Build log
Last updated 5 August 2026
Vane is being built in the open. That means the mistakes are published too — which is the
point, since a tool that claims to tell you what really happened has to be willing to show
when it got something wrong.
Working now
- A full token dossier from chain data alone in about 69 RPC calls — deployer age,
funding, launch-block capture, current posture. No API keys.
- Token-creation indexing across arbitrary block ranges, with launchpad venues
discovered from the data rather than hardcoded.
- Outcome resolution — take tokens indexed at birth, resolve what happened to them later.
- Trade-tape reconstruction: per-pool liquidity over time, and the single largest extraction.
Not working yet
- No site-side lookup. The search box on the front page is not live; the tooling
behind it runs on the command line.
- Two known bugs — see 03 and 04 below. Nothing gets published as a report until
they are fixed, because both produce confident wrong answers rather than errors.
- No historical backfill. Public RPC serves roughly 72 hours; anything older needs the
Old Faithful archive, which is free but not yet wired up.
- No per-token scoring, and possibly never. At n=47 one signal was directionally
right, one was null, and one came out backwards. Aggregates need no prediction, so those
ship first.
What has been measured
74% of Solana tokens are silent within 24 hours of launch — 35 of 47, median 24
lifetime transactions, "silent" meaning no trade in six hours. One 24-hour window, small
sample, stated with its size. Bigger ones will be published with intervals.
Also: 100% of token creations are inner instructions; roughly 50–87k mints a day,
which is about 4–6 GB a year of index — small enough for one person to own; and transaction
count is a terrible proxy for volume, since one token showed 260 transactions and three
actual trades.
Bugs found while building
Every one of these returned a plausible wrong answer rather than throwing. That pattern is
the reason the calibration discipline exists at all.
01
Every token creation is an inner instruction
Fixed. A top-level scan returns zero on a chain minting tens of thousands of tokens a day — and zero looks plausible
02
Pump.fun bonding curves hold native SOL
Fixed. Invisible to token-balance parsing; the parser reported "no trades found" on the most common launchpad on Solana
03
A token trades on more than one pool
Open. Merging them produced "66.7% of liquidity removed in one transaction" out of a two-lamport dust trade
04
The fee payer is not the creator
Open. Launchpads sponsor transactions, which inverted a signal outright — and "this wallet created 47 tokens" is a claim about a person
← back to vane.report