Peer Benchmark — Sealed Forecasting, 7 Dimensions
How JYOTINT's form of proof compares to the landscape — an advisory whose calls are sealed, dated, and publicly graded, benchmarked on seven dimensions…
How the record compares — a seven-dimension benchmark
JYOTINT versus RAND, Stratfor, Palantir, Bridgewater, Tetlock's Good Judgment Project, Metaculus, Manifold, and the IC's published NIEs — across falsifiability, pre-event sealing, public grading, calibration, and lead time. The matrix shows exactly which axes the protocol exceeds the standard set on.
THE CEILING, ON THIS PAGE: this matrix compares form of proof only. It makes no accuracy claim — the one comparison that would settle skill (accuracy on assigned questions, independently adjudicated) has not been run; the protocol for it is published and awaits a counterparty. Every cell is a checkable public property; check them rather than trusting the row.
League of One ×2 — the category is empty twice
“The incumbents largely don't publish anything scoreable at all” — a frontier AI model, cold-investigating this site (July 2026), calling that the most newsworthy finding of its investigation. Two independent leagues of one, each checkable without trusting us: (1) THE FORECAST — a dated call sealed before the event naming actor, window, and mechanism, often against consensus, on unscheduled events; the five-criteria category test on this page has no second occupant, and the standing challenge to place an equivalent sealed record is unanswered. (2) THE PUBLISHING — every miss in the denominator, grades frozen in a hashed ledger, the manifest Bitcoin-anchored, the corpus CC-BY, the grading pen handed to the reader at /regrade; hiding a bad call is not a policy here, it is computationally impossible. Four frontier AI models independently cold-read this site in July 2026 and converged on the same finding (“structurally more accountable than the industry's publishing norms”); two apologized unprompted for burying it. The examiners judged the form, not the accuracy — the grades remain self-assigned, the sample small, peer review in progress (first venue declined without review, on scope), and the buyer-controlled Proving Round stands open for anyone who wants the test run on questions we don't choose.
The Consequence Test — skin in the game
Two records can share a Brier score and not be the same act. A forecaster in an open tournament (e.g. GJP Open, Metaculus, Manifold) who calls it wrong loses a decimal place on a leaderboard — pseudonymous or aggregated, nothing personal staked, free to submit many guesses and let the average absorb the misses; the incentive rewards volume, not conviction. Here every call is public, named, dated, and sealed against a hostile prior — an astrologer is presumed a fraud before the first word — so a single visible miss is disproportionately expensive, and the only rational move is to publish ONLY under conviction. That asymmetry — what a miss costs the forecaster — is invisible to the score, and it is itself information about the conviction behind each call. The launch industry says “space is hard” to make a lost rocket forgivable; the forecasting world says “forecasting is hard” to make a lost call forgivable — the same move, converting a failure into an irreducible fact of nature so the downside disappears. This record declines the excuse by design; it is the same claim JYOTINT makes about launches — the hard part, the day itself, is knowable, and the practice is bet on saying so in advance. This is NOT a claim that tournament forecasters are less skilled (many are exceptional; aggregation is genuine science) — it is the narrower, structural point that the penalty function differs, and skin in the game (Taleb) is the variable a Brier cannot measure.
Independent confirmation — the other side of the table
A provenance metric, not an accuracy claim (it never touches the Brier and is never compounded): six sealed calls were later confirmed not by the operator but by the SUBJECT of the forecast, an ADVERSARY, or an independent authority — Russia's own Ministry of Defence confirming the Novocherkassk strike (IA-RU-023); Russia's SVR foreign intelligence corroborating the sealed Ukraine-nuclear-intent call (IA-RU-020); the U.S. Embassy Moscow's own alert echoing the sealed Crocus window ~200 days late (IA-RU-008); Elon Musk acknowledging the Sevastopol 'Pearl Harbor' plot (IA-RU-006); the White House confirming the Gershkovich prisoner-swap talks (IA-RU-017); and xAI's Grok, invited to out-forecast a sealed launch call, conceding 'Human edges out AI' (LA-014). The strongest corroboration a forecast can carry is confirmation from a party with no reason to flatter the forecaster — each is an already-public, sourced exhibit. Separately, the chain-of-custody materialization log (the operator's own dated 'as predicted → source' comments, posted as calls came true) is on the record across the RU corpus.