AI tools give you answers.
Decisions need an argument.

Perpensa convenes a network of adversarial AI perspectives around your decision — researching as the debate branches, challenging every claim, and keeping the full reasoning traceable from evidence to conclusion.

re melius perpensa — the matter, more thoroughly weighed

The Perpensa confluence Strands representing different AI voices converge into a single decision line.

why perpensa

One model, one pass, one confident answer — that isn't deliberation.

High-stakes, one-off strategic decisions don't fail for lack of information. They fail because the reasoning was never seriously attacked before it was acted on. Research agents hand you a memo; decision-automation platforms handle high-volume operational calls. The deliberative layer in between — where options are weighed, dissent is surfaced, and conclusions earn their confidence — still happens in someone's head or a meeting room. Perpensa is software for that layer.

how it works

Four passes between a question and a defensible decision

01 · branch

Branching research

Research that reacts to what it finds — forking into the questions that matter instead of marching through a fixed outline.

02 · debate

Adversarial deliberation

Contrasting expert voices argue the options. Disagreement is treated as signal, and new evidence is summoned mid-debate.

03 · challenge

Red-team pass

A dedicated challenger attacks the emerging conclusion — assumptions, incentives, failure modes — before reality does.

04 · trace

Reasoning lineage

Every conclusion carries its provenance: which evidence, which arguments, which challenges it survived. Defensible by construction.

inside a deliberation

You're in the room, not on the sidelines

Interject, redirect, summon evidence, or overrule — the voices work for you. This is what a live exchange looks like:

Market entry · branch 3 of 5 red team active

Advocate — pricing power holds if we anchor on validated reasoning lineage rather than feature parity.

Skeptic — lineage is invisible before purchase. Anchor on outcomes a buyer can verify in a pilot.

Analyst — pilot-conversion data from adjacent categories supports the skeptic: proof beats promise 3 to 1.

RT-03 · challenge

All three anchors assume a rational evaluator. What does procurement reality say — who actually signs, and what do they fear?

claim:4.2 ← evidence:17 ← challenge:RT-03 · 2 resolved · 1 open lineage: traceable end to end

the network

Seven kinds of voice, one shared record

Each perspective has a stable identity and color across the debate, the visualizations, and the final artifact — so you always know who said what, and why it's in the record.

AdvocateBuilds the strongest case for each option
SkepticFinds the weight the case can't carry
Red teamAttacks the conclusion before reality does
AnalystBrings the numbers, mid-debate, on demand
SynthesistFinds the position the debate is converging on
PrecedentSurfaces how similar decisions actually went
ModeratorKeeps the deliberation honest and moving

Evidence platforms

Search, aggregation, synthesis, cited memos. They answer: what does the evidence say? Perpensa consumes their output happily.

Perpensa

Deliberation, challenge, traceable reasoning. It answers: what should we decide — and can the reasoning survive attack?

market intelligence ends where perpensa begins

status

In development

Perpensa is being built in the open-ish. If deliberation-grade decision support is your problem too — as an operator, a design partner, or just a fellow traveler — say hello.