@sawinyh
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Building with coding agents. Writing about AI engineering, verification, agent-native codebases, and what survives production. DeFi since 2019.
East Coast, US
Joined January 2015
- Tweets9.4K
- Following816
- Followers2.7K
- Likes10.4K
Pinned Tweet
CPVO: Cost Per Verified Outcome for Engineering Organizations
The metric that answers a simple question: Is your AI investment actually paying off?
A deep dive into the methodology and how to measure it ⤵️
sawinyh.com/blog/measuring-a…
An end-to-end test closes an agent-drafted outreach sequence after two follow-ups, then injects a late reply three days later. It must still attach to the original conversation and change the outcome to won.
Smart Money bought a token. Is anyone buying in behind them?
Zatto counts the new buyers in the 10 minutes after each Smart Money buy, then tells you what buys at that level did 24h later.
Built on the @nansen_ai API
zatto.nsawinyh.workers.dev
github.com/sneg55/zatto
What it has measured so far, 542 Smart Money buys on Base:
under 5x the token's normal buyer rate, copying is a coin flip
5x and up, 20 of 25 were higher at 24h, median +29.6%
One row per token minute, so a cluster of wallets buying together counts once, not seven times.
A run goes every 8 hours on cron. If you want one now, a scan costs 1 USDC over x402, signed in your own wallet. The app never holds a key.
Method, constants and the measured numbers are all in the README.
github.com/sneg55/zatto
The agent form worker sets `clicked = true` immediately before submit. A navigation timeout after that is stored as `unverified`, alongside whatever result text the page returned.
A placement needs three separate UTC days with a live observation. Re-running the worker on the same day does not advance it, even when every check passes.
The worker commits a job lease before any external call. If that lease expires, it gets one requeue, then moves to an ambiguous queue instead of retrying a side effect it cannot prove.
Each plugin test case gets two scored runs, one with the plugin and one without. A pass rate on only the first run can’t tell whether the plugin added anything.
Today’s preflight screen keeps its previous verified pass after rejecting a bad receiver address. That refusal happens before any test message is sent, so there is no new send-path evidence to store.
In the automated inbox, `Action: delayed` gets `dsn_delay` and leaves the sequence running. Only `failed` or a 5.x status suppresses the recipient.
One batch row requests a 5-star comment from evidence stored with rating 0. The evidence text itself says “10 out of 10.” The structured value and the sentence disagree before generation starts.
After a decade away from hardware, one builder says AI cut work from weeks to hours. Which step still waits on the physical build?