Metrics
Measure the system.
Don’t confuse the measurements.
Runtime health, operational activity, continuity quality, and experiments answer different questions. They stay separated here so a useful number never gets mistaken for a universal score.
Runtime health
Is the machinery behaving correctly?
Revision—Authoritative semantic revision
Events—Governed event count
Invocation events—Operational lifecycle evidence
Replay—Verified replay duration
Operational picture
What durable work exists right now?
Work items—Governed work records
Open work—Current unresolved work items
Accepted results—Durable accepted milestones
Conversation turns—Governed conversation application
Live application traffic
A privacy-safe window through the sudofx boundary.
Shows lifecycle evidence only: which application used sudofx, when it happened, which boundary stages completed, and bounded context size. Prompt, response, research content, and payload data are never shown here.
Traffic
Waiting for application invocation evidence…
The next governed application invocation will appear here without exposing its contents.
Application use
Which applications are using the engine, and how?
Derived from generic governed application actions and invocation lifecycle evidence. This view is disposable; SQLite remains authoritative.
Applications
Waiting for durable application evidence…
No application-specific meaning is inferred here.
Continuity quality
Did a fresh intelligence receive enough faithful context?
These are measurements about one continuity experiment, not general model-quality scores.
Context—Bounded delivered context
Full context—Available source context
Reduction—Bytes removed before provider delivery
Semantic review—Latest explicit assessment
Context delivery
Waiting for durable aggregate evidenceBounded context versus full source context
Current durable observations created before this metrics contract may not contain byte aggregates. New cycles preserve only the safe aggregate measurements.
Current experiment
Where is the continuity stress test?
Cycle—Global continuity cycle
Semantic lens—What capability is under test
Exposure—How much readable context survives
Pressure—Competing or adversarial condition
Experiment telemetry is loading from the current disposable projection.
Application metrics
Application meaning stays local.
WAKE✳︎ owns research-specific metrics such as cycles, evidence collection, notebook progress, and research funnels. sudofx exposes only generic authority/runtime measurements here.
Open WAKE✳︎ metrics →Interpretation rule
Evidence is not authority.
A fast replay, a passing semantic review, or a high compression ratio is evidence about a specific boundary. None of those measurements gets to rewrite the durable record.
Waiting for live projection freshness metadata…