Live ledger — real records, collected so far.

TIERS AND API

Access

Prediction precision does not scale with price — freshness does. A paid user genuinely sees a tighter interval, because theirs recomputed twenty minutes ago with a new registry hit folded in while the free user’s is from last Tuesday. Same honest model, same honest uncertainty, and the gap widens exactly when it matters most.

Free
$0

Corpus and correction. Free users report bad entity links and wrong dates, and every claim card carries a report control that routes to the review queue.

RECOMPUTE

Weekly snapshot

Start free
Pro
~$14/mo · placeholder

Real-time unresolved claims, the full trail, and predictions that recompute daily rather than weekly.

RECOMPUTE

Daily

Request Pro
Analyst
~$79/mo · placeholder

Feature attribution on every prediction, the full source register with scores, exports, and the API.

RECOMPUTE

Hourly

Request Analyst
TerminalSOLD, NOT SELF-SERVED
from ~$600/mo · sold, not self-served

Live recompute on every new piece of evidence, the full posterior, final-chapter windows, and private sources. Sold to licensors and rights agencies, where per-seat value is measured against a six-figure acquisition decision.

RECOMPUTE

Live on new evidence

Request Terminal
GATE ONFREEPROANALYSTTERMINAL
Recompute frequencyThis is the ladder that mattersWeekly snapshotDailyHourlyLive on new evidence
Prediction horizonNext event onlyNext event6 months12 months, full pipeline
DriversFeature attributionAttribution + full posterior
Serialization riskRisk bandFinal-chapter window
Unresolved claims7-day delayReal-timeReal-timeReal-time
AlertsEmail, daily digestEmail + push, instant+ webhook+ Slack / Teams
Source registerGrades onlyFull, with scoresFull + private sources
ExportCSV / JSONBulk + S3 delivery
API10k req/mo1M req/mo, SLA
Seats1115+
Trail viewFirst 3 nodesFullFullFull
Watchlists5100UnlimitedUnlimited
PagesTerminal · Sources · CalibrationTerminal · Sources · CalibrationTerminal · Sources · Calibration
WHY THE LADDER IS SHAPED THIS WAY

The naive implementation — show free users a deliberately widened interval and paid users the real one — breaks the second discipline rule outright. If displayed intervals do not match the model’s actual uncertainty, the calibration page is a lie for at least one tier. It is also the kind of thing that gets discovered, because someone will eventually compare two accounts.

THE FREE TIER'S REAL JOB

Corpus and correction. Free users report bad entity links and wrong dates, which is why every claim card carries a this is wrong control that routes straight to the review queue. Entity resolution across four scripts is the highest-severity risk in the system — it silently corrupts everything downstream — so that queue is worth more than the subscription revenue the tier forgoes.

TERMINAL

The gap between Analyst and Terminal is deliberate and large. Terminal is sold, not self-served, to licensors and rights agencies — it is where per-seat value is measured against a six-figure acquisition decision, and where private sources and bulk delivery become the point.

APIREST + JSON
  • ClaimsThe ledger, paginated, each claim with its computed confidence and code
  • TrailsThe full provenance graph behind a claim
  • EntitiesThe catalogue, with every name form
  • SourcesThe source register, with grades and scores
  • PredictionsIntervals with their coverage and drivers
  • CalibrationReliability diagrams and Brier scores, per model

Cursor pagination. Keys carry per-tier token buckets. Resolved claims export in a schema.org ClaimReview-compatible shape alongside the native format — it costs a day and buys interop with fact-checking and news tooling. Access details come with the key.

WEBHOOKSHMAC-SIGNED
  • claim.state_changed

    Fires on every append to claim_event.

  • prediction.updated

    Fires when a recompute moves a point or interval.

  • watchlist.triggered

    Fires when a watched entity's claim changes state.

HMAC signatures and exponential-backoff retry. Grey-class signals never appear in a webhook payload, an export, or the API — their existence can move a confidence score, but their content leaves no trace outside the evidence bucket.