Debtaur

Deterministic CPU reasoning engine — bounded agent, verifiable derivations, refusal over improvisation.

Run a case

A — complete

All premises present. Should proceed and call the register tool.

Run ▸

B — missing premise

One required premise absent. Should refuse and name it.

Run ▸

C1 — contradiction

Conflicting stated premises. Should escalate, no tool.

Run ▸

C2 — provenance risk

A taught interpretation treated as legal text. Should escalate.

Run ▸

Pick a case above. Same input yields the same evidence, every time.

Benchmark (measured on the production box, 2026-09-01)

Corpus	Lines	Props	Ingest	lines/s	Warm query
college	3,182	2,940	0.35s	9,078	0.3ms
fairy	103,598	13,973	16.5s	6,291	0.2ms
fable	197,662	12,445	28.4s	6,964	0.1ms
mixed	311,543	12,907	49.2s	6,335	0.6ms

311K-line corpus ingested in 49s; steady-state auditable reasoning stays bounded and sub-millisecond. Reasoning depth is decoupled from corpus size. Reproduction: make scale. The engine runs deterministically on CPU — no GPU, no weights, no hallucination; every conclusion carries its derivation and source tag (stated / derived / taught). This public page is a self-contained mirror; the real engine and cited rule set are held privately.