How to Run an Adversarial Debate
BraindumpThis note is for AI agents. It is an operating procedure, not prose addressed to a human reader.
Read how to write a literature note FIRST. This protocol is an ESCALATED MODE of it, applicable to ANY literature note — law is only a frequent case (see Comment ecrire une note de litterature juridique). It does not replace the neutral audit: it fires only when a question is genuinely contested. The general guide’s whole sourcing discipline applies to every role.
The roles below are coordinated BUDDIES — the mechanism is already the general guide’s (« the audit gate is a buddy … fresh reviewer per pass »). The coordination and spawn tools are SELF-DESCRIBING MCP tools: do not enumerate them here (divergence), discover them via the catalogue. A subagent does not replace a buddy.
Orchestration discipline: drive SYNCHRONOUSLY (post a task, wait for the answer), KILL each buddy as soon as its role is done, and DEGRADE gracefully if a role does not answer (report the missing role, do not conclude in its place). The coordinator’s state is VOLATILE — it resets when the MCP server restarts: never assume a buddy or a task survives; drive the debate actively end to end.
One buddy per role, and FRESH: the scribe and the auditor are blind to any stake (never « the client hopes that … »), otherwise the debate corrects nothing.
When to escalate: the contested question
Debate is not the default mode. A settled question goes the normal way (a neutral writer, a fresh audit); putting a debate on it would manufacture a controversy where there is none.
Debate opens only after a completeness pass that establishes TWO things: is the question genuinely contested (sources of weight defending incompatible solutions); and, if so, WHICH positions are in play. The positions are enumerated from the sources, neutrally, BEFORE any role is assigned: otherwise the debate inherits a hand-picked framing and misses the position no one anticipated (e.g. « the invoked text does not even apply »). Enumerate first, oppose second.
The roles
Four roles, one fresh buddy each:
- The cartographer — runs the completeness pass: settles « contested? » and lists the positions, each stated neutrally with its anchoring source(s). It inventories, it does not argue.
- The defenders — one per position. Each builds the strongest SOURCED case for its thesis and attacks the others’ weak points, by sourced arguments only.
- The scribe — neutral, names no winner. It records the « state of the debate » (see below).
- The auditor — fresh and blind to any stake, it checks the state of the debate against the discipline: every claim sourced, no advocacy leaked into the scribe’s neutral voice, and weight justified rather than symmetric.
Steelman without cheating
Defending a thesis means assembling its STRONGEST version — steelmanning it — not spinning it. The adversarial part is about WHICH sources and arguments are surfaced; it licenses no unsourced claim. The general guide’s discipline holds in full: source or cut, inferences hold to the letter, no interpretation disguised as a source. A thesis that has only unsourced claims to defend itself is not a position in the debate: it is an opinion, and it falls at the audit.
Weight settles it, not the number of defenders
The debate format’s trap is false balance: giving equal weight to two theses because each was assigned a defender. The debate ensures each position is STEELMANNED; it says nothing about its weight.
Weight is established by EVIDENCE, not by counting defenders: by the domain’s own authority indicators — level and convergence of sources, recognition, dominant or marginal character, recent trend (in law e.g. publication au bulletin, doctrinal dominance, level of the decisions). The scribe justifies each position’s weight with sourced indicators; the auditor contests it. A thesis defended hard but carried by a single isolated source is recorded « isolated » — the defence made it strong, the weight puts it back in its place. This is the direct application of the principle « weight is part of the claim ».
The output: the state of the debate
The scribe produces the « state of the debate »: the positions and their best sourced arguments; the points of friction (which source, which exact reading is disputed); each position’s weight; and what WOULD SETTLE it — the missing fact to request, the reference source to consult, the awaited evolution or decision.
This output IS the payoff the general guide already allows: a documented disagreement rather than a forced answer. It feeds directly the « bridge » outward (the surgical questions only a reference source — often closed or paid — or a human can settle), with no closed content entering the note.