ZEROlobby
22/22 strict⌘Kvisitor
studio · build → ship
explore · scrub · compare · attest
exploration-only · no commits
founder shell · lab· exploration plane · AGI-VIS §exploration← back to control plane

lab · what is the system doing to itself?

The 4 governance planes (control / judgment / verification / futures) are decision surfaces — they answer "what should I do." The lab is an inspection surface — it answers "what should I be looking at that I'm not." Tukey's exploratory data analysis, applied to the shell's own behavior.

benchmark drift · last 24h · across 6 doctrine gates
composition8.4 → 8.2
−2.4%
state8.8 → 8.9
+1.1%
intervention8.5 → 8.5
flat
provenance9.1 → 9.0
−1.1%
memory7.9 → 8.1
+2.5%
mode8.6 → 8.5
−1.2%
intervention density · last 24h · by sovereign-action verb
approvelast 24h
12
refuselast 24h
3
pauselast 24h
1
escalatelast 24h
0

event tape · 2026-05-13 · scrubbable

click any marker · arrow keys to scrub · home/end for extremes
08:00z18:00z

trust decomposition · current confidence + 8-class breakdown

72%confidence
high
  • disagree
  • sparse
  • irreversible

per AGI-VIS §9 + ZERO-UNCERTAINTY-DECOMPOSITION · the single 72% confidence decomposes into 3 contributing classes · 1 high-severity (irreversible) requires paired intervention affordance per §4.

discovery prompts · Tukey · §1

  • · where is the model unstable in the last 7d?
  • · where is memory retrieval skewed by stale evidence?
  • · where do interventions cluster · by time · by class · by role?
  • · where do latency and failure distributions hide?
  • · where does benchmark performance drift exceed 2%?
  • · where does trust degrade over time?
  • · where do autonomous actions diverge from prior patterns?