Deviation-scoring agent
Escape generic
LLM gravity.
deviate scores every artifact against a generic-LLM baseline across four deterministic engines — evidence density, orthodoxy rupture, memory residue — before it ships. No model calls, no randomness: the same text gets the same score every time.
[higgsfield: a gold particle escaping a dashed orbital ring on a black field, thin engraved line art, noir studio lighting]
deviate · 4-engine gravity scorer · deterministic corpus v1
Why teams install
What deviate earns back
Kills the “sounds like ChatGPT” problem
Scores anchor (evidence density) and rupture (orthodoxy-breaking) against a fixed corpus — not a vibe check from a second model.
‹b6fccee2›Deterministic, not a coin flip
Pure regex/lexicon density scoring. No model calls, no seeds-as-RNG — the same artifact scores identically every run.
‹831c2260›A real gap between rich and filler
Seeded rich artifacts clear 100.0 / 94.1 on anchor / rupture; filler artifacts flatline at 0.0 on both — every run, same margin.
‹1e7d048f›How it works
Three steps. No faith required.
01
Score
Run the artifact through 4 engines: graviton (gravity escape), anchor (evidence density), rupture (orthodoxy-breaking), echo (memory residue).
02
Discriminate
Compare anchor and rupture against the seeded 60/40 threshold — a discrimination margin, not an absolute scale.
03
Gate
Ship only if both anchor and rupture clear 60. Anything that reads under 40 on either engine is blocked.
Proof band
One packet. Every claim accounted for.
packet rig/deviate · corpus rich-3 claim "evidence+rupture-seeded artifacts score ≥60; filler scores <40 — deterministically" anchor 100.0 · margin +40.0 over 60 floor rupture 94.1176 · margin +34.1176 over 60 floor filler 0.0 anchor · 0.0 rupture — both floors held gate PASS — discrimination verified across 5-rich/5-filler corpus themis SEALED — ef33e344
Every deviate 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 an AI-detector?
No — deviate doesn't guess whether a human wrote it. It measures evidence density and orthodoxy-rupture against a fixed corpus, deterministically. Same text, same score, every time.
Why 4 engines instead of one score?
Generic AI failure modes aren't one thing — they're missing evidence, safe conformity, and forgettable prose. graviton / anchor / rupture / echo isolate each one so you know exactly what to fix, not just that something's off.
Why trust the number?
Every run seals a ProofPacket — hash-chained, verifier-signed, queryable from your dashboard. The scoring code and the seeded corpus behind the audit trail are the same ones behind the score, not a black box.
Buyers
Who runs deviate
Head of Content
B2B SaaS · 40-person content team
VP Editorial
media brand · 12 writers
Brand Lead
agency · 200+ client deliverables/mo
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
100 artifacts a month scored. Public scorecard, full engine breakdown.
Team
Priced after a short conversation
Unlimited scoring, private corpus, CI gate. priced after a short conversation for the first 100 customers.
Pro
Priced after a short conversation
Dedicated verifier, custom lexicons, SSO.
Early customers keep the terms they sign; team pricing is set after a short conversation. No countdown timer — once slot 100 fills, the price changes for everyone after.
Runs well with
