Collect
InputsOrder book and volume from exchanges, holder and wallet activity on-chain, public exchange data, community signal from X, Telegram, and Discord, and what the research and news cycle is saying.
Machines are better at logic and data. People are better at trend and feel. Annapurna is built on that division.
Hard signals
The model handles what can be measured.
Liquidity
Order book depth
Holder distribution
Volume quality
Human signals
We handle what has to be judged.
Community
Narrative
Mission
Market trend
What we measure, and how we measure it.
Depth at range
Bid-side depth within a fixed band of mid, measured per venue rather than aggregated across them.
Volume-to-depth divergence
Volume a book cannot support is flagged, not counted.
Holder concentration
Distribution across wallets, with treasury and exchange addresses excluded.
Float against FDV
What is actually tradable, against what the market is being asked to price.
Attention and liquidity gap
Where interest runs ahead of the depth to absorb it.
Measure
MetricsRaw inputs become comparable metrics: depth at range, volume quality, holder concentration, and the gap between attention and liquidity.
Judge
Model and judgmentThe model reasons over the numbers. We weigh what it cannot: community, narrative, mission, and where the market is actually turning. Neither side decides alone.
Report
OutputA written report on where the project stands, what the market is pricing, and what would have to change.
Each read ends in a written report, sent to the team it is about.
Want Annapurna's read
on your project?
Contact →
