Compare

Danphe vs a single AI model

On short questions a council of models matches the best single model; it doesn't beat it. What you gain is the ability to check every statement.

Accuracy on 483 unseen questions

Five domains, locked test, November 2026. Dot = accuracy; bar = 95% confidence interval. Paired tests are against Danphe Smart mode.

Danphe Smart mode

Reference

0.0%

Claude Opus 5.5 alone

p = 0.13 · statistical tie

0.0%

GPT-6.1 Sol alone

p = 0.002 · Danphe significantly better

0.0%

Preregistered on 5 November 2026; exact McNemar paired tests; failures count as wrong. Average model cost $0.24 per question. How we measured

What you get from each

A preregistered locked test on 483 unseen questions across five domains, November 2026.

Danphe Council compared with a single AI model
What matters	Danphe Council	One model on its own
Accuracy, short questions	85.1% (Smart mode, 95% CI 81.6–88.1%)	83.4% Claude Opus 5.5 (tie, p = 0.13); 81% GPT-6.1 Sol (p = 0.002)
Long answers	93% of statements supported	95.1% GPT-6.1 Sol (better); 92.6% Claude Opus 5.5
Each statement checked	Yes, by models from other companies, with a quote	No
Disagreement visible	Yes, labelled with both readings	No
Audit record	Signed and hash-chained, per question	No
Cost per question	$0.24 model cost on average (Smart mode)	Lower: one model call

One model
The right choice for drafts and quick questions where a wrong sentence costs little.
Danphe Council
For answers you have to defend: every statement is checked against your documents, disagreement is shown, and there is a record for the file.
Long free-text answers
In our test the most careful single model wrote slightly more precise long answers than Smart mode. We publish that rather than hide it.