fairness."
}Certify your first prediction
Submit a prediction payload and receive a signed certificate in seconds. Any party can verify it without trusting the issuer.