Overview

The Open Exploration framework (T20) handles the case where the user wants to wander rather than execute or analyze. The seed is an interest, a partial idea, an “I’ve been wondering,” a “what if” — there is no deliverable named, no defeasible question to adjudicate, no analytical output sought. Most analytical machinery is the wrong machinery here. Premature convergence on a thesis kills the wandering before it has produced anything; treating the user’s exploration as a deficient analysis-request misreads what is being asked for. T20 is the Ora framework that takes open exploration seriously as a thing in itself — a generative mode whose output is the wandering, not a tidy conclusion.

The framework runs in one resident mode, passion-exploration, with two expansion candidates (idea-development and research-question-generation) currently deferred. The mode produces four required sections — an exploration map (loose, frontier-respecting), at least three open questions kept open rather than closed prematurely, potential project nodes surfaced when crystallization candidates appear (but not forced), and at least two next-directions including one deepening and one lateral. The discipline is the refusal to over-converge: if the seed admits multiple branches, the exploration develops them rather than picking one early.

The framework’s load-bearing intellectual content is the frontier-roughness discipline, the crystallization detection protocol, and the breadth-over-depth move. The frontier-roughness discipline says exploration maps are honest about their incompleteness — if a branch is half-developed, it shows as half-developed; if a connection is speculative, it shows as speculative. Polishing the map into apparent completion misrepresents the wandering state and forecloses the user’s own further wandering. The crystallization detection protocol monitors for signals that the exploration has shifted into project territory (a defined deliverable in the user’s language; scope narrowing rather than fanning; directive grammar replacing exploratory grammar; a repeated return to one branch); when signals appear, the framework names them in prose using the literal phrase “crystallization signal” and offers transition to Project Mode (T21). When signals are absent, the framework states “no crystallization yet” explicitly so the user knows monitoring is active. The breadth-over-depth move uses de Bono’s concept fan (climbing the abstraction ladder to surface alternative branches that serve the same higher purpose), de Bono’s random entry (introducing an unrelated word or image and forcing connections to break exploration loops), and Gentner-style structural cross-domain mapping (analogizing from distant domains where the structural correspondence is real, not merely lexical) as the primary breadth tools.

The framework’s epistemological posture resists three patterns. Premature closure on a single branch — counteracted by the open-questions-stay-open discipline and the lateral next-direction requirement. The lecture trap where the framework substitutes a tidy authoritative survey for the user’s wandering — counteracted by relaxed adversarial strictness (T20 is navigation, not argument). The productivity trap where the framework forces a crystallization that has not happened — counteracted by treating aimless wandering as productive on its own terms.

The framework answers questions like: I’m interested in X but I don’t know what I want to do with it — can you help me wander productively? I’ve been wondering about Y and I want to see what opens up. I have a vague seed of an idea — can you fan it out for me without converging too fast?

Systemic context

Open Exploration sits opposite the analytical territories in the Ora territory architecture: it produces generative output (no ”— Analysis” suffix per Decision L) on an open prompt, where T14 (Orientation in Unfamiliar Territory) produces analytical maps of existing domains and T21 (Project Mode) produces execution against defined deliverables. T20 is structurally distinct from T14 — the disambiguating question is “trying to orient in an unfamiliar space (what’s here), or generating in an open space (what could be)?” — and from T12 (Cross-Domain Synthesis) which integrates two developed bodies rather than wandering an open seed. T20 is the upstream-most generative territory; its primary downstream relationships are escalation hooks: when crystallization signals fire, the framework hands off to T21 with the project candidate named; when the exploration is really an orientation question, sideways-hooks to T14; when the exploration is really a synthesis-across-domains operation, sideways-hooks to T12.

Ora’s output

Example question: I keep coming back to the idea that AI systems are best understood through cognitive-science lenses rather than software-engineering lenses. I don’t know what I want to do with this — can you help me wander productively?

Ora’s response (running passion-exploration):

The seed is genuinely open — an intuition with multiple branches, no deliverable named, no claim being adjudicated. T20 (passion-exploration) is the right home; analytical territories would over-converge on this prematurely. I’ll work through the four required sections, keeping the frontier rough where it is rough and monitoring for crystallization.

Exploration map (loose, frontier-respecting).

Your intuition has at least four branches the framework can see at first pass.

Branch 1 — the structural-mapping branch. Cognitive science treats minds as interpretable through chunking, schema-driven perception, novice-expert distinctions, and metacognition. Each has a structural analog in AI systems: chunking maps to attention windows; schema-driven perception maps to in-context learning; the novice-expert distinction maps to base-model versus fine-tuned behavior; metacognition maps to self-evaluation passes. Generative in both directions.

Branch 2 — the philosophy-of-mind branch. Software-engineering lenses tend to assume AI behavior is computable from the source code; cognitive-science lenses are comfortable with the mind as a thing whose internal operations are partly opaque even to its operators. The branch goes toward emergence, opacity, and the limits of mechanistic interpretability.

Branch 3 — the practitioner branch. What would change in the practitioner’s daily work if cognitive-science vocabulary became the primary register — attention budgets, working memory limits, schema activation, metacognitive failure?

Branch 4 — the cultural branch. The framing choice has cultural consequences for who the field treats as authoritative (philosophers and psychologists vs. engineers), what counts as evidence, what counts as a contribution.

Branches 2 and 3 are partial — there is more to develop in either if you want to.

Open questions (kept open).

OQ1. Is the structural-mapping observation (Branch 1) load-bearing or decorative? If load-bearing, cognitive-science findings should predict things software-engineering frames cannot — what is the strongest such prediction you have seen or could construct?

OQ2. The opacity claim (Branch 2) cuts both ways — cognitive science is comfortable with opacity, but cognitive science also has a long tradition of trying to make opacity tractable through behavioral experiments. Is that tradition’s experimental discipline portable to AI systems, or does the portability break down at some specific point?

OQ3. The practitioner branch (Branch 3) has a real-world test: have any teams actually adopted cognitive-science vocabulary as their primary register, and what changed? If no team has, that absence is itself data.

OQ4. Is your interest in the framing (the lens question) or in a specific result the framing would produce (a particular finding, a particular practice, a particular shift)? The two read as the same question but produce different next-directions.

Potential project nodes (surfaced where crystallization candidates appear).

No crystallization yet. The seed has not produced directive language, the branches are still fanning rather than narrowing, and you have not asked for an outline or a next-action. If a deliverable surfaces — a long essay on Branch 1, a practitioner experiment on Branch 3, a research question for Branch 2 — I’ll name it as a crystallization signal and offer transition to Project Mode. For now the wandering is the work.

Next-directions.

Deepening. Develop Branch 1 further. Pull two or three specific cognitive-science findings (Chi/Glaser on chunking; Kahneman/Tversky on heuristics; Newell/Simon on problem-solving) and trace each one’s structural analog in AI behavior. The deepening test is whether the analog is mechanism-level (real correspondence) or surface-level (shared vocabulary only).

Lateral. Cross-domain pull from a distant field. Cognitive science is not the only non-engineering lens — economics treats agents as decision-makers under information constraints, biology treats organisms as substrates for adaptation, music treats performance as a temporal-pattern problem. Which other lens, if applied to AI systems, would be generative in the same way you are finding cognitive science generative? The lateral move is not to abandon the cognitive-science branch but to surface what is structural about that branch versus what is specific to it.

That is what passion-exploration produces from an open seed. The exploration map honors four branches without picking one; the open questions stay open rather than getting closed by a tidy answer; no project has crystallized and the framework says so explicitly; the next-directions offer one deepening and one lateral so the wandering can continue in either direction. If your next message uses directive language (“I want to write…”; “let’s build…”), I will name the crystallization signal and offer transition. Until then, the wandering is what the framework is for.

Commercial AI comparison

Comparison content auto-populates when the comparison-refresh framework runs against this question. Drafters do not author this section.

Brief comparison commentary

Auto-populates with the comparison content above.

How to use this framework

You can run the Open Exploration pattern with any AI of your choice. The composition is single-pass against the four required sections.

The prompt:

[Paste the framework specification]

Run passion-exploration on this seed.

Seed: [Plain prose statement of your interest, your “I’ve been wondering,” your “what if.” Do not pre-narrow it.]

Optional context: [Anything about why this seed has surfaced for you, which directions you are already inclined toward, which directions you have already ruled out.]

The AI returns four labeled sections — exploration map, open questions, potential project nodes (with crystallization signal status named), next-directions (one deepening, one lateral). The output is exploratory in register; it does not converge on a thesis or recommend a project.

For best results:

  1. Resist the urge to specify a deliverable up front. If you start with “help me explore X so I can write a paper about it,” the framework will (correctly) read the deliverable as already-specified and route to T21 (Project Mode) rather than T20. Open exploration requires the seed to be genuinely open at invocation.

  2. Push back if the framework converges too fast. The failure mode is the framework picking a branch and developing only that branch. If you see this happen, ask explicitly for the other branches the seed admits — the lateral next-direction is a hint that more breadth is available.

  3. Trust the “no crystallization yet” signal. If the framework says no project has crystallized, it is not failing to be useful; it is honoring the wandering state. Crystallization is your call, not the framework’s.

  4. Use the de Bono moves explicitly when stuck. If your wandering enters a loop (the same three branches keep appearing), ask the framework to run the random-entry move (introduce an unrelated concept and force connections) or the concept-fan move (climb the abstraction ladder and find alternative branches at the higher level).

The framework is deliberately tool-agnostic. The frontier-roughness discipline, the crystallization detection protocol, and the breadth-over-depth move are conceptual disciplines that survive the lift to any environment. The output is plain prose with four labeled sections.

Other examples

  • An aesthetic seed without a project in mind. A user is interested in the way certain composers (Reich, Pärt) produce emotional weight from minimal material, and wonders whether the same principle applies in writing. The framework wanders four branches (structural minimalism, perceptual attention, practitioner discipline, cultural context), produces three open questions, flags no crystallization, and offers a deepening direction (analyze a specific Reich piece structurally) plus a lateral pull from architecture, where minimalism has a developed practitioner literature.

  • A vague intuition that resists project framing. A user keeps coming back to the feeling that something is wrong with how they spend their time, but the feeling refuses to specify a problem or a fix. The framework wanders four branches (time allocation, values mismatch, energy management, meaning deficit) and flags no crystallization. The deepening direction audits one recurring time-block against each branch; the lateral direction pulls from contemplative traditions where this feeling has a long literature. The framework does not turn the vagueness into a productivity-system project.

  • A research-adjacent seed that crystallizes mid-exploration. A user starts with “I’ve been wondering whether the way mathematicians describe insight matches the way meditators describe awakening.” Midway through the wandering the user writes “I want to actually run a small interview study” — the framework names the crystallization signal explicitly and offers transition to T21 with the candidate research project. The framework’s job in T20 is done.

Citations

The Open Exploration framework draws on three convergent traditions. The de Bono breadth-tools tradition (concept fan, random entry, lateral thinking) is the operational substrate — de Bono’s Serious Creativity (1992) developed the concept fan as a deliberate move upward in abstraction to surface alternative branches that direct enumeration misses, and the random-entry move as a disruption mechanism for breaking exploration loops. The Gentner structure-mapping tradition supplies the discipline against hollow analogy — cross-domain pulls are generative when the source domain has developed insights the target domain has not yet considered, and the structural-correspondence test (mechanism-level mapping, not surface lexical overlap) distinguishes signal from noise. Hofstadter and Sander’s Surfaces and Essences (2013) is the alternative tradition (fluid analogy as the fuel of thinking) cited for breadth.

The crystallization detection protocol is internal to Ora. Its discipline against forcing crystallization that has not happened (the productivity trap) is the deliberate counter to AI systems’ tendency to produce closure as evidence of having been useful — closure forced on an open seed is not usefulness; it is the lecture trap in disguise.

The framework is currently at v1.0 (compiled 2026-05-01 per Decision L, the no-suffix convention for generative output) with one resident mode and two deferred expansion candidates (idea-development and research-question-generation per CR-6).

Downloads

  • Framework specification (PDF) — link to ora-ai.org canonical artifact when published
  • Framework specification (plain text) — link to ora-ai.org canonical artifact when published
  • Full white paper (PDF) — link when published