Continuous reasoning, at a cost that makes sense
Compact Contracts compress raw logs, metrics, traces and events into structured representations built for cross-layer reasoning — roughly 6,000× less volume, with the diagnostic context intact.
Reasoning over raw telemetry does not scale.
A large estate produces more telemetry than any model can read continuously. So reasoning runs only on demand, on a sample, or once an incident is already visible — exactly when it is least useful.
Generic summarization makes it worse: it keeps what reads well and drops what correlates. The signal that links a storage stall to a routing change is usually the least interesting line in either log.
A representation layer, not text compression.
Each Compact Contract preserves the relationships between network, compute, cloud, application and infrastructure signals that the reasoning pipeline needs — and discards the redundant volume that does not contribute to correlation or root-cause tracing.
Contracts are typed and versioned, so what the model sees is predictable and auditable.
- Entity, time window and topology position on every contract
- Deltas and anomalies instead of repeated steady-state values
- Links to raw evidence retained for investigation
entityhall B · row 4
window5 min
deltabaseline −1.2 dB
relatedvia topology
evidencelinked, not discarded
Research
Method, results and limits for the techniques on this page.
Compact Contracts
Roughly 6,000× less telemetry volume, with diagnostic context preserved.
Read paper →ResearchCurve-1: a reasoning architecture
The four-stage pipeline behind cross-layer root-cause ranking.
Read paper →DocumentationPlatform concepts
Topology graph, contracts, hypotheses, incidents and verification.
Open docs →