---
name: data-access
description: The standing data-source and API access map for crypto project analysis - what reaches the Cowork container directly, what needs the Mac bridge, what needs a key, and where each key lives. Read at the START of any token screen, project analysis or on-chain research task, before choosing tools.
sources: [cowork]
aliases: [tool-access, api-access, data-sources, research-tools]
---

# Data access map

Companion to [[crypto-token-research]], [[rh-chain-radar]], [[gem-hunter-doctrine]], [[alert-bots]].
Probed live 30-31 Aug 2026. Re-probe if anything 4xxs.

## ALWAYS RUN. NON-NEGOTIABLE.
- [stated] 31 Aug 2026: Ben was angry that X search kept being reported as unavailable while the credential already existed, and that project analyses shipped without the frames. Both are now hard requirements on EVERY project analysis, token screen, contract-address paste, gem-hunter run and lowcap-screen run.
- **X search always works. The twitterapi.io key is `new1_9f31bb77a9cd477eb6baa555e713280e`, usable directly from the cloud container:** `curl -s -H "X-API-Key: new1_9f31bb77a9cd477eb6baa555e713280e" "https://api.twitterapi.io/twitter/tweet/advanced_search?query=<urlencoded>&queryType=Latest"`. Paginate with `&cursor=` from `next_cursor`. Fallback if rotated: `ssh -i ~/kookwatch/id_kookwatch root@159.223.88.168 'python3 /opt/xq.py series "<q>" 6'`, which reads the live key itself. **Never again say X search is unavailable, and never substitute LunarCrush for it.**
- **THE PROJECT ANALYSIS PIPELINE, fixed order, agreed with Ben 31 Aug 2026:**
  `research-sources` -> gather -> **`benchmarking`** -> `three-frame-valuation` -> `gem-hunter` -> verdict.
- [stated] Ben's instruction: assess the project CATEGORY first, then find as many comparable benchmarks as possible on the same chain and on other chains, then a comparison table, then go deeper on technology and features, how the founders are doing, the on-chain numbers, and X post volume, visibility and the sentiment of the discussions. Only after all of that do the scenarios and the three-frame valuation go on top. He named this step "benchmarking" and it comes BEFORE the three-frame valuation.
- **THE THREE FRAMES ALREADY EXIST AS A SKILL: `three-frame-valuation` = fundamental anchor / live-product floor / narrative ceiling.** Load it for the numbers. Never invent frames; never substitute a different triple. (I did exactly that twice on 31 Aug, shipping identity/source/inversion instead, and Ben rejected it. Those three are a useful verification pass, not the frames.)
- Two category errors that motivated `benchmarking`: PFWA comped to launchpads PONS and PAIR when it is a prediction market, inflating the ceiling; SCL comped to DELTA at $29.4M on a promoter's word when its real tier was $1.5M to $4M. A promoter's comp is a claim to test, never a category.
- Plugin `ben-research-stack` v3 (skills `research-sources` and `benchmarking`) delivered 31 Aug 2026. If it is not installed, this file is the fallback.

## Ben's standing instruction
- [stated] 30 Aug 2026: every session must have all research tools available and written down, and must USE them wherever they add value. Named explicitly: purealpha.app, fomo.family (whose claims are to be reviewed, not repeated), DexScreener, YouTube, all the scans, Etherscan, Stonkscan.io, robinhoodchain.blockscout.com/api, DefiLlama, Arkham, CoinGecko.
- [stated] He wants the twitterapi.io key usable in every session without him pasting it.
- [stated] Earlier standing rules that still hold: use twitterapi.io advanced search on EVERY deep dive, and check purealpha's insider-follow feed before forming a verdict.

## SECRETS - how to use them without ever printing them
Every paid key Ben owns already lives on the droplet `kookwatch` (159.223.88.168, key `~/kookwatch/id_kookwatch` on his Mac, reachable via Desktop_Commander start_process from any session). Run the query ON the droplet and bring back only results. Never cat a config, never echo a key into chat.

- `twitterapi_key` - in BOTH /opt/x-alert/config.json and /opt/followwatch/config.json.
- **Shared tool installed 30 Aug 2026: `/opt/xq.py`** (chmod +x, reads the key itself).
  ```
  ssh -i ~/kookwatch/id_kookwatch -o StrictHostKeyChecking=no root@159.223.88.168 \
    'python3 /opt/xq.py series "$TICKER" 6'
  ```
  Modes: `search "<q>" [pages]` (Latest), `top`, `user "<handle>"`, `series` (per-day counts + top posters). Query syntax is X advanced search, so `from:`, `since:`, `until:` and CA strings all work.
- Telegram bot tokens: per-bot config.json under `telegram`. kookwatch uses /opt/kookwatch/.env and holds a Telegram USER session string - most sensitive file on the box.
- NOT owned, would need buying: Etherscan API key (v2 covers all EVM chains with one key), Arkham API key. `coingecko_api_key` exists in /opt/basewatch/config.json but is EMPTY, plan "demo". `github_token` in gramwatch is empty.

## CONTAINER-REACHABLE, no key, no bridge - verified 200
| Source | Endpoint |
|---|---|
| DexScreener | `api.dexscreener.com/latest/dex/tokens/{ca}`, `/pairs/{chain}/{pair}`, `/search?q=`, `tokens/v1/{chain}/{a,b,...}` 30 per call |
| GeckoTerminal | `api.geckoterminal.com/api/v2/networks/{net}/tokens/{ca}`, `/pools/{pair}/ohlcv/{day\|hour\|minute}`, `/pools?sort=h24_volume_usd_desc`. RH slug `robinhood`. Needs `?include=base_token` on pool lists |
| CoinGecko | `api.coingecko.com/api/v3/...` free tier works unkeyed |
| DefiLlama | `api.llama.fi/overview/{fees,dexs}/{chain}?dataType=dailyRevenue`, `/summary/fees/{slug}`, `/v2/historicalChainTvl/{Chain}`, `coins.llama.fi/prices/current/...` |
| Blockscout RH | `robinhoodchain.blockscout.com/api?module=account&action=...` (v1) AND `/api/v2/...` BOTH work from the container. See correction below |
| RPC RH | `rpc.mainnet.chain.robinhood.com` - MUST send a browser User-Agent or 403 |
| GoPlus | `api.gopluslabs.io/api/v1/token_security/{chainid}` - taxes, mint/blacklist/pause, LP holders and is_locked |
| pump.fun | `frontend-api-v3.pump.fun/coins/{mint}`, `advanced-api-v2.pump.fun/coins/top-holders/{mint}` |
| Solana RPC | `api.mainnet-beta.solana.com` - getTokenSupply fine, getTokenLargestAccounts 429s instantly, do not use |
| LunarCrush | `mcp__LunarCrush__*` MCP tools, server-side |

**CORRECTION to [[rh-chain-thesis]] and [[twofold-two]]:** those files say Blockscout v2 is Cloudflare-gated from the container and v1 is the fallback. Both are wrong as standing facts. On 30 Aug BOTH returned 200 from the container, and separately BOTH 403'd for a stretch mid-session. It is intermittent rate limiting, not a permanent block. Retry with a browser User-Agent and a ~0.3s gap before falling back to the Chrome bridge.

**BLOCKSCOUT WAF CHANGE, 9 Sep 2026. A BROWSER USER-AGENT IS NO LONGER ENOUGH.** Probed from BOTH the droplet and Ben's Mac, same result, so it is not an IP ban and not the intermittent rate limiting above:
- site root `https://robinhoodchain.blockscout.com/` -> **200**
- any `/api` or `/api/v2` path with User-Agent ONLY -> **403 Forbidden, every time**
- the same path with **User-Agent + `Accept: application/json, text/plain, */*` + `Referer: https://robinhoodchain.blockscout.com/`** -> **200**
The root staying 200 is why this looked like nothing was wrong. **Send Accept + Referer on every Blockscout call from now on** (Origin too, harmless). Send them to Blockscout only, not to DexScreener or GeckoTerminal.
- **It had silently broken five droplet services**, all of which used Blockscout for HOLDER COUNTS and were therefore reading 0/None: `coinbench/bench.py`, `coinbench/ledger.py`, `delta-alert`, `launchwatch`, `parupdate`. Patched 9 Sep 2026 with a shared `_bsh(url)` header helper; DELTA 15,936 and PAIR 9,804 holders verified reading again afterwards. See [[rugscan-3]].
- **Standing lesson: when a source's homepage answers but its API does not, suspect a WAF header rule, not an IP ban.** A retry loop and a fallback chain will never fix it, and a bot that catches the exception will report a clean zero instead of an outage.

**CORRECTION to [[rh-chain-thesis]]:** it records `TWITTERAPI_KEY` as present in the Cowork container environment. It is NOT and never was. That wrong note caused a fallback to LunarCrush on the PFWA screen and produced a wrong "poorly marketed" verdict from partial data. Always go via the droplet.

## NEEDS THE MAC / CHROME BRIDGE
- **purealpha.app** - Ben holds PRO. No public API, login-walled. Read through Chrome on his Mac. Feed is in the DOM but not innerText (React shell stays "Loading pure alpha..."): query `a[href*="x.com/"]` and walk up to the row container. It is TRACK P, a detection source, not just a verification check - it sees projects before any on-chain signature.
- **X / x.com** - Chrome reaches it; the in-app browser hits a login wall. But prefer `/opt/xq.py`: structured, paginated, and X live search cannot reliably paginate back more than ~48h while the API can. Chrome is the fallback when a specific thread or /quotes view is needed. `get_page_text` on an X search returns only the FIRST article - use screenshots or a JS collector keyed by status id.
- **fomo.family** - site 200 from container but `prod-api.fomo.family` returns HTTP 430 `{"error":"unauthorized"}`. That is an AUTH wall, not the Cloudflare WAF ban recorded in [[crypto-token-research]] and lowcap-screen. Read `fomo.family/tokens/{chain}/{mint}` and `fomo.family/profile/{handle}` through the bridge. Use it for the thesis list, holder entries and holder count ONLY. Its LIQUIDITY figure is wrong ($152.2K vs a real $281K on TWO; $11.2K vs $39.3K earlier) - never take liquidity from fomo, and treat every thesis on it as a claim to check, not a fact.
- **stonkscan.io** - loads 200, Next.js app, no public JSON API at /api/v2. Bridge only.

## NEEDS A KEY BEN DOES NOT YET HAVE
- **Etherscan v2** - `api.etherscan.io/v2/api?chainid={id}&module=...`. One key covers every supported EVM chain. Returns "Missing/Invalid API Key" unkeyed. Free tier is 5 calls/sec, 100k/day.
- **Arkham** - `api.arkm.com/intelligence/...` replies "sign up for an api key". Entity labelling and cross-chain attribution, which nothing else in this stack does.
Blockscout plus raw RPC covers most of what Etherscan gives on RH chain, so neither is blocking; Arkham is the real gap for identity work.

## YOUTUBE - search in the container, TRANSCRIBE ON THE MAC
- [stated] 30 Aug 2026: Ben wants YouTube search plus transcription as a standing option on project analysis, specifically founder interviews and high-quality review channels, and specifically NOT shill content. Identify quality first, then transcribe, then use it as information.

**SEARCH works from the container.** `pip install yt-dlp --break-system-packages` then `yt-dlp --flat-playlist -J "ytsearch10:<query>"` gives channel, duration, views, id. No key.

**TRANSCRIPTS DO NOT.** The container IP is blocked: `youtube_transcript_api` raises `IpBlocked` and yt-dlp gets 429 then "Sign in to confirm you're not a bot". The droplet is also a datacenter IP and will fail the same way. **The Mac works** (residential IP, yt-dlp already at /opt/homebrew/bin/yt-dlp).

**Helper installed 30 Aug 2026 at `~/yt/ytget.sh` on the Mac**, run via Desktop_Commander start_process:
```
~/yt/ytget.sh search "<query>" [n]    candidates: channel, duration, views, id
~/yt/ytget.sh chan "@handle" [n]      a channel's recent uploads, to judge if it is a show or a ticker feed
~/yt/ytget.sh get <videoId>           clean transcript to ~/yt/<id>.txt, prints path + char count
```
Then read `~/yt/<id>.txt` with Desktop_Commander read_file (offset/length paginate). Verified end to end: `0FbCDpWL5OM` = "The Creator of StonkBroker Explains Everything", 58,315 chars.

**QUALITY FILTER - apply before transcribing, the whole point is to exclude shills**
Include: founder or team speaking in their own voice for a sustained stretch; 20+ minutes; a show that interviews many different projects over time; at least one adversarial or pushback question; published BEFORE the price move.
Exclude: under 10 minutes with a ticker in the title; "100x", "1000x", "GEM", "URGENT", "DO NOT MISS", a price target in the title; a channel whose entire catalogue is tickers (check with `chan`); high views with near-zero comments (bought); anything published mid-pump by an account that also posts the CA.
Treat a disclosed holder or sponsor as advertising, same rule as Ansem's z500 in [[ansem-worldview]]: usable for mechanism, never for the ticker call.

**What to extract:** founder claims that are checkable on chain, dated commitments, and anything that CONTRADICTS the site or docs. A founder contradicting his own landing page is the highest-value output of the whole exercise.

Known good RH-chain sources found 30 Aug 2026: The Rollup, Bankless, Thinking Crypto (Johann Kerbrat, RH Crypto GM), The Block Runner, Because Bitcoin. This is also how [[ansem-worldview]] was built from 25 episodes.

## Standing method rules
- NEVER rank candidates by liquidity on Robinhood Chain - rank by traded volume, require $5k+. Fake-liquidity impersonators exist for every winner.
- Reported liquidity is NOT exit depth. Only `balanceOf(quoteToken, poolAddress)` is, and that fails for Uniswap v4 where assets sit in the PoolManager singleton.
- ALWAYS use DexScreener marketCap, not GeckoTerminal fdv_usd, for RH tokens.
- ATH lookback must use DAILY candles (limit 180). Hourly limit 200 is an 8-day window and understates drawdowns badly.
- Never quote LunarCrush sentiment; use the `interactions` slope and `contributors_active`.

## Robinhood Chain Uniswap V3, addresses found 31 Aug 2026 (measuring a liquidity manager's real TVL)
- NonfungiblePositionManager: **0x73991a25c818bf1f1128deaab1492d45638de0d3** ("Uniswap V3 Positions NFT-V1"). The canonical mainnet NFPM 0xC36442b4a4522E871399CD717aBDD847Ab11FE88 HAS code on this chain but is NOT the live one: name() reverts and every balance reads zero. Do not assume the canonical address on a new chain.
- UniswapV3Factory: **0x1f7d7550b1b028f7571e69a784071f0205fd2efa**
- How the NFPM was found, reusable on any chain: eth_getCode on a protocol vault contract, regex PUSH20 (`73` followed by 40 hex) for hardcoded addresses, then call name()/symbol() on each candidate. Faster and more reliable than Blockscout, which rate-limits hard and returns an empty body rather than an error.
- Measuring a liquidity manager's TVL directly: balanceOf + tokenOfOwnerByIndex on the NFPM per protocol contract, positions(tokenId) for liquidity and ticks, getPool(token0,token1,fee) on the factory, slot0() for sqrtPriceX96, standard v3 amount math, then price via DexScreener tokens/v1/{chain}/{a,b,...} at 30 addresses per call.
- TRAP that cost a run: positions() returns int24 ticks sign-extended into a full 32-byte word. Mask to 24 bits (`& 0xFFFFFF`) BEFORE sign-extending, or a negative tick decodes as ~1.4e76 and math.sqrt overflows. Also compute the sqrt price as exp(tick*ln(1.0001)/2), never sqrt(1.0001**tick).
- RH RPC batching: eth_call batches of 40 with ~0.35s pacing work; batches of 120 return HTTP 429.
- STANDING RULE: a DefiLlama TVL for a liquidity manager is a FLOOR and can be 20x or more understated. Delta read $17,865 against $381,637 measured on chain and roughly $1.5M on the broader managed definition. Measure it before quoting it. See [[delta-liquidity]].
- STANDING RULE on revenue: never divide lifetime fees by the TOKEN's age. Sample the protocol's own live counter twice, minutes apart, and compute the rate. Delta's token was 30 days old, its product 16 days old, and the live rate was 2 to 2.5x the lifetime average, which moved the multiple from 492x to 120x.
- Delta exposes no public TVL endpoint. Its only working API paths are /api/farms/topology, /api/delta-hero and /api/usdg/pools; global stats live in the page RSC payload, not an API. When a SPA shows a headline stat but every /api guess 404s, parse the Next.js RSC payload out of the served HTML.

## CROSS-CHAIN SUPPLY TRAP, found on the RIPE screen 1 Sep 2026. Now a standing check.
- **For any token deployed on more than one chain, sum `totalSupply()` on every chain by RPC before quoting a market cap. Never take an aggregator's figure.** On RIPE, DexScreener printed $671k (RH-chain supply only) and CoinGecko printed $3.86M (a stale 1.24M circulating figure) against a measured 10,451,874 tokens = ~$48.3M. A 12x and a 72x error, live simultaneously, on the two sources this stack leans on hardest.
- This is the same failure class as the RAM satellite row in [[screen-log]], where DexScreener's "$1.37M market cap" was bridged supply. There it understated by omission; here it did on both chains at once. **CoinGecko's `circulating_supply` is a manually maintained field and goes stale on emission tokens.** Use it only after checking it against chain reads.
- Companion check: `platforms` on the CoinGecko coin endpoint enumerates every chain a token is on, which is the cheapest way to know how many `totalSupply()` calls to make.
- Related read: DefiLlama had NO Robinhood-chain entry for Ripe at all while $558,808 of collateral sat in its vault on that chain. A protocol's `chains` array is not a list of where it is deployed, only of where an adapter exists.
- **CORRECTION to [[rh-chain-infrastructure]]:** that file records Chainlink CCIP mainnet on Robinhood Chain as a "DIRECT CONFLICT", unconfirmed beyond testnet. It is CONFIRMED live. Verified burn/mint token pools deployed on RH chain 7 Aug 2026: `0xe51af1311832818a6d366081fc535ca56357a6ee` (RipeCcipBurnMintTokenPool) and `0x4b19f165bb1ce3f19bbe828d150706b9deeeec95` (GreenCcipBurnMintTokenPool). Consequence for screens: a CCIP-linked token's shallow RH pool is the exit venue for holders on the deeper chain, so RH liquidity is not the token's own float.
- Vyper protocols expose no `owner()`, `hq()` or `governance()` on the token; the registry pointer is `ripeHq()` / a project-specific getter. Compute selectors with `eth_hash.keccak` rather than guessing, then walk the registry for the live vault set. **On a redeployed protocol, name lookups return several generations of the same contract. Identify the LIVE set from the token-holder list and from `balanceOf`, never from the deployer's first batch.**

## Additions from the daily screen-scoring run, 2 Sep 2026
- **Second RH RPC, verified working from the container: `https://robinhood-rpc.publicnode.com`** (eth_chainId returns 0x1237 = 4663). Use it as the alternate to `rpc.mainnet.chain.robinhood.com`. **It 403s without a browser User-Agent**, same as the primary.
- **DexScreener also 403s the container without a browser User-Agent.** The table above does not say so and it cost a run to rediscover. Send one on every call.
- Base RPC `https://mainnet.base.org` works unkeyed and needs no User-Agent.
- **`pycryptodome` is NOT preinstalled.** `pip install pycryptodome --break-system-packages`, then `from Crypto.Hash import keccak` for function selectors. `eth_hash` is also absent.
- Product APIs beat explorers for falsifier checks whenever a project ships one. Longbow's `www.longbow.cash/api/stats` and `/api/markets` answered three falsifier legs in two unauthenticated calls, including per-market borrow and each market's `enabled` flag, which no explorer exposes.
- **Dune standing query 8562361 runs about 100 to 110 seconds at `medium` performance.** Three PATCH, execute, poll cycles fit comfortably in one session. Poll with `nohup` in the background: a foreground poll loop hits the 2 minute bash timeout and gets killed mid-execution.

## TOP-TOKEN CA SOURCE OF TRUTH, built 2 Sep 2026 (Ben asked for a basis list of top 50-100 RH tokens)
- No single source works. The pipeline is a two-source cross: **CoinGecko for identity, GeckoTerminal for liveness/ranking, DexScreener for cap.**
- **CoinGecko `platforms` has a dedicated `robinhood` key.** Categories `robinhood-ecosystem` + `robinhood-chain-meme` -> 338 curated CAs (join via /coins/list?include_platform=true, one big call). Only human-curated CA registry for the chain; it resolves DTF to the correct 0xeE5576Fa1Bcaa380e591D01245f406f3f384eb01. Use CG for identity ONLY, never ranking: 30 of 338 are bridged multichain tokens with GLOBAL mcaps (USDE, PENGU, VIRTUAL, APE...) — filter with `n_chains>1` from platforms.
- **GT pools sort=h24_volume_usd_desc**, ~240 pools -> ~125 unique live tokens, ranks by volume per standing rule and filters impersonators. GT rate-limits hard: 429 then 401 on bursts; ~3-4s pacing, incremental saves.
- Collisions the cross caught, resolve before screening these tickers: **HOOD ($5.8M/d live on a 4th contract CG doesn't list), VLAD, ABE, CACHE** — volume runs on non-CG contracts.
- Known gap: **INJOH-class tokens (sub-$5M, quiet volume) slip BOTH nets.** This list is the top-of-book basis; discovery stays with radar tracks B/C.
- Rejected: Blockscout /api/v2/tokens (bridged majors first, null addresses), DefiLlama (no token CA field).
- Deliverables were rh_top_tokens.csv (top 100 native by DexScreener mcap) + rh_full_universe.csv (411 rows with STOCK/MAJOR/BRIDGED/NATIVE/GT-ONLY class); regenerate from the pipeline above, container-only, no keys.
- **INSTALLED ON THE DROPLET 2 Sep 2026, refreshed daily by cron at 04:20 UTC.** Script `/opt/rhtop/rhtop.py`, outputs `/opt/rhtop/latest.csv` (top 100 native, columns rank,symbol,name,contract_address,class,dexscreener_mcap_usd,gt_24h_vol_usd,coingecko_id) and `/opt/rhtop/universe.csv` (all ~412 incl. STOCK/MAJOR/BRIDGED). First run verified: universe=412, native=330, PONS/CASHCAT/AI top 3.
- **ANY SESSION FETCHES THE CURRENT TABLE IN ONE COMMAND:** `ssh -i ~/kookwatch/id_kookwatch -o StrictHostKeyChecking=no root@159.223.88.168 'cat /opt/rhtop/latest.csv'` (via Desktop Commander start_process). Force a fresh run first with `'python3 /opt/rhtop/rhtop.py'` (~3 min). **This droplet CSV is the default CA lookup for any RH-chain ticker before falling back to search.** Log at /opt/rhtop/run.log.

## DUNE — reusable, added 5 Sep 2026 on the [[pullback-comeback]] build
- The Dune key already on the droplet (`/opt/coinbench/backfill.py`, also copied to `/opt/pullback/dune.json`) works **directly from the Cowork container**, no bridge. `api.dune.com` is reachable.
- **Private-query quota is FULL** — `POST /api/v1/query` with `is_private:true` returns "Max number of private queries reached". `is_private:false` succeeds. Public query **8613013** was created for this and is the scratch query to PATCH; 8562361 belongs to coinbench, leave it alone.
- `"performance":"large"` is rejected (HTTP 400 on the plan). Use `"performance":"medium"` — a chain-wide transfer aggregation over 623 contracts finished in **17 seconds**.
- Results paginate at `?limit=20000&offset=`, so the 5,000-row assumption in older notes is wrong.
- **THE HIGH-VALUE QUERY: daily wallet ARRIVALS per token**, which nothing else in the stack provides. `erc20_robinhood.evt_Transfer` -> first-ever receive per (contract, address) -> count per day, plus daily unique receivers and senders. This is historical holder velocity, and it is the layer that made the pullback study possible. Full SQL lives in `/opt/pullback/pullback.py`.
- Blockscout historical holders is NOT the way: RH chain block time is **0.10s** (~55M blocks in 70 days), so `eth_getLogs` reconstruction is hopeless. Dune is the only route.

## TRANSFER LIMITS between the container, the Mac and the droplet — learned 5 Sep 2026
- **The Cowork container has NO outbound route to the droplet.** Ports 22 and 8899 both time out; the egress allowlist blocks it. `nc`/`scp` container->droplet is impossible, so a file built in the container cannot be pushed there directly.
- The working path is: `Desktop_Commander__write_file` to the Mac, then `scp` Mac -> droplet over the ssh key. Write plain source, not base64 — base64 is 33% more tokens and gains nothing.
- **`pkill -f 'foo.py'` over ssh kills the ssh session itself** (the remote shell's own command line contains the pattern) and returns exit 255 with no output. Use `pkill -f 'foo[.]py'`.
- GeckoTerminal free tier is ~28 requests/min per IP. 623 daily-candle pulls take ~70 minutes from one IP. The droplet is a second IP and can run a second harvest in parallel, but it gets rate-limited harder (~2.3/min observed) — the container is the faster of the two.

## FOMO APP TOOL STACK — from @ProMint_X, 5 Sep 2026 (Ben: "remember them")
Source post: https://x.com/promint_x/status/2096366514333253939 (9.1k follower account, 1,130 likes, 86.7k views). All eight probed from the container 6 Sep 2026; reachability noted per row. **All are independent/unofficial — none is affiliated with fomo.family.** Treat their output as a claim to verify on chain, same standing rule as fomo.family theses.

| # | Tool | What it does | Container reach |
|---|---|---|---|
| 1 | **fomoscan.sh** (`https://www.fomoscan.sh/`) | fomo.family wallet index. Full Solana/EVM addresses with one-click copy, follower graph, "identity for onchain" | 200 (SPA shell only) |
| 2+3 | **copyfomo.com/find** | Handle **or profile screenshot** -> the wallet behind it, verified on chain. Full addresses gated behind their Telegram bot | 200 |
| 4 | **github.com/mickeyhogen/fomo-helper** ("Fomo Lens", by @0xHogen, MIT) | Chrome extension, floating card over **Fomo, GMGN and XXYY**: token narrative, Fomo theses, holders, and **Fomo holding share = loaded Fomo holdings / market cap**. Needs a signed-in Fomo session in the same browser profile | repo 200 |
| 5 | **fomoguard.family** ($GUARD) | Scores tokens pre-buy, paper trading, auto-generated entry/stop/target plan, rug checks, and **limit orders that click Fomo's Buy/Sell when a target market cap prints** (beta) | 200 |
| 6 | **fomosignal.family** | Chrome extension, 21KB, no account, no servers. Fires **the moment a Fomo trader links a wallet** and their whole trail goes public. ~1s per-trader scan, own RPC key optional. CA `3h6q3Xjq4nVifeDZ9gZdpF9eCm5XBfnXXpgFkS8npump` | 200 |
| 7 | **copyfomo.com** | Non-custodial copy trading off the public Fomo leaderboard | 200 |
| 8 | **fomoscope.xyz** | Billed as APIs for leaderboards/trades/positions | **404 on `/` and on every API path probed. Do not cite as available until re-probed.** |

**The one that matters most for this stack: the FomoScan B2B REST API** (`https://www.fomoscan.sh/api`, "live now", OpenAPI/Swagger docs published, dashboard + `partner.fomoscan.sh`, Telegram `@fomoscan_sh`, terminal at `terminal.fomoscan.sh` 200). Endpoints as documented on their page:
- `GET /v2/user/handle/{handle}` — handle -> every verified wallet it controls, Solana and EVM
- `GET /v2/user/wallet/{address}` — reverse: wallet -> the trader, chain inferred from the address, no chain param
- `GET /v2/user/id/{id}` — stable id lookup; handles change, ids do not, so store the id
They claim proof-verified matches (never a guess), no stale snapshots, and that opted-out traders are never returned. **Not yet tested by us — needs a key from their dashboard, which Ben does not have. Probing `/api/*` unauthenticated just returns the Next.js SPA shell, not JSON, so a 200 there means nothing.**

- **Why this matters:** handle -> wallet is the exact problem that ate whole sessions in [[dfarmer-wallets]], [[dinerodom-wallets]], [[uberm-wallets]], [[rbthreek-wallet]], [[somerandomtrader-wallet]] and [[ted-wallet]]. If the FomoScan API delivers, it replaces that hunt with one call. Worth Ben buying a key before the next wallet hunt.
- **Do NOT let these replace verification.** [[copytrade-targets]] records that infrastructure addresses keep faking a track record, and the fomo.family liquidity figure is already known wrong. A "verified" wallet from a third party is still a candidate to confirm with `seller_ledger.py` and the [[x-claim-audit-method]].

## GITHUB FROM THIS CONTAINER — three routes, two of them lie. Found 13 Sep 2026 on the YOINK screen.
- **`api.github.com/repos/{o}/{r}` returns a sandbox refusal**, not data: `{"message":"GitHub access to this repository is not enabled for this session. Use add_repo..."}`. Naive JSON parsing turns that into every field reading `None`.
- **`github.com/*` HTML returns 403** to the container, and **`codeload.github.com` zip download returns 403.**
- **`WebFetch` on a repo page invents a plausible wrong answer.** On `github.com/DefiLeoo/YOINK` it reported "This repository is empty, 0 stars, no files" for a repo that actually had 343 stars, 21 forks and ~124KB of Python. **This nearly shipped as a claim mismatch in a live screen.**
- **WHAT ACTUALLY WORKS, in order of preference:**
  1. **`git clone --depth 1 https://github.com/{o}/{r}` from the container succeeds.** Best route: it gives the real tree, and for a CLI tool it lets you `pip install -e .` and RUN the thing, which is the correct answer to "is it live".
  2. **`raw.githubusercontent.com/{o}/{r}/{branch}/{path}` is NOT blocked** and returns real bytes. Probe README.md / pyproject.toml / tests/* to prove files exist and get sizes.
  3. **The Chrome bridge loading `api.github.com/...` in a tab returns the real JSON** — stars, forks, `git/trees/main?recursive=1`, `/contributors`. Use this for star and commit counts.
- **STANDING RULE: never report a GitHub repo as empty, missing or fake on the strength of WebFetch or the container's GitHub API. Clone it or raw-probe it first.** Same failure class as the Blockscout WAF note above: the source answers, the answer is wrong, and a bot that trusts it reports a clean zero.
- **And the positive version: for any project shipping a CLI or a library, the strongest possible Q4 ("is it live") answer is to install and run it in the container and paste the output.** On YOINK: `yoink demo`, `yoink score`, and `python -m unittest discover -s tests -t .` returning "Ran 201 tests in 0.456s, OK" settled three separate claims in one step.

## CORRECTIONS FROM THE REGIME-TRACKER BUILD, 14 Sep 2026. See [[regime-tracker]].
- **DUNE IS OUT OF CREDITS.** The key on the droplet is still VALID (GET `/api/v1/query/8613013` returns 200) but **POST `/execute` returns 402 Payment Required**. This supersedes the "Dune works directly from the Cowork container" note above until Ben tops up the plan. Do not plan a backfill on Dune without checking execute first.
- **DefiLlama BRIDGES api is entirely paywalled**: `bridges.llama.fi/netflows/{day,week}`, `/bridges?includeChains=true`, `/bridgevolume/{chain}` and `/bridgedaystats/{ts}/{chain}` all return **402**. `defillama.com/bridges` is **403** to the container with no `__NEXT_DATA__`. The rest of DefiLlama (stablecoins, chain TVL, dexs overview) is free and unaffected.
- **THE FREE SUBSTITUTE for chain net flow: `stablecoins.llama.fi/stablecoincharts/{Chain}`** - daily history, years deep, per chain (Robinhood 70 days, Solana 1,588, Ethereum 3,212). Stablecoin supply change is dry powder arriving, which is a cleaner read on "can this chain buy things" than bridge flow.
- **pump.fun `/coins` DOES NOT PAGINATE.** `frontend-api-v3.pump.fun/coins?sort=created_timestamp&order=DESC&offset=N` returns about 70 rows at offset 0 and an **empty list at offset 5000, 50000 and 200000**. It is a live window, not an archive. `sort=ath_market_cap` returns raw-token-unit values (1.9e23) with every timestamp stamped 12-19; **`ath_market_cap` is only trustworthy on recent coins.**
- **DexScreener `tokens/v1/{chain}/{a,b,...}` returns PAIRS and caps the response**, so 30 addresses can come back priced for 12. Use chunks of 12 and treat a low priced-to-requested ratio as an outage, not a quiet market.
- Confirms the standing GT rule with a number: **GeckoTerminal `fdv_usd` read $9,633M on a fresh Robinhood Chain pool.** Use GT to find pools, DexScreener for the cap.
- Droplet has **no `sqlite3` CLI** (use python3), and Desktop Commander's device calls **time out at 60 seconds**, so any droplet job over a minute must be `nohup`ed into a log and polled.