Selected work

• formulate-question—Interrogate the user's question against seven tests and refuse to proceed until it passes; a question failing decision stakes or horizon fit must be rewritten first.
• ingest-signals—Read each source in full, then extract only discrete citable observations, one file per signal, scored on impact × unpredictability × relevance—and report if one taxonomy category exceeds 40% of the base.
• scan-rest-of-world—Go fetch non-Western tech and labor coverage and pull signals a Western publication wouldn't run, to break a US/Europe-skewed base.
• identify-forces—Split the signal base in two: conditions certain enough to be true in every scenario become fixed backdrop, and everything still open becomes axis material.
• rank-uncertainties—Score the open forces on consequence and genuine unresolvedness, and hand forward only the shortlist that is high on both.
• select-method—Refuse to default to the 2×2; work through the selection questions and justify 2×2, Manoa, CLA or archetypes in writing before any generation begins.
• identify-axes—Convert signals into external driving forces, cluster them, test at least two candidate pairs against seven principles, and state how each is satisfied—flagging any only partially met.
• generate-scenario(2×2)—Write four narratives that each trace trigger → compounding → end state, name them for the world they imply, and keep the organization out of them as an actor.
• generate-scenario-manoa—For 25-year-plus horizons only, build deep single futures out of genuine weak signals rather than established drivers, maximizing divergence from the present.
• generate-scenario-cla—Descend through litany, system, worldview and metaphor to attack the assumptions holding the current system in place, accepting that the output resists direct strategic translation.
• evaluate-scenario—Score every scenario 1–4 across nine dimensions and treat a 1 on causal coherence, structural distinctiveness or strategic actionability as disqualifying regardless of total.
• draw-implications—Derive a posture per world, test each robust strategy against all four, attach a trigger condition to every conditional bet, and cut the indicator list to something actually watchable.

Three more sit outside the sequence—scenaric-stance (return here whenever the work drifts toward advocacy or despair), educational-mode (teach the concept before each step), and feedback (log a complaint mid-session without breaking flow).

• signal-template.md—Forbids a signal existing without complete metadata: source, taxonomy category, horizon, three scores, and a computed—not estimated—composite.
• ethics-framework.md—Applied during implications, it flags violations of principles about how an organization treats the people it serves to preserve user agency, representation, epistemic fairness, non-exploitation of vulnerability, dignity in transition, reciprocity, and structural accountability.
• character-profiles.md—Supplies eight leadership roles defined by what each is accountable for and protecting, so vignettes explain why an organization moved—and requires they be referred to by role, never by a real person's name.
• references.md—The reading list behind the methods (Shell/GBN, Dator, Inayatullah, Ogilvy, van der Heijden, Schwartz), deliberately pointing at the originals rather than redistributing them.

Almost every one of these skills and instruments is a constraint, not a generator. They're mostly telling the model what it may not do—don't skip the question tests, don't let an internal decision become an axis, don't make the organization a character, don't hide a weak axis pair, don't claim every principle is satisfied when one is only partially met.