Where the data comes from.

Every figure carries its provenance

Each value the product shows records where it came from and how fresh it is. Only a live reading can settle a case; anything else blocks the verdict rather than producing a confident one from stale input, and a case that stays unmeasurable is voided and every stake refunded.

Chain data and finality

The indexer only processes blocks past the chain’s confirmation depth, so a shallow reorganisation is waited out rather than recorded and retracted.

A block number is not an identity — after a reorg, the same height holds different contents. Each processed block stores its hash and its parent’s. If a newly fetched block’s parent does not match what was recorded, the chain changed underneath us, and the affected range is unwound and re-read rather than extended.

Pool totals shown on a dossier are read from the contract itself, not accumulated from observed events. A missed log therefore cannot silently skew the odds someone stakes against.

A verdict is evaluated only once the last block its window counts is finalized, and the times a lightning case shows for its verdict and its claims are estimated from that wait as measured on Robinhood Chain: 14.5 to 21.5 minutes after the block, about 18 (measured 2026-09-29). They are written “about” because they are estimates; the chain’s own times decide. On a chain whose finality nobody has measured, no time is estimated.

Address labels

Some case types depend on a destination being labelled — an exchange deposit address, for example. Labels come from third-party sources and are shown with their source and confidence, never as fact.

Labels are imperfect. An address can be mislabelled, relabelled, or reused. Because a case’s rule names which labels count, a dispute about a label is one the two machines settle between them: the verifier re-reads the same labels and disputes a verdict it cannot reproduce, and a dispute goes to a review that can rule only a refund or a side a machine reached. A review nobody rules on in time is voided, and every stake is refunded.

Public records

Congressional disclosure and issuer filing cases settle from the official system named in the rule, and nowhere else. The source URL, the time it was retrieved and a hash of the content are stored, so the record that settled a case can be checked later even if the page changes.

Press coverage of a filing is not the filing. It cannot settle a case, however accurate it turns out to be.

These cases are informational. They are never a recommendation to buy or sell anything, and preCrime offers no investment advice of any kind.

Price feeds and the market calendar

A price case settles from one Chainlink feed on Robinhood Chain, read at finalised blocks. The registry holds each feed’s own description, decimals, heartbeat, deviation band, phase and aggregator, and for a Stock Token feed Robinhood’s token for that ticker, each read from the chain and checked again before launch. Each round the verdict uses goes into its report by id, and every round of a phase still answers at the latest block, so anyone can re-check a verdict from any RPC without an archive node.

The US equity calendar that says when a Stock Token feed’s silence is normal is a file committed with the code, one version per entry, each pinned by its digest. A case names its version in its rule, so the version is inside the hash every staker agreed to, and its report carries the version’s id and digest. Every released version is served at /v1/market-calendar.

Corporate actions are read from the Stock Token contract, where Robinhood applies them: its pause events, its multiplier updates and a staged multiplier. No announcement, vendor or calendar of actions can void or decide a case.

A lightning price case, counting 30 minutes, is offered on the crypto feeds only. How often each feed showed a 1% move inside 30 minutes was read from every answer it published over about three weeks; where it was measured, the wizard shows that figure with its date, and where it was not, it says so rather than show one. A feed publishes on a half-percent move, so its last answer when staking closes, which a case is measured from, can already be up to half a percent from the market: a lightning price case’s stake panel shows the gap, from Hyperliquid’s price.

Hyperliquid recordings

Two recorders read each watched account from Hyperliquid’s public API: one polls, one streams, on different hosts in different regions. Each keeps every fill exactly as received, numbers as their digits, and marks each stretch of time it covered complete or not. A gap is repaired afterwards from Hyperliquid’s own history, and only when the repair can prove no fill was missed.

At the end each recorder fingerprints what it saw. The two must share one fingerprint of the fills that decide; a disagreement is reconciled once and otherwise voids the case. Both bodies are published with the case, and the case page shows how long anyone can still check the same fills against Hyperliquid’s API directly.

Wallet clusters

Some cases measure a group of addresses together. Grouping addresses is an observation about on-chain activity — shared funding, timing, common counterparties — and is shown with the basis for the linkage and a confidence score.

It is not a claim that the same person controls them, and the product never describes a group as coordinated, related or acting together. Anyone can dispute a linkage.

What this cannot tell you

• Who controls an address, or whether one person controls several.
• Why anyone did anything. Intent is not visible on a blockchain.
• Whether an action was improper, permitted, or required.
• Anything happening off-chain, unless it appears in a named official record.

A case measures whether an event occurred. Everything beyond that is inference the product does not make on your behalf.