Risk-coherence agent
Risk scores that
mean something.
monotonic tests whether your system's own risk predictions actually track reality — a 90-day Spearman correlation against operational load, not a point-accuracy score that flatters the model. rho > 0.6, p < 0.05, or it doesn't ship.
[higgsfield: a rising gold-dotted trend line against a faint dashed regression line, black graph field, engraved line art]
monotonic · 90-day Spearman gate · seeds 42/7 verified
Why teams install
What monotonic earns back
Point accuracy lies. Rank correlation doesn't.
A risk model can nail today's number and still be useless — what matters is whether risk rises and falls with real operational load over time.
‹3e549bac›A constant risk score fails, loudly
The planted-failure predictor reports the same "risk" every day. rho collapses to exactly 0.0 and the gate goes red — no partial credit.
‹08f670f7›Stable across seeds, not a lucky run
rho holds 0.75–0.83 across six independent seeds with permutation p pinned at its floor (≈0.001).
‹3e549bac›How it works
Three steps. No faith required.
01
Observe
Collect 90 days of the metrics an operator can actually measure: defect escapes, test-cycle time.
02
Correlate
Spearman rank correlation between your predicted risk and the observed load index — correct tie-handling, no third-party dependencies.
03
Gate
Pass only if rho > 0.6 AND a seeded 1,000-shuffle permutation test clears p < 0.05.
Proof band
One packet. Every claim accounted for.
packet rig/monotonic · seed 42 claim "predicted risk tracks observed operational load over 90 days" window 90 days · 1000 permutations rho 0.7688 · floor 0.6 p-value 0.0010 · ceiling 0.05 gate PASS — rho and p both clear with margin themis SEALED — 08f670f7
Every monotonic run emits a packet like this — hash-chained, verifier-signed, and queryable from your own dashboard. How ProofPackets work →
Objections
Answered before you ask
Isn't this just a dashboard metric?
Dashboards show you a chart. monotonic runs a statistical test — Spearman rho plus a seeded permutation p-value — and seals the verdict. It either clears the gate or it doesn't.
Why rank correlation instead of RMSE or accuracy?
Point accuracy rewards a model that nails today's number and says nothing about whether risk tracks reality over time. Rank correlation is the honest question: does risk go up when load goes up?
Why trust the p-value?
1,000 seeded shuffles, not an asymptotic approximation. The permutation test and the correlation are both pure-Python, reproducible, and sealed into the ProofPacket — you can re-run the exact math.
Buyers
Who runs monotonic
Head of Engineering
60-eng multi-agent platform
VP Reliability
fintech infra
Risk Modeler
insurance underwriting desk
Pricing
Start free. Pay when it pays.
Generalist ships today. Domain Expert adds vertical memory + priors. Custom is built to your stack.
Free
Priced after a short conversation
1 system monitored, 30-day window. Full statistical output.
Team
Priced after a short conversation
Unlimited systems, 90-day rolling window, permutation audit trail.
Pro
Priced after a short conversation
Dedicated verifier, custom load metrics, SSO.
monotonic's audit trail (rho, p-value, permutation seed) is built for the evidence bundle SOC 2 wants. Formal attestation lands Q4 — early customers get the pre-release evidence pack now.
Runs well with
