How this work conceives and tests artefacts, proposals, and applications (a "DNA-economy cell", a coordination artefact, a candidate move). It is the disciplined, non-firehose alternative to an autonomous adversarial loop — and, because this is a Living Epistemic Work, the method and every run are published in full. The verdict is never handed down as a conclusion; the working is shown, including where the case was wounded and where it leapt. That is what "openness" means here: the evaluation apparatus is as transparent as the claims it evaluates (siblings: /objections, /attack).
Why this and not a loop
An autonomous adversarial loop optimises for survives-the-attack, not for truer / more alive / in-voice. Unsupervised it over-hedges (death by a thousand concessions), entrenches the frame, and drifts from voice. And it firehoses — volume defeats assessment. The Six Hats (de Bono) fix both: they sequence generation and critique so neither dominates, and Blue is the selector seat — so the human audits a small, structured synthesis, not a pile. (Full reasoning: see the memory thread on recursive self-improvement.)
The fitness function
Every hat measures against the same standard: Fuller's mission as CE's viable objective — prefer configurations that bring greater life to all and less to none, without ecological offence. No-regret over moves, not maximise-a-number. Each candidate is gated by the CE critics: displaced-costs, substitution-limits / binding-constraint, asymmetry. CE is generative and possibility-sized — we want directions of movement and best-next-actions, not legal certainty.
The sequence (de Bono ordering, yellow before black)
- Blue (scene) — frame the question, the fitness function, what each hat should focus on.
- White — facts and data only; anchor or correct the proposal; confidence flagged on every figure; web-checked; never invented.
- Yellow — the disciplined optimistic case. Is it even worth doing? Lead with possibility (yellow before black, so a real idea isn't strangled in the cradle).
- Black — failure modes; attack this specific yellow case (not generic pessimism); rank by how badly each wounds it; name which CE gate fails.
- Green — creative ways through; take the specific black wounds and leap to better configurations; antifragility done right (attack absorbed, then leap). Name walls that genuinely don't yield in-horizon.
- Red — the factors logic can't price (tech-rate, behaviour, momentum, "people being dicks", tailwinds). A briefing for a gut, not a verdict.
- Blue (close) — reconvene, sequence the best-next-actions, name the binding constraint, state the CE-honest verdict.
The orchestration (how the swarm runs)
A Workflow (deterministic): each substantive hat is a small swarm (3 agents competing truest-in-mode — insight, not intensity; caricature is discarded), then a Blue conductor synthesises the swarm into one hat-output before the next hat sees it. State threads forward via an accumulating dossier, so Black attacks this Yellow and Green solves these Black wounds — a relay, not six disconnected runs.
The two human seats
Red (the gut) and the final Blue (the verdict) stay human. The agents do the legwork — facts, the case, the wounds, the leaps, the non-logical briefing — and the human makes the call. This is the line we hold between what is delegable (White/Yellow/Black/Green, and Red-as-briefing) and what is irreducibly the author's (the gut, and the decision). It is also how the firehose is tamed: the human reads a structured six-part synthesis and wears two hats, not a pile.
Worked example
NOTES-ogwen-six-hats.md — the first full run, on the Ogwen DNA-economy cell. It is the proof the method produces decision-grade output: White corrected the proposal against real data, Black found a genuine fracture (the upregulation engine fails arithmetically at Turn 1→2), and Green leapt to a re-sequencing that survives the wounds (break "N funds N+1"; the cell owns the pattern, legibility and aggregated demand, external balance sheets own the atoms; the network is the capstone, not the leap of faith; sell the pattern).
Possible next step (not done)
This method + its runs are candidates for a public /method or /evaluations surface — openness made navigable: anyone can see how a proposal was tested, hat by hat, and where the human judgment entered. Flagged, not built.