What Prime Radiant actually does

When you want a complete analysis of any subnet, the AI Board runs the full team in parallel. Five analysts dig into their specialties — OnChain reads every snapshot, Fundamental runs Deep Research across 40 quality questions with cited sources, Sentiment watches Discord and Twitter, Graph maps team and partnership connections, Technical computes price structure for entry / target / stop. Their findings go through Data Guard, our skeptic — it validates every citation, flags hallucinated claims, and refuses analyses that don't hold up to scrutiny. Then Fund Director synthesizes everything into one decision: buy / hold / cut, with concrete entry, target, stop-loss, and a conviction score from 1 to 10. Every recommendation is persisted and auto-resolved against real market data every four hours — wins and misses, both published.

• Parallel analyst execution — Five specialists work simultaneously, full session typically completes in 60–90 seconds.
• Data Guard validation — Refuses outputs that fail citation or consistency checks before they reach you.
• Persistent track record — Every recommendation auto-resolved against real market data — track record is measurable, not anecdotal.

The failure mode of LLM-based subnet analysis is paying for tokens on candidates that should never have been considered. Prime Predictor scans every active subnet through a verified historical pattern dataset and ranks by accuracy, not raw confidence. Each survivor passes through six gates: dereg-risk gate filters subnets near deregistration; live-score gate drops anything with weak Onchain / Fundamental / Sentiment for the chosen risk profile (Conservative / Moderate / Aggressive); cooldown gate skips subnets recently analyzed without meaningful price movement; whale-pressure gate downranks subnets in caution or danger bands or eliminates them if top holders are actively dumping. Only the top five survivors are shown to you, with effective scores, whale chips, dereg badges, score chips, and cooldown context. The run pauses for manual approval; no LLM cost is incurred during the pause. After approval, the full 7-agent AI Board analysis runs on each approved candidate via SSE streaming, cancellable mid-stream.

• Six independent safety gates — Dereg-risk, live-score, cooldown, whale-pressure all filter before you spend a token.
• Manual approval pause — Zero LLM cost during the pause. You approve before the AI Board runs on candidates.
• SSE streaming with cancel — Watch the AI Board reason in real time, cancel mid-stream with no orphan calls.

Whales rotate out before they dump. Cohorts shift before prices move. Whale Tracker turns the top 200 holders of every subnet into a single health metric: Whale Pressure Score, 0–100, where higher is healthier. Four components feed the composite. Distribution Score measures ownership concentration via HHI and Gini with top-1 share as tiebreaker. Cohort Flow Score tracks signed net flows across five wallet cohorts — validator, whale, owner, retail, unknown — over 24h, 7d, and 30d. Free Float Score captures the health of staked-versus-liquid supply. Dormancy Score detects long-dormant coldkeys waking up to sell, using ADD z-scores. The composite drops into five bands: healthy / stable / watch / caution / danger. Each band automatically influences Predictor candidate scores and AI Board confidence.

• Cohort labeling that doesn't lie — Validators identified by owner-coldkey registry, not delegation patterns; owner-team coldkeys auto-resolved.
• Team Stake metric — See the absolute TAO position held by the subnet's owning team.
• Danger band Telegram digest — Twice daily, every subnet in danger band is flagged proactively.

Add view-only SS58 addresses — no private keys, ever — and your portfolio becomes the AI team's workspace. Every position is tracked at the lot level: each (netuid, hotkey) pair maintains its own lots, partial sells compute realized P&L correctly, the open lot tracks unrealized P&L with weighted-average cost basis. The Portfolio Advisor agent looks at your full portfolio in context and produces an action plan: specific buy / sell / hold actions per position with reasoning, as toggleable to-dos. Price-zone alerts auto-route to Telegram when a position crosses critical levels. Every AI Board recommendation made for your wallets is tracked at the AI Track Record view — outcomes auto-resolved against real market data. You see what worked, what didn't, and where current price sits relative to the original entry / target / stop.

• Lot-based realized P&L — Partial sells handled correctly, no cost-basis confusion.
• Rebalancing action items — Specific actions per position, not vague advice.
• Per-wallet track record — See how the AI team performed on your actual capital, not a backtest.

Bittensor isn't just a list of subnets — it's a network of teams, projects, and dependencies. The Knowledge Graph turns that network into a queryable structure. Entities cover subnets, teams, individual contributors, projects, and partners. Edges describe relationships like founded by, invested in, depends on, competes with. Every entity property and every edge has a source link, pulled from Deep Research outputs and curated mentions via OpenAI GPT-5 nano. The graph rebuilds weekly with versioned snapshots so you can see how the network shifts. The Graph Analyst agent inside the AI Board uses this structure to flag partnership concentration, dependency risk, and competitive overlap when scoring any individual subnet.

• Cited edges only — Every relationship has a source link — no edge in the graph is unsupported.
• Weekly versioned rebuild — Snapshots let you see how the Bittensor network shifts week over week.
• Wired into AI Board scoring — The Graph Analyst uses this structure to flag concentration and dependency risk.

The Dashboard is where you start. Every active Bittensor subnet on one screen, with 12 sortable columns — onchain score, fundamental score, sentiment score, WPS band, dereg risk, price, 24h flow, 7d flow, APY, holder count, market cap, last analysis age. When you've added view-only wallets, your positions are highlighted inline so you can see at a glance which subnets you hold are scoring well and which are flashing risk. Every signal on the Dashboard is the same one feeding the AI Board and the Predictor — no hidden second tier. Free during early access on Watcher tier; full per-column drill-down on Pro and Unlimited.

• 12 sortable columns — Onchain, fundamental, sentiment, WPS, dereg, price, flows, APY, holders, mcap, last-analysis.
• Inline wallet context — Your positions highlighted on the same screen as every signal — no app-switching.
• Same data the agents see — Dashboard isn't a different view — it's the agents' input surface.