Send a batch of traces. Get a readout in two weeks.

What to send

A batch of one long-horizon agent's traces, anonymized to whatever level you need. Any of these works:

• Tool-call logs with typed tool returns (the returns matter more than the model's text; the read treats a rejected call as a no-op).
• LangGraph state history for one or more threads (get_state_history export).
• OpenInference traces from Arize Phoenix, or a comparable span export.
• A coding agent's edit log with the editor's success flags, plus the repository it started from.
• A durable-execution workflow's step stream with inputs and outputs.

Include one line on the commitment to look for: the early decision a later step must honor. Agents whose state exists only in free-text logs are out of scope for the read; if you are unsure, describe the agent and we will say.

What you get back

A written readout, within two weeks of receiving traces, that locates where each run's committed state held, where a step acted against it, and which of the two causes bound at each break. Nothing runs in your production path, nothing needs model access, and ERA's scoring stays behind the hosted instrument.

Send traces →

Design partners

ERA is working with a small number of teams running long-horizon agents in production, on any cloud or framework where the read can reconstruct committed state. After the offline read, a design-partner engagement demonstrates the matched repair on an agreed slice of your agent, head-to-head against the unmanaged agent on coherence, cost, and latency, and both sides agree the success criteria before the run.

Introduce your team →

Data room

The methodology, the dataset census, the formal proof, the technical FAQ, and filings, for technical evaluators and prospective partners. Access under a mutual NDA.

Request access →

Code and integrations

The committed-state layer and the framework integrations are open source. We welcome issues and pull requests at github.com/ERA-Fathom.

Press, research collaboration, and everything else

contact@embeddedriskanalytics.com. We read every message and reply to the ones we can help with.