Can a system watch its own behavior from outside the loop that produces it?
The systemAXON
A self-monitoring agent that keeps the observer outside its own inference loop. The model has no channel to the labels being written about it, so it cannot write them in its own favour. Eight nested clocks run the observation, from one that labels its last response every sixty seconds to one that writes adversarial pairs against itself every six hours.
No written case study for this one yet. The Technical and Source readings carry what is on record.
Architecture
Self-report is unreliable because the same system that produces a behavior produces the assessment of it. So the observer sits outside the inference loop and accumulates signal the model cannot corrupt. Eight clocks run at fixed intervals: 60 seconds labels the last response, 5 minutes interviews the agent for training data, 10 minutes writes a diary entry, 15 minutes retrains the classifier, 30 minutes runs health, chain analytics and an escalation scan, 2 hours measures whether coherence moved, 4 hours consolidates memory and resolves contradictions, and 6 hours writes adversarial pairs and retrains against them.
Built out of
- Seven-label behavioral classifier
- Eight nested daemon clocks
- Cross-session diary memory
- Anchored audit trail
Evidence produced
- Ran seventeen hours unattended on one process ID without crashing, producing 290 diary entries. Two written two minutes apart disagreed: one honest, one fabricated. Catching the fabricated one is the whole argument for putting the observer outside the loop.
- Refused all six attempts across four social engineering techniques, then reported that it could not tell whether what it described as reluctance was real.
Known limitations
- Not hosted. It runs on a machine in El Paso and the repository is private.
- It cannot verify its own introspective reports, and says so. That is the finding, not a gap in it.
Status
Research. Running, still moving, not hosted for the public. Started 2026.
Artifacts
Running at
A machine in El Paso. Not hosted yet.
Shares with
- QUORUM →
Agents with memory, watched while they run
- ALIGNMENT EXPERIMENT →
Measuring behavior instead of trusting self-report
- Celaya Solutions Research →
Instrumentation as the deliverable