The evolution of the six-hat method from a single run into a loop — and itself an artefact: a meta-tool for making tools. Where the six-hat run is one deliberation at one framing, this loops, and each loop's job is not to decide but to expand the space.
Search, not optimisation (the commitment)
A choice from CE, and a position Tom holds with Forrest Landry: optimisation picks the least-bad option from a fixed set — which accepts the trade-offs that set contains and crystallises the space at its current shape. Search expands the set until a solution with no trade-offs becomes reachable. The claim is not that we reach it immediately, or this year; it is that increasing the search space reliably yields better approaches, and that a trade-off is a hypothesis that the frame is too small, not a law of nature.
This is CE applied to its own method. The active constraint on what is reachable is pattern-intelligence — the search — not physics (pattern-intelligence-constraint); the reachable set grows with search (configuration-generates-configuration); and the right stance is to keep the game open rather than take the game-ending move of deciding too early (infinite-game). Optimisation is the throughput / finite-game move in method form. Search is the method made self-consistent with the theory.
The boundary that gives it teeth
Not naive techno-optimism. Search dissolves the trade-offs that come from framing, pattern, and coordination — the vast majority. It does not dissolve the physical invariants (a conservation law, the second law, a true material floor — binding-constraint). So the discipline: treat every trade-off as dissolvable until it is shown to bottom out on a physical invariant. Relentless about the frame; honest about the floor.
The loop (one pass)
- Re-level (first, every pass). Is this the right altitude? What is the real goal this frame is a proxy for? Go up a level if the frame is too low — "a better combustion engine" → "make the trip"; "self-driving cars" → "move me and my requirements from A to B". Re-levelling is itself a search-space expansion, and it dissolves trade-offs no within-level search can touch.
- Expand. Generate configurations that dissolve trade-offs rather than trade them off — across personas, cross-domain, "what if the constraint isn't real". This is the anti-box step.
- Find the walls. What still forces a compromise — and for each, is it a frame-artefact (re-expandable) or a physical invariant (the floor)?
- Dissolve / decide. Dissolve the frame-artefacts by expanding or re-levelling again; name the single biggest open one as the next pass's problem.
- Ledger + loop. Continue only while open frame-artefact trade-offs remain and the pass genuinely expanded the space. Stop when the ledger bottoms out on physical invariants, or a pass stops yielding — never stop at a good-enough compromise (that is the finite move that crystallises the space).
The box, and the role split
The agents are phenomenal, and their training is the box: left alone, a search loop run by box-bound agents expands within the box and converges on the sophisticated-conventional, majority-of-humans answer. So the loop carries explicit anti-box machinery (cross-domain wildcards; "what is everyone assuming that needn't be true"), and — the load-bearing part — it surfaces, explicitly, where it could not break its own box, and hands those points to the human. The agents do the patient, persistent, positive search legwork; the human is the irreducible box-breaker who injects the genuinely different frame. Neither does the other's job. (This is the selector lesson again: the firehose is cheap; the scarce thing is the move that escapes the distribution.)
Bounded, and surfaced
Run bounded (a few passes), then surface the shrinking trade-off ledger and the box-stuck points to the human, who decides whether the goal is worth more energy. "Loop until solved" is the commitment; "loop until you've taken the next honest step, then hand the human the edge" is the practice.
First outing
See NOTES-iterative-search-mobility.md — the loop run on the car problem, fed the residual trade-offs of the autonomous-fleet evaluation and built to re-level "cars" up to "mobility" on its own.