docs: expand Pump.fun and the Solana launchpads, add sniper detection

[buddies2705]

Pump.fun is the most asked-about subject in the support channel, and the panel was showing six queries for it.

The cap was hiding working queries

The corpus holds 63 Pump.fun queries. 12 were shown. 31 more were verified working and hidden by a per-venue cap of six.

The cap is now set from demand instead of uniformly — Pump.fun 16, Raydium 12, LetsBonk.fun / Meteora / Four Meme / Uniswap 10, PancakeSwap 8, everything else unchanged at six.

Pump.fun goes 6 → 15 queries and 6 → 15 streams, surfacing work that was already written and tested: graduating tokens in the last five minutes, tokens created by an address, creator fee transfers, PumpSwap trades and pools, market-cap monitoring, King of the Hill, and the 95–100% bonding-curve stream.

One genuine gap

Telegram asks "how do I find snipers" repeatedly — sniper criteria, "trying to find snipers", "pair this with pumpfun buys". The corpus had zero queries matching sniper or snipe.

Added first buyers of a token, in ascending time order, which is what sniper detection is actually built on: who bought in the first seconds, how much, and at what price. Written against a live Pump.fun token and executed before saving.

Near-duplicate filter

Raising the caps surfaced three variants of "top pump fun tokens by market cap" that differ only in wording. Titles that normalise to the same string now collapse to one, so a bigger cap adds coverage rather than noise.

llms.txt

New Solana launchpads and memecoins section, plus Hyperliquid and Phoenix perpetuals under protocol APIs. Every docs.bitquery.io link in the file was checked for a 200.

Verification

733 query entries across 32 groups, 412 stream entries across 17.

gate	result
play-link slug + tooltip on every entry	pass
slug resolves, not deleted, not unpublished	pass
zero BalanceUpdates	pass
entry on the chain its group claims	pass
stored endpoint matches query dialect	pass
no V1 query on a migrating chain	pass
no duplicate or near-duplicate titles	pass
every entry executed successfully, live	pass
no unexplained drops vs the original 298	pass
non-ASCII in any panel comment	0 of 1,141
widest comment line	78 chars
Docusaurus production build	success

The 17 newly surfaced subscriptions were executed over WebSocket before shipping; none had a prior verdict, and the audit refused to pass until they did.

Unrelated finding worth flagging

docs.bitquery.io has not been redeployed since #251 and #252 merged. The IDE reads refs/heads/main directly so the panel is current, but the docs website still serves the old build — /docs/start/bitquery-for-ai/ returns 404 and the live llms.txt lacks the entries added in #251. Nothing to fix in this PR; the deploy just needs to run.

🤖 Generated with Claude Code