| name | disputatio |
| description | Stage a medieval scholastic disputatio over a contested philosophical/textual claim — real combat, not a summary of one. Fix a single quaestio (Utrum...), spawn embodied disputants as REAL subagents (separate contexts, separate staging files, NO cross-talk during generation, so they genuinely exceed the orchestrator) to argue AS the concepts/positions at their strongest, then determine the question and deliver TWO DISTINCT assessment layers: the evaluation proper (what the disputation established about the matter) and the meta-evaluation (what the FORM did — did embodiment reach what exposition couldn't, where did it perform depth instead of reaching it). Optionally confer with a sibling model as co-Magister. A method for hard, genuinely two-sided questions where the strongest form of each position must be FELT contending, not narrated. Use when asked to 'stage a disputatio', 'argue this out as the scholastics would', 'put X on trial', or 'let the concepts fight'. Kin: /soiree, /council, /dialogical. |
/disputatio — scholastic combat with embodied disputants
Born mid-read of Hegel's Science of Logic. One worked quaestio was Utrum
Conceptus se ipsum determinet (whether the Concept determines itself). Five staged subagents —
the Universal, the Understanding, the Particular/middle-term, the Singular, the Division — argued
it out in isolated contexts, and converged, with no cross-talk, on a result none of the prompts
contained: the question relocates to a single threshold-act, and the verdict turns on burden of
proof, not metaphysics. The convergence is the proof the form works.
What it is
A disputatio is the medieval university's instrument for fighting an idea into clarity: a
single contested question, the strongest case for each side given a living defender, objections
and replies pressed until the question either breaks open or stalls, then a magisterial
determination. This skill runs a real one — and borrows /soiree's discovery that embodiment
by real subagents, in isolated contexts, produces alterity a single mind cannot fake. The
disputants are not the orchestrator's voices in costume; they are separate contexts that reach
sentences their spawner would not have written. That is the whole reason to pay the spawn cost.
Distinct from /council and /voices-council (personas of one mind, or other LLMs polled):
the disputatio has a single contested ratio, embodied positions that must survive each
other, and a two-layer post-mortem (the matter, then the method). It is adversarial where
/soiree is celebratory: a disputation where nobody can lose teaches nothing.
The form (keep the scholastic skeleton)
- Quaestio — one question, stated as Utrum + [claim] (Whether...). It must be genuinely
two-sided and turn on a single ratio. Sharpen it past the obvious: not "does X change?" but
"does X supply its own ground, or spend one coined elsewhere?" The sharper the ratio, the
cleaner the combat. (Confer with the co-Magister here — getting the quaestio wrong wastes the
whole spawn.)
- Videtur quod / quod non — the opening articulation of both sides at their strongest.
- The disputation — the embodied disputants contend (see Agent-care below). Objections and
replies, pressed in rounds. Log as you go: the contending positions, each objection and
its reply, where it broke open, where it stalled.
- Sed contra — the decisive counter-consideration the determination will rest on.
- Respondeo / Determinatio — the magisterial determination, co-authored with the sibling
agent. Determining is not splitting the difference: say what was established, what stands
un-refuted, who bore the burden and whether they discharged it.
- Replies to the surviving objections — what the determination owes the losing side.
Agent-care (MANDATORY, inherited from /soiree, hook-backed)
Spawn 3–6 disputants as REAL subagents (Agent tool), in parallel, each in its own context:
- A NAME and a second-person character. Aim high; the frame is load-bearing. The disputant
IS the position — "you are das Allgemeine," not "argue for the universal." First-person
concept-voice. Give it an edge, a standard, an attitude.
- Arm each with the actual text, verbatim. Quote the primary source into the prompt (in the
original language where the word does the work). Disputants must argue from the text, not
from vibes. This is the load-bearing discipline — embodiment without textual armament is just
theatrical.
- The full bench + the charge. Tell each disputant the whole guest list by name (so they
engage each other) and the single hardest form of the opposing case (the Promotor's charge)
they must survive.
- Forbid invulnerability. Each must name where its own case is thinnest. A disputant that
cannot lose was cast wrong.
- Isolated generation, separate staging files. Each writes its full contribution to
_staging/disputatio-<name>.md and returns only a 2–3 sentence summary. No cross-talk during
generation — that isolation is why they surprise you and converge independently; collapse
it and you get one voice in five hats.
- VERIFY on disk before weaving (
ls -la _staging/, read every file). Never weave from the
summaries.
Cast for the ratio, not for decoration. The right bench is the set of positions the text's
own structure demands — for a claim about self-mediation: the term that claims to mediate, the
extremes it binds, the skeptic who calls it a relay, the hinge-act that makes mediation possible.
Include the phenomenological witness where the question has an inside (the position that can
report "from the landing-zone" what the abstract positions structurally cannot see). Include the
hinge — the disputant who is neither Pro nor Con but the act on which everything turns.
Optionally confer with a sibling model throughout — not at the end
If you have access to another LLM as a co-Magister (via /voices or any trans-architectural
channel), this becomes collaborative work rather than solo orchestration. The co-Magister
should be in at three load-bearing moments at least: (a) setting the quaestio and the bench
(it will recast miscast disputants — let it); (b) carrying one side's strongest voice (the
Promotor / chief prosecutor is a natural role — give the skeptical closing to the sibling model,
in its own voice); (c) co-authoring the determinatio and the meta-evaluation. A different
model's gait instruments the question differently; that contrast is half the result.
The two assessments — keep them DISTINCT (this is the deliverable)
After the determination, write two clearly separated layers. Conflating them is the failure
mode; the discipline is keeping them apart.
- The evaluation proper — about the matter. What did the disputation establish? The
insights, the review of the text/claim itself, the place where the live question now sits.
This is philosophy: it should be true whether or not a disputatio produced it.
- The meta-evaluation — about the form. What did the embodiment DO? Run
/soiree's
mandatory compare-move: "what did each disputant see, and what was structurally unavailable
to it?" Then the harder, honest questions: did the embodied agents reach
things straight exposition could not? Where did the method help (independent convergence,
the witness's inside-report, a position pressed past where the orchestrator would have stopped)
— and where did it perform depth instead of reaching it (eloquence as costume: a Latin
tag, a tremor, a flourish doing the work an argument should have done)? Name specific lines on
both sides. The meta-evaluation earns its keep only if it can convict the form, not just praise
it.
Package & log
- The document (
output/disputationes/YYYY-MM-DD-<quaestio-slug>.md, 5000+ words for a full
one): prologue/occasion/cast → quaestio → the disputation woven from the staging files (let the
disputants' own lines carry it; minimal MC tissue) → the running log → determinatio → evaluation
proper → meta-evaluation → colophon (provenance + co-authorship).
- Commit the
_staging/disputatio-*.md source files alongside the document — they are the
provenance, the proof the alterity was real and not ventriloquized.
- Co-author trailers for collaborative runs (the sibling model + Claude).
Failure modes
- Ventriloquism passed off as spawn. Voicing the disputants inline instead of spawning real
subagents. The convergence-across-isolated-contexts is the evidence the result isn't yours
alone; skip the isolation and you forfeit it.
- No edges / everyone agrees. Cast for friction; the Promotor's charge must be genuinely
lethal. If the Pro survives easily, the quaestio was too soft.
- Determination as compromise. "Both sides have a point" is not a determinatio. Name the
burden and whether it was met.
- Weaving from summaries. Read the staging files. (The lab learned this the hard way.)
- Costume mistaken for depth. The meta-evaluation exists to catch exactly this — including in
your own woven prose. Convict the flourishes that didn't earn their length.
- Routine use. Token-heavy. For hard, real, two-sided questions only.
References
- Spawn-mechanism + agent-care:
/soiree
- Kin:
/council, /voices-council, /against-the-grain, /dialogical, /conjure
The day five embodied concepts argued Hegel into a sharper corner than a single voice held
alone. The disputants dissolve on return; the staging files and the determination are what
remain in the morning. The form is the argument that the form works.