Embedded Risk Analytics reconstructs what an agent committed over a run and names where it later worked against itself. The read answers calls from this page with no key, no server, and no build step. Pick one of our examples or paste your own, and the findings render below.
The live read
These are compact examples. The full DBOS, CrewAI, and DeepAgents runs live in the census.
The hosted read answers unkeyed calls at 50 a day from one address. Your own 50 cover anything this demo does.
Fathom now goes by Right Rudder. Old links and the fathom-read package keep working.
The read
The read tracks six ways an agent turns against its own committed state.
Stale reference
The agent acts on a fact it already removed or renamed away.
Superseded value
The agent writes or answers with a value it already replaced.
Authored contradiction
The agent reintroduces a token into a record it had already migrated.
Residual
The agent ends the run with a record still carrying a value it replaced elsewhere.
Duplicate commit
The agent adds an entity a collection already holds.
Post-commit mutation
The agent changes a record after committing it.
Each finding names the operation and the committed fact it broke. The read returns the findings and their kinds, and it never reads the content of what the agent committed.
Measured
In our reread of DBOS's own Hacker News research agent, vendored unchanged and run on gpt-4o-mini through OpenRouter across five topics at ten iterations each on 2 September under fathom-read 0.4.1, the agent as it shipped worked against its own record. With the repair in front of the one step where it proposes its next queries, and nothing else changed, the redundant work fell away.
As it shipped
With the repair
Across the 45 follow-up steps, the repair let 23 proceed untouched, filtered 21 to the agent's own non-contradicting alternative, and regrounded 1. The agent's own first-choice next query already sat in its committed list on 21 of those 45 steps. Every number comes from a public run, with the trace, the mapping, and the command in the census under rows/dbos-hn-agent.
The repair
The read has a companion that acts. The repair takes the run so far and the actions the agent proposes next, evaluates each proposal alone against the committed state, and returns one of three decisions.
Proceed
Every proposal survives, and the step runs as the agent intended.
Filter
Some proposals survive, and the caller takes the agent's own non-contradicting alternative.
Reground
No proposal survives, and the response hands back the committed facts each rejected proposal broke.
The response holds decisions and facts, and never scores. The Right Rudder page carries the full before-and-after and the key form.
The census
The coherence census holds full reads of production agents, each carrying its trace, its mapping, and the exact command. Three carry complete runs today.
The whole census lives at github.com/RightRudderAI/coherence-census.
Install the read
right-rudder ships on PyPI at 0.7.0 with fifteen adapters, the client, and a command line that runs right-rudder read, right-rudder reground, and right-rudder key. langchain-right-rudder adds the read to a LangChain agent as middleware, and a capture that extends it to the agent's sub-agents. Both install from PyPI.
Work with us
We run no-cost reads for teams building long-horizon agents. If a coherence read on your own workflow would help, the contact page reaches us.
