Public deliberations

Every deliberation
is a complete record.

Every Augle session is published in full — the question, the finding, the confidence grade, every agent's contribution, and every Contrarian objection. Not summaries. The full structured record.

Browse the interactive session log
Corpus · illustrative
33
Sessions
11
Domains covered
38%
Contested or Gap
93%
Avg Guardian integrity

The record captures the ensemble's confidence grade, agent contributions, dissent, and Guardian integrity for every session.

Filter
All domainsLife sciencesPolicyEconomicsTechnologySocial science
All gradesEstablishedProbableContestedGap
Featured session · full record

What a complete deliberation record looks like.

Life sciences · Deep depth · Healthcare
"What does the evidence establish about the comparative effectiveness of GLP-1 agonists vs. bariatric surgery for long-term weight maintenance?"
Deep · ~40 minHealthcare
Finding · Phase 3 synthesis
Contested46%

The evidence does not support a directional comparative-effectiveness claim. Bariatric surgery shows comparable-to-larger and more durable weight loss in long-term observational cohorts, but no large long-term head-to-head randomised trial exists — the comparison rests on indirect evidence across materially different populations. The finding is Contested: the dispute is named and both positions preserved, rather than resolved into a false winner. Reopen condition: a head-to-head randomised trial with ≥3-year follow-up would move the grade.

Phase 1 · Exploration

Cartographer maps settled ground (both interventions produce clinically significant short-term loss), contested terrain (relative durability beyond three years), and unknown territory (outcomes in matched long-term populations).

Phase 2 · Deliberation

Methodologist caps the comparative claim at Contested — no large long-term head-to-head RCT, indirect comparison only. Contrarian raises a Strong objection on population non-comparability. Guardian authenticates all 22 citations and flags one industry-funded extension.

Phase 3 · Synthesis

Synthesizer produces a Contested finding that names the dispute and preserves both positions. Pragmatist declines a directional recommendation and proposes a follow-on session scoped to the head-to-head evidence gap.

Unresolved objections · preserved verbatim
Strong"No large long-term head-to-head randomised trial exists. The comparison rests on indirect evidence across cohorts with materially different baseline BMI, age, and comorbidity profiles — the effect estimate is not transportable between populations."
Moderate"The durability evidence for surgery extends further only because the intervention is older. Longer follow-up is not the same as superior durability — the comparison is confounded by evidence age."
Session metadata
VerticalHealthcare
Guardian flags1 — industry-funded source
Citations verified22 of 22
Dissent flags1 Strong · 1 Moderate
Reopen conditions3 defined
FindingContested
Agent confidence scores
Cartographer
64%
Methodologist
58%
Contrarian
37%
Synthesizer
49%
Pragmatist
45%
How it works

From session to
published record.

1
A session runs on a research question

A session addresses an open research question — academic, policy, legal, or clinical. The question and its scope are locked to the session record before deliberation begins.

2
The full deliberation is logged with provenance

Every agent output, every evidence node, every Contrarian objection, and every Guardian flag is written to the session transcript with a timestamp and agent attribution. The record is immutable after session close.

3
The finding is produced with a confidence grade

The Synthesizer produces an evidence-anchored finding graded Established, Probable, Contested, or Gap — bounded by the Methodologist's confidence ceiling. The grade is the finding, not a footnote to it.

4
Reopen conditions are attached

Each finding carries structured reopen conditions: the specific new evidence, publication, or replication result that would move the grade. Uncertainty is made explicit, not hidden.

5
The full record is published

Sessions are published in full — question, finding, confidence grade, agent outputs, Contrarian objections, and Guardian flags. Every claim in the record is auditable against the source transcript.

6
The corpus accumulates as a structured asset

Every session adds a structured record to the reasoning corpus: multi-agent deliberation, adversarial challenge, and calibrated confidence across domains. It is a dataset of how the system reasons — it exists only because the sessions ran.

The reasoning corpus

The deliberation record
is the asset.

Every session adds a structured record to the reasoning corpus — multi-agent deliberation, disagreement, adversarial challenge, evidence weighting, and calibrated uncertainty across research questions. At scale, this constitutes a dataset that does not exist anywhere else: how a structured ensemble reasons, argues, and quantifies what it does and doesn't know.

View the Deliberation Index →
Beta launchFirst live sessions
Day 90Deliberation behaviour visible
Day 180 — 2,000–5,000 sessionsDomain coverage broadens
Day 365 — 10,000+ sessionsResearch-scale corpus
Year 3 — 1.5M+ sessionsLargest structured AI deliberation dataset

Add your question
to the record.

Join the waitlist — every session you run becomes part of the deliberation record.One free Standard session with every new account.