Frequently asked questions

No. Prime Radiant is an analytical tool. We surface signals, score subnets, and synthesize AI-generated theses to inform your research. Investment decisions are yours. See our full Risk Disclosure for details on what Prime Radiant is and is not.

No. Prime Radiant is an independent product. We use Taostats and tao.app as data sources, but we have no formal relationship, sponsorship, or endorsement from the Bittensor Foundation, the Opentensor Foundation, or any subnet team. We are not a validator and we do not solicit delegations.

Absolutely. Email us at welcome@primeradiant.pro. We read every message and respond to most within 48 hours. We're actively building toward a public roadmap and community channels — if you'd like to be involved early, just reach out.

It's a reference to Isaac Asimov's Foundation — the Prime Radiant is a device that visualizes the predictive mathematics of psychohistory, turning the chaos of human history into legible structure. Same vibe here: take the chaos of Bittensor subnet data, run it through a team of specialized analysts, surface structure, turn structure into signal, turn signal into decision. Plus, the name sounds cool, and that matters too.

(Note: the Asimov reference is unofficial — we're not affiliated with the Asimov estate or any Foundation media properties.)

Yes. Every AI Board recommendation is persisted in our database with entry, target, stop, and conviction level. Outcomes are auto-resolved every four hours against actual market data — wins, losses, and zone transitions are all tracked. Our pattern library auto-verifies its similarity predictions at +7 days, +14 days, and +30 days against real subnet snapshots. We treat publishing misses as a feature, not a bug. See Product → H-003 for the gating logic.

Several safeguards. First, the Data Guard agent reviews every other agent's output before the Fund Director synthesizes a recommendation — it checks citations, internal consistency, and refuses analyses that don't pass review. Second, we run offline backtests with permutation null hypotheses and p-values to gate live wiring of any new signal. The H-003 signal, for example, fires in production only if its backtest verdict is ACCEPT and a kill-switch hasn't been engaged. Third, all numeric scoring (WPS, dereg risk, live-score gate) is deterministic — LLMs do not override these gates. Production safety is built into the architecture, not bolted on.

A composite health score from 0 to 100, where higher means healthier. WPS combines four sub-scores: distribution (how concentrated is ownership), cohort flow (are validators, whales, and owners net buying or selling), free float (how much supply is actually liquid versus staked), and dormancy (are previously sleeping coins waking up to sell). The composite is bucketed into five bands: healthy (80–100), stable (60–79), watch (40–59), caution (20–39), and danger (0–19). The Predictor downranks candidates whose subnets are in caution or danger bands. Full breakdown in our Product → WPS section.

A combination, depending on the task. The AI Board defaults to Google Gemini 2.5 Pro. Deep Research uses Perplexity Sonar Deep Research with strict JSON retry. The Knowledge Graph uses OpenAI GPT-5 nano. We route each task to the model best suited for it, with automatic fallback between providers if any one is unavailable.

Deregistration risk is a first-class signal in Prime Radiant. We score every subnet by pruning rank percentile and bucket them into four risk bands (critical, danger, moderate, safe). Subnets in critical or danger bands are filtered out of conservative recommendations automatically. Re-registration is detected through a three-path cascade — on-chain deregistration cache, price-crash signatures, and holder-count resets — and re-registered subnets have their pre-reregistration history excluded from pattern analysis. Full breakdown in our Product → Deregistration handling section.

Prime Radiant ingests data from two primary sources: Taostats, the official on-chain data provider for Bittensor (subnet pools, prices, emissions, holder balances, validator data), and tao.app, which provides subnet metadata and social analytics (Discord and Twitter sentiment scores). Both are public, official sources used widely across the Bittensor ecosystem. We refresh every four hours. Whale movement tracking uses our own 200-holder snapshots per subnet, captured directly from Taostats. Full pipeline in Product → Data sources.

Yes — a REST API is available on the Unlimited tier. You can fetch subnet signals, trigger AI Board analyses, and pull whale-pressure data programmatically. Authentication is via a personal bearer token you generate in your account settings.

Yes, anytime. No annual lock-in, no cancellation fees. You retain access until the end of your current billing period, then your account moves to read-only. Specific pricing and trial terms will be announced when subscriptions launch — see our Pricing page for the current tiers.

Yes — view-only. You add SS58 addresses to your account, and we read public on-chain data associated with those addresses to build your portfolio view. We never ask for private keys, seed phrases, or signing capability. We cannot move funds. We are an analytics tool, not a custody or trading service.