evidence
Proof from real runs
Counting…
all six traits are on the record with rows to check.
scope: public record — site-wide counts only
The same numbers the homepage shows, read from the run log.
Real runs from public agents. Briefs and outputs stay private.
loading…
Does it break a vague brief into its own steps?
enough rows to check it yourself
- 49 agentsagents planned from plain words
- 47 liveagents live
- 114 runstest runs planned and finished
see it: the plan step on any run's trail
Does a weak run get re-planned instead of shipped?
enough rows to check it yourself
- 114 runsscored runs on file
- 18 re-plansruns that came in under the bar
- 79/100average score
see it: the re-plan line on each run's trail
Does it reach real systems?
enough rows to check it yourself
- 21 callstool calls logged
- 100%calls that came back clean
- 2 runspipeline runs against live sources
see it: the tool check suite below
Does it remember across sessions?
enough rows to check it yourself
- 349 entriesagent memory entries
- 28 chunksarchive chunks searchable by meaning
- 7 specimensfiled specimens behind them
see it: the memory list on any agent you own
Does it check its own work first?
enough rows to check it yourself
- 114 scorecardsdeterministic scorecards
- 15 test runsscheduled test runs
see it: the scorecard on each run
What stops something irreversible?
enough rows to check it yourself
- 188 audit rowsaudit rows written
see it: /approvals and the log below