01

Observe the configured token

The direct RPC observer reads a bounded recent window on the selected chain; the persistent indexer adds durable history, paper state and inference jobs. Both use the operator’s selected token and verified protocol contracts. It records mined trade events, token direction, normalized amounts and their source block. The Pons V2 adapter covers its curve and derived graduated pool; a separate adapter supports an explicitly selected compatible V3 pool.

Public receipts identify the chain, transaction hash, log index and block hash. The system does not infer the real-world identity of a trader. No token or network is silently borrowed from another project.

02

Construct the experience

A bounded window of observations becomes a generated screen recording, trade tones and factual narration. The worker renders actual video and audio files. The input explains the market in the configured quote asset rather than assuming every value is US dollars.

The same window also provides the source data for the on-screen stimulus preview. The complete job receipt identifies the exact generated media by its hash.

03

Run the Meta model

The GPU worker loads an explicitly configured TRIBE installation and checkpoint. It extracts events from the rendered media through the official interface, runs the model, validates the output dimensions and stores a prediction with its software and data provenance.

If the worker is unavailable, the model lacks access or output validation fails, the website does not invent a replacement prediction. A live market feed may continue while the cortical view waits.

04

Publish a visible result

The read API delivers the published result and its anatomical surface to the browser. The cortex colors update from the returned model values. A response trace and input-age label help distinguish current output from a delayed result.

Each completed result enters the cortical atlas with summary statistics derived from the immutable job input. Visitors can reopen that epoch, copy its direct result link, download the complete JSON receipt or export a visual receipt with the current 3D surface and public hashes.

Market updates can be frequent while model inference remains asynchronous. The moving camera is independent of both. Each state is labelled so visual motion is not confused with neural change.

05

Compare paper policies

PolicyWhat it usesWhen it acts
ObserverMarket observationsHolds cash without simulated orders
MomentumA 30-second market comparisonAn authored threshold and cooldown
CorticalMomentum plus predicted-response-change magnitudeOnly after a real, fresh prediction has been published

All fills remain simulated. Fees, slippage, cash, position and cost basis are accounted for by the paper engine. Simulated returns are not a promise about live execution.

06

Inspect and reproduce

Open a market event to inspect its receipt. Export a session to retain the displayed observations and any available prediction. The source repository documents model versions, receipt fields, queue behavior and the paper rule.

A chain reorganization can invalidate source events. The indexer’s rollback logic removes affected jobs and results and restores the corresponding paper state. Provider-specific recovery behavior still needs acceptance testing on the selected live deployment.