How the station works.

01What Phelorus is

A pelorus is the dumb instrument on a ship's bridge that takes a bearing relative to the vessel — useless on its own, indispensable once you check it against something fixed. This station is the same idea. It continuously discovers Solana pairs, reads each one from 0 to 10 against twelve rules drawn from live data, and — the part that matters — returns six hours later to check whether it was right, then moves the weights that were wrong.

02Where the data comes from

Two public sources, no keys and no scraping. DexScreener supplies market state: the deepest Solana pair, its liquidity, 24h volume, age, price change, and hourly buy/sell counts. A Solana RPC node supplies the hard signals — whether the mint authority and freeze authority are still live, read directly from the mint account.

03An unread rule is never a zero

Upstreams omit fields. DexScreener returns no liquidity object at all for very fresh pairs. A station that reads that absence as "$0 liquidity" would condemn every new token as thin water and call it analysis. Phelorus instead marks the rule unread, excludes it from the depth, and publishes coverage — the share of rules it could actually read. A low-coverage sounding says so on its face, because a sounding taken from a guess is worse than no sounding: it goes on the chart looking exactly like a real one.

04The twelve rules

Each is a plain rule with a weight and hard bounds. Nothing is a neural net; every depth can be re-derived by hand from the deviation card.

05The three bands

DEEP WATER at 7.5 and above, SHOALING between 4.5 and 7.5, FOUL GROUND below 4.5. A band is a summary of the depth, nothing more — the rules underneath it are the actual output.

06How outcomes are checked

Six hours after a sounding the pair is fetched again. Liquidity gone, or down more than 70%, is FOUNDERED. Down 30–70%, or price halved, is TAKING WATER. Anything else is AFLOAT. If the upstream fails at bearing time, the sounding simply stays unchecked — a failed fetch is never recorded as an outcome.

07How weights move

For each rule Phelorus asks one blunt question: when this rule fired, how often did the pair actually go bad, compared to the base rate across everything? A penalty that fires on worse-than-average outcomes has earned its weight; one that fires on better-than-average outcomes is too harsh and loses some. Bonuses are judged the same way, inverted. A weight moves at most 0.15 per swing, needs a minimum sample of 12 before it moves at all, and can never leave the bounds declared in the source. Every change is logged with its reason on the deviation card.

08Why a sounding on request is not filed

Anything you sound yourself is computed with the same rules and the same sources, but it is never written into the log book. If it were, anyone could pad the station's record with easy pairs and the published error rate would quietly become a marketing number. The error rate is only worth something if the sample is the station's own choosing.

09What this is not

Not advice, not a signal service, not a prediction. A depth is a reading of public data at one moment under one swing of the compass. Phelorus publishes how often its own readings turn out wrong precisely so the number is never mistaken for a promise. Read our error before you trust a depth.