Query endpoints
After this page you can pull anything the Telegram bot or MCP server can show you, directly over HTTP with a query-scoped key.
Each endpoint mirrors one of the nine questions:
| Endpoint | MCP twin | Answers |
|---|---|---|
GET /v1/why/{sig} |
why |
Decoded failure story for a signature |
GET /v1/diagnose-drop/{sig} |
diagnose_drop |
Drop classification + evidence |
GET /v1/baseline/{program} |
baseline |
Program vs ecosystem vs per-provider rates |
GET /v1/release-diff/{program} |
release_diff |
Pre/post-deploy regression table |
GET /v1/top-issues/{program} |
top_issues |
Bot-segmented top issues |
GET /v1/cu-headroom/{program} |
cu_headroom |
Per-instruction CU p95 + trend |
GET /v1/wallet/{program}/{wallet} |
wallet_history |
Wallet-scoped history |
GET /v1/provider-report/{program} |
provider_report |
Measured drop rate per RPC provider |
The why/diagnose-drop/status/baseline/release-diff/top-issues/cu-headroom/
wallet/provider-report read routes are the free, no-auth public surface (rate-limited per
IP). The public Landing Score dataset the leaderboard renders is a committed
snapshot (apps/web/src/data/leaderboard.json), not a live API route.
Example: why
Section titled “Example: why”# public read route — no key required (rate-limited per IP)curl http://127.0.0.1:8790/v1/why/4h2K…{ "sig": "4h2K…", "outcome": "failed_onchain", "program": "DemoAMM…", "instruction": "swap", "ixPath": "swap", "cpiDepth": 0, "errCode": 6001, "errName": "SlippageExceeded", "cuUsed": 48211, "cuRequested": 200000, "payer": "9xQe…", "payerBotScore": 0.02, "issue": { "fingerprint": "v1:9c41ae…", "count7d": 14, "firstSeen": 1766880000000 }}outcome is one of landed, failed_onchain, dropped, preflight — the
failure taxonomy as an enum.
Example: leaderboard dataset
Section titled “Example: leaderboard dataset”The public Landing Score data the leaderboard renders is a committed snapshot
(apps/web/src/data/leaderboard.json), not a live API route. It carries sample: true (marking
it a committed snapshot rather than a live feed) and a sampleRate per program — the dataset is
honest about sampling everywhere, not just in prose.