TxWhy

Every claim, one click

Everything we say about TxWhy, next to the link that checks it. Numbers marked live come from the service itself as this page is rendered.

It works on mainnet

  1. A real agent swap built on a stale quote failed simulation, was rebuilt from a fresh quote for the same trade, verified locally, and landed.

    Solscan, Sep 23, slot 449732359
  2. An agent paid $0.001 USDC for a verified repair over x402 on mainnet, with no account.

    Solscan receipt
  3. The paid endpoint is listed in the x402 Bazaar, the facilitator's public catalog.

    PayAI discovery (search TxWhy)
  4. Run a repair on live mainnet state right now, no wallet needed.

    /repair, four one-click demos

It is measured, not claimed

  1. The worker samples nine of Solana's busiest programs around the clock and replays real landed failures through the engine every 90 seconds, publishing every verdict with its exact reason.

    live 29,030,467 sampled since Sep 21, 36.7% failed, 128,490 replayed, 12,424 rebuilt and verified, 22.8% of the failures real people hit

    /failures
  2. The same index as JSON, free to reuse, CC BY 4.0, no addresses stored.

    /api/v1/index
  3. Usage is counted by the service itself, with tests excluded, and published.

    live 103 repairs for 19 distinct callers, 84 diagnoses

    /stats

You never have to trust it

  1. The rule for what a repair may change is code you run yourself, with no network access: same fee payer, same signers, every other instruction byte for byte, one swap replaced or one limit moved within the cap.

    src/lib/verify.ts
  2. The verifier exists in TypeScript, Rust and Python, all tested against the same real-repair fixtures and 48 attack cases, on every push.

    GitHub Actions, CI on main
  3. The server runs the verifier on its own output and refuses to return a repair that fails it; a repaired transaction is only returned after it passes simulation.

    src/lib/repair.ts
  4. Hardened like infrastructure: 122 malformed inputs against the repair API and 52 against the MCP server, never a 500.

    scripts/fuzz-repair.mjs and scripts/fuzz-mcp.mjs

Someone else has used it

  1. An external builder ran his own failed swaps through the CLI over four rounds, Oct 4 to 7, and every point he raised shipped the same day. Quoted with permission.

    TESTERS.md
  2. Solana Agent Kit, the default agent toolkit, has no repair step; our pull request adds one.

    sendaifun/solana-agent-kit#620

It ships in every stack

  1. npm: @txwhy/sdk, with sendWithRepair for web3.js and @solana/kit, a one-line wallet wrapper for any dApp, and the npx txwhy command line.

    npmjs.com
  2. PyPI: txwhy, the same verifier and send loop for solana-py.

    pypi.org
  3. crates.io: txwhy-verify, the offline verifier in Rust.

    crates.io
  4. MCP server with three tools for any agent, streamable HTTP.

    /api/mcp
  5. Telegram bot with wallet failure alerts.

    @txwhy_bot
  6. Browser extension that adds the missing paragraph on Solscan and Solana Explorer.

    extension/
  7. 573 error-code pages, so a search for 0x1771 lands on an answer.

    /errors
  8. MIT, one repository, every commit in the open.

    github.com/SUNDRAM07/txwhy

Something here that does not check out? Open an issue at github.com/SUNDRAM07/txwhy. The claim comes down before the page does.