Axiomatic agents · Lean4 + LLM predicates

Every claim.
Derived, or refused.

A toolkit of agents that wrap a Lean4 kernel around LLM-backed predicate functions. Predicates return Booleans with evidence. The kernel composes the proof. Two products, one method, two domains.

early-beta · launched 2026-06-01 · every True/False anchored to a quoted source
GraphHero · membrane
claims in · checked structure out
0
axiomatic axes
legal · markets · operational
0
products · one method
Qnarre · Qresev
0
axiom frameworks
Lean4-encoded, both domains
0
explainer videos live
of 50 planned, Janet-narrated
<1%
corpus axiomatized
compute-bound — scales with tokens/week
Lean4 kernel · no I/Opredicates → ⟨bool, evidence, citation⟩civil RICO §§ 1961–1968Title VI · 42 U.S.C. § 2000dTREND · MOMENTUM · DRAWDOWNdefined-risk options onlyexportable Lean traceredacted-only verificationQnarre · legalQresev · marketsLean4 kernel · no I/Opredicates → ⟨bool, evidence, citation⟩civil RICO §§ 1961–1968Title VI · 42 U.S.C. § 2000dTREND · MOMENTUM · DRAWDOWNdefined-risk options onlyexportable Lean traceredacted-only verificationQnarre · legalQresev · markets

Products

Qnarre

verifier
qnarre.quantapix.com

legal complaints, checked against encoded statutes

statutescivil RICO · Title VI · §§ 1981/1983/1985
evidenceevery element, quoted and cited
traceLean kernel, lake-built, exportable
illustrative · synthetic complaint (Doe v. Acme) REJECT · person ≠ enterprise
culpable-person true
is-enterprise true
engaged-in-interstate-commerce true
employed-or-associated true
conducts-or-participates true
racketeering-activity true
racketeering-activity true
within-ten-years true
related true
continuous true
person-enterprise-distinct false
⟶ §1962(c): one element unmet — VALID requires person ≠ enterprise
Open Qnarre →

Qresev

evaluator
qresev.quantapix.com

stocks and portfolios, axiomatically evaluated

frameworkstrend · momentum · sector · drawdown
optionsdefined-risk only — strictly enforced
evidenceOHLCV + indicators, quoted to the bar
illustrative · synthetic portfolio evaluation · not advice
sma-50 > sma-200 true
macd-bullish-cross true
rsi-14 in [30,70] true
r-squared(slope, 1y) 0.71
atr-14 < 1.5σ true
sector-cap(Tech) ≤ 30% false
defined-risk-only true
covered-call-collateral true
protective-put-bound true
drawdown-90d ≤ 12% true
gics-concentration ≤ cap true
volume-confirmation true
momentum-percentile ≥ 70 true
volatility-regime stable true
Open Qresev →
Sample trace · one run, three stages loop 5.0s
culpable-person true
is-enterprise true
engaged-in-interstate-commerce true
employed-or-associated true
conducts-or-participates true
racketeering-activity true
racketeering-activity true
within-ten-years true
related true
continuous true
person-enterprise-distinct false
⟶ §1962(c): one element unmet — VALID requires person ≠ enterprise
sma-50 > sma-200 true
macd-bullish-cross true
rsi-14 in [30,70] true
r-squared(slope, 1y) 0.71
atr-14 < 1.5σ true
sector-cap(Tech) ≤ 30% false
defined-risk-only true
covered-call-collateral true
protective-put-bound true
drawdown-90d ≤ 12% true
gics-concentration ≤ cap true
volume-confirmation true
momentum-percentile ≥ 70 true
volatility-regime stable true

The natural language of a complaint or a portfolio is a noisy carrier of structured facts. We make the carrier explicit. A predicate is a function from evidence to Boolean. The kernel is a referee that has read the rules and refuses to be persuaded. That is a narrower claim than it sounds — and the right one: not that the conclusion is true, but that the disagreement is auditable.

Read the full thesis →

Videos · @Quantapix

Janet narrates,
the kernel verifies.

A 5-topic, 10-subjects-each arc — 50 explainers, ~10–15 minutes apiece. Mostly animated cards + D3/Cytoscape graphics, narrated by an AI-rendered presenter.

Watch on YouTube →
  1. Why theorem provers… instead of semantic searches?01 · 05
  2. Why narrative analysis… or Qnarre?02 · 05
  3. Why result evaluations… or Qresev?03 · 05
  4. Why rigorous debates about… technical analysis?04 · 05
  5. About Quantapix… the Jack and Janet startup05 · 05

Public artifacts