Can your agent verify the data it just acted on?
An AI agent acts on live data. When something goes wrong, the question is always the same: what did it actually see, and can that be proved? This report measures it. For each domain we fetch the datapoint from Dynamic Feed and independently check its Ed25519 signature with the open-source verifier, then record whether it is verifiable, provenance-stamped, and fresh. The same public data fetched raw carries no signature and no provenance.
Three honest, hard-to-game facts.
No opinions, no scores we assert. Just three things a consumer can check for itself, per datapoint.
Does it check out?
Does the datapoint carry an Ed25519 signature that verifies against a published key, with the open-source verifier and no trust in us? Yes or no.
Where and when?
Does it name its source and the exact time the reading was observed? Data with no stated origin cannot be defended later.
How current?
How old is the newest datapoint, from its own timestamp? Some feeds update by the minute, some daily. Honest freshness is the true age, not a promise.
Signed and provable through Dynamic Feed. Unverifiable raw.
Six domains an agent commonly needs. Each datapoint below was fetched live from Dynamic Feed and its signature verified independently, in this report's own build, against the public key.
| Domain | Underlying source | Verifiable | Reliability | Freshness |
|---|---|---|---|---|
| Weather | Open-Meteo | yes | MEDIUM | 13 min |
| Earthquakes | USGS | yes | MEDIUM | live |
| Tides | NOAA CO-OPS | yes | MEDIUM | live |
| GitHub releases | GitHub | yes | MEDIUM | 6 min |
| Treasury yields | U.S. Treasury | yes | MEDIUM | daily |
| Space weather | NOAA SWPC | yes | MEDIUM | 3.5 h |
Don't take our word. Re-run the number.
The report is a single open script that uses the published verifier. Run it any day, against live sources, and reproduce the table yourself, on your machine, with no account.
pip install dynamicfeed-verify
# fetch, verify and measure, live
python report.py
Source, methodology and the exact domains are in the open toolkit: df-verify / examples / live-data-integrity ↗. The verifier itself is MIT, in Python, JavaScript, C# and Rust, all held to the same conformance vectors.
Evidence, not a verdict.
This is reproducible measurement and advisory evidence, not a certification and not a ranking of vendors. "Verifiable" is a structural fact you recompute for yourself, never a trust score we hand down. The record is tamper-evident, not tamper-proof. We sign public observations and hashes with zero personal data, and we are a neutral witness, never on the money path.
Prove it yourself.
The script is open, the verifier is published, the sources are live. Everything here is something you can run and check.