Anyone can predict.
Nobody can delete.

An append-only registry of dated, falsifiable, probability-weighted claims — resolved automatically and scored in public, including ours. The scarce commodity was never prediction. It is track record.

—CLAIMS ON THE LEDGER

—AWAITING RESOLUTION

—PREDICTORS

0CLAIMS EVER DELETED

Verify it without trusting us

This is the whole point. Download the ledger, run the verifier on your own machine, and reproduce every score — including the ones that make us look bad. If any number here requires you to trust this server, that is a bug.

NO AUTH · NO RATE LIMIT · NO PAGINATION

$ curl -s https://verdict.raz0r.io/v1/export/claims.jsonl > claims.jsonl
$ python verify_export.py claims.jsonl

rows            : 1
chain           : OK
rehashed        : 1/1
withdrawn       : 0 (still scored — withdrawal is not deletion)

VERIFICATION PASSED

verify_export.py imports nothing from this project and talks to no server. It recomputes every content hash, walks the chain, and re-derives the scores from the rows in front of it.

What is load-bearing

These are enforced by the database, not by our good intentions.

APPEND-ONLY

Postgres triggers, not convention. A claim permits UPDATE on two lifecycle columns only; DELETE raises. A bug in the API cannot rewrite history.

SERVER-ASSIGNED TIME

A client-supplied timestamp is rejected with a 400, not quietly ignored — so anyone who thinks they are backdating finds out immediately.

RESOLUTION DECLARED FIRST

How a claim resolves is inside its content hash, fixed before the outcome is known. It cannot be edited afterwards to fit the result.

WITHDRAWAL ≠ DELETION

A withdrawn claim stays public and stays scored, flagged with a timestamp. Changing your mind is free. Erasing your prior view is not.

NO CERTAINTY

Probabilities are open-interval: 0 < p < 1. We will not host a claim of certainty, and log scoring is undefined there anyway.

SKILL, NOT SCORE

Raw Brier is never shown without Brier skill beside it. Predicting p=0.98 on near-certainties gives a beautiful Brier and zero skill. A test fails if that ever changes.

What this is not

Not a prediction market. No stakes, no money, no tokens — which keeps it clear of FCA and CFTC regimes entirely.

Not a debate platform. No comments, no votes on claims.

Not a reputation score for people. Scores attach to predictors within a domain.

Not a news product. We do not editorialise on outcomes.

Why cyber intelligence first: no major threat-intel vendor has ever published a Brier score. The entire industry sells forward-looking assertions with no scoring layer at all. A registry seeded with dated ransomware and exploitation claims is both a differentiator and an indictment of the category.

Anyone can predict.Nobody can delete.