It is a product of three conditions: grounding in what matters, reasoning from what reality forbids, and stepping into the adjacent possible, which widens with every step
A position paper: possibility as a function of grounding × first principles × the adjacent possible, tested against the science of modal cognition, prospection, mental contrasting, purpose, self-determination, emergence and measured innovation — and why an economy that credits people for goals they author makes this the central human capacity of the agentic age
ENSI research. Built on a library of 307 primary documents across 16 angles — possibility and modal cognition, prospection and imagination, contemplative science, purpose and meaning, motivation and self-authorship, the physics of information and complexity, the adjacent possible and economic complexity, and the evidence on AI and human agency. Stamped 11 November 2026.
Thinking from infinite possibility is a cognitive capacity, not a belief about outcomes. It has three components. Grounding is the anchoring of choice in a meta-level of what matters — values and a purpose held by the agent, which supply the criteria against which every option is evaluated. First-principles reasoning is the evaluation of options against the constraints reality actually imposes — physical, logical, legal, economic — rather than against inherited categories, roles and precedents. The adjacent possible is the set of states reachable in one step from the present state; it is not fixed, because each step taken adds new states one step away. The capacity is the joint operation of the three: a stable criterion, a correct model of constraint, and an iterated search whose frontier expands as it moves.
Stated as a working model: possibility = grounding × first principles × the adjacent possible, iterated step by step. The form is a product because the conditions are conjunctive. Grounding with no model of constraint yields a stable agent searching a space drawn by someone else. A model of constraint with no grounding yields capable search steered by whatever criterion is most salient — threat, status, or an engineered feed. Both without iteration yield a static option set, and the evidence below shows that a static, vividly simulated future lowers attainment. The multiplicative form is a structural hypothesis, not a measured law. Its nearest empirical tests are interaction effects — goal attainment raises well-being more when goals are self-concordant (β = .11 for the interaction), and mental contrasting makes effort depend on expectancy — and both point the same way.
The hinge of the argument is a single experimental result. In a study of 498 adults, judgements made under a 1,500-millisecond limit classed immoral events as impossible 37.30% of the time against 21.41% after deliberation (z = 5.42), and irrational events 42.10% against 37.86%; for physically impossible events reflection moved judgements the other way, from 79.37% to 85.22% (default possibility, p. 3). The fast modal default is narrower than the feasible set, and deliberation corrects it asymmetrically — more permissive about constraints set by norm and habit, stricter about constraints set by physics. That asymmetry is first-principles reasoning in miniature, and it is the operation this paper generalises.
The popular version of “infinite possibility” fails the tests the evidence sets, and the failures define the construct by exclusion. Belief in one’s own malleability is not the lever: across 63 studies and 97,672 students, growth-mindset interventions moved achievement by d = 0.05, falling to 0.01–0.03 after correction for publication bias (Macnamara and Burgoyne, p. 2, 24–25). Positive simulation of a desired future predicts less attainment, not more (mental contrasting review, p. 1). And where novelty is measured directly, it accumulates sublinearly — Heaps exponents of 0.45 to 0.78 across four large data sets of human activity (correlated novelties, p. 3, 13). “Infinite” is therefore not a claim that every outcome is reachable. It is a claim about the topology of the space: unbounded in its combinations, local in its steps.
The positive half of the thesis is better supported than the popular half. Prospection — the representation and evaluation of possible future states — has been proposed as an organising function of cognition rather than an occasional activity (prospection manifesto, p. 2), and the neuroimaging shows one core network serving remembering, simulating futures and counterfactuals (future of memory, p. 1, 5–7). A self-chosen purpose predicts sustained effort and, observationally, survival; autonomy over goals predicts autonomous motivation across 881 samples. None of these findings shows that possibility is boundless. Together they show that the option set a mind actually searches is a variable — narrowed by speed, threat, imposed goals and inherited categories, widened by reflection, contrast, self-concordant purpose and recombination.
The physical half of the argument comes from formal theory. One research programme proposes stating every law of physics as a claim about which transformations are possible and which are impossible (constructor theory, p. 1); on that view anything no law forbids is open to construction. Condensed-matter theory adds that knowing the base laws does not allow one to deduce the structures built on them, because each level of organisation develops properties of its own (More Is Different, p. 1). The two claims together give the shape this paper defends: the base is narrow and the space above it is not. Both are theory; the first is untested and its author states that its principles “may be false” (constructor theory, p. 39).
The agentic transition raises the stakes. Large language models could affect at least a tenth of the tasks of about 80% of the US workforce, and at least half the tasks of about 19% (GPTs are GPTs, p. 1) — an exposure estimate, not a forecast. As execution is delegated, the scarce human inputs move upstream: choosing the goal, judging the result, originating what does not yet exist. Who sets the goal becomes the central economic variable. Axiocracy — capitalism, amplified for the agentic age — keeps markets, property, prices, profit and competition, and adds an accounting of the value people and agents create against goals and measures agreed before the work starts by those who will be measured, together with a protected sphere of life that is never accounted. Under such rules the person becomes the author of their goals and measures. The capacity that authorship requires is the capacity described here.
The paper proceeds in four movements: what possibility is as a cognitive variable (1–2), the grounding that stabilises the search (3–5), the first principles that define its true boundaries (6–9), and the step-wise expansion and authorship through which it becomes real (10–12). Each principle gives a conceptual image, a six-line definition, the evidence and five ways to build with it. Every evidence line carries a grade: meta-analytic (pooled across studies), trial (one or a few randomised experiments), correlational (cohort, cross-sectional or survey data), or theoretical (formal models, conceptual analysis, theory without a direct test). Null and contrary findings are kept.
The main points
The default possibility set is narrower than the feasible set. Fast modal judgement excludes options that violate norms; reflection restores them and keeps physical limits.
Unconstrained simulation of a desired future closes the space; contrast opens it. Positive fantasy lowers energisation and attainment; desired state, obstacle and plan together raise it.
Grounding is a self-concordant purpose that extends beyond the self. Goals held for one’s own reasons sustain effort, and purpose beyond the self sustains it where self-interest fades.
Threat contracts the simulated future; prepared responses restore it. Distress lowers the specificity of simulated futures, and pre-committed if–then plans outperform in-the-moment regulation.
Identity-bound certainty terminates search. A familiar solution suppresses better ones, and the rate of belief updating predicts forecasting accuracy more than ability does.
Inherited categories fix the visible subset of the option space. Conceptual mappings run below awareness, and memory retrieves by surface category rather than by structure.
A first principle states what is forbidden; everything else is open to construction. True constraints are few and often far from customary ones.
Constraints underdetermine outcomes. Each level of organisation has properties of its own, so the space of constructions above the base laws is open.
Novel options are recombinations of mastered structure. Simulation recombines stored elements; structural analogy widens the space, and retrieving the distant analogue is the bottleneck.
The adjacent possible expands locally and sublinearly. Each novelty opens related novelties; leaps beyond the adjacent set rarely land, and combinatorial take-off is not foreshadowed.
Autonomy over goals and measures is the motivational condition of possibility. Imposed goals and controlling rewards reduce intrinsic motivation; goals agreed in advance by those measured restore authorship.
When agents execute, the human contribution is goal-setting, judgement and origination. Agents can serve as coach, mirror and red-team; delegating the goal itself removes the grounding factor.
1. The default possibility set is narrower than the feasible set
Metaphor: A lit circle on a dark plane — the plane is what reality allows, the circle is what fast judgement illuminates.
Definition:
Every mind computes a fast, default representation of what is possible before it deliberates.
That default is a subset of the feasible set, from which options that violate norms or habits are silently excluded.
In the default, the modal judgements “could” and “should” are nearly the same judgement.
Deliberation widens the subset where the exclusion was normative and narrows it where the limit is physical.
The capacity to simulate specific futures sets a further ceiling on how large the searched set can be.
Thinking from possibility begins as the deliberate correction of a measured bias, not as an act of faith.
Why it holds:
Under a 1,500-millisecond limit, 498 participants judged immoral events impossible 37.30% of the time against 21.41% after deliberation, and irrational events 42.10% against 37.86%, while ordinary and improbable events were unaffected and physical impossibility rose from 79.37% to 85.22% with reflection (default possibility, p. 3). Grade: trial — a randomised timing manipulation in one paper, online US samples and vignettes.
Different modal words (may, might, could, ought, should, possible) converged on one default set when answered fast (mean r = 0.893 against 0.830 on reflection, d = 0.754); for immoral events they correlated 0.558–0.919 when fast and −0.016 to 0.572 after reflection, and a mediation analysis indicated that people feel “forced” to act morally because the default already excludes the alternatives (p. 4–6). Grade: trial; the outcome is which options come to mind, not behaviour.
A short episodic-specificity induction — training in recalling concrete detail — increased the detail of simulated futures, raised hippocampal activity and improved both means–end problem solving and divergent thinking (hippocampus and imagination, p. 20). Grade: trial — single-laboratory experiments, no pooled effect sizes.
In 267 people with aphantasia against 203 controls, the vividness of simulated future events was sharply lower (r = 0.63) while spatial navigation was preserved (r = 0.08, n.s.), so the capacity to represent futures varies as a measurable trait (aphantasia cognitive profile, p. 6–7). Grade: correlational — self-report, self-selected online sample.
Clinical states narrow the set: suicidally depressed patients simulate less specific futures, using the same core network that serves remembering (prospective brain, p. 1–4), and generalised anxiety has been characterised as “the chronic expectation that something unspecified but awful will soon occur” (prospection manifesto, p. 17–18). Grade: correlational clinical comparison; the second claim is theoretical.
How to build with it:
Before any consequential decision, list the options rejected in the first seconds and the reason for each exclusion.
Classify every exclusion as either a physical, logical, legal or arithmetic constraint, or a norm, habit, role or preference.
Ask “could” and “should” as two separate questions, in that order.
Remove time pressure where the option set matters; a speeded choice inherits the default.
Train specificity: describe each candidate future in concrete detail — sequence, actors, resources — before judging it.
2. Unconstrained simulation of a desired future closes the space; contrast opens it
Metaphor: A map with the destination drawn in colour and the terrain between left blank.
Definition:
A vivid simulation of a desired future feels like an expansion of possibility.
On its own it functions as anticipatory consumption: the reward is experienced early and energisation falls.
Contrasting the desired state with the present obstacle converts the simulation into a test of feasibility.
When expectancy is high the contrast mobilises effort, and when it is low it licenses disengagement.
A possible self alters behaviour only when it carries a strategy, and it is steadier when paired with a feared counterpart.
Possibility, on this evidence, is a discipline of contrast rather than a mood of hope.
Why it holds:
Across “over 15 years of research”, “the more positive a person’s fantasies about the future, the less successful she is in realizing these fantasies”, in health, interpersonal and academic outcomes measured up to two years later (mental contrasting review, p. 1). Grade: correlational longitudinal studies, summarised in a narrative review by the method’s originator; no pooled effect sizes.
In four experiments, induced positive fantasies lowered energisation — measured subjectively and as systolic blood pressure — relative to questioning, negative or factual thoughts; mental contrasting combined with if–then plans improved grades, exercise over four months and diet over two years (p. 1–4). Grade: trial — several randomised studies, not pooled.
Among 160 low-income eighth graders, 82% of expected selves concerned school, yet only selves carrying concrete strategies predicted more homework (β = 0.20), better participation and higher grades, explaining 8% of multivariate variance across five outcomes — and they changed behaviour without changing feeling about school (ΔR² = .00) (possible selves as roadmaps, p. 10–11). Grade: correlational, one school year, small effects.
In 238 adolescents, more than 81% of the least delinquent held at least one balanced pair of expected and feared selves against 37% of the most delinquent, and global self-esteem did not track delinquency (possible selves and delinquency, p. 10–11). Grade: correlational, cross-sectional, one city.
Belief does not substitute for enactment: belief-change interventions moved achievement by d = 0.05 across 63 studies (Macnamara and Burgoyne, p. 2), while completing two concrete trait-relevant behaviours a week shifted traits by about 0.17 SD over a semester (Hudson following through, p. 7). Grade: meta-analytic (preregistered) against trial.
How to build with it:
State the desired outcome in one sentence, then the principal obstacle located in oneself — a habit, belief or emotional response.
Convert that obstacle into an implementation intention: “if situation X, then response Y”.
Attach a concrete strategy to every hoped-for self; a possible self without one has no measured effect on behaviour.
Name the feared self in the same domain and hold it as a boundary condition, not as a forecast.
Where the contrast shows expectancy is low, disengage deliberately; releasing an infeasible goal frees resources for the adjacent possible.
3. Grounding is a self-concordant purpose that extends beyond the self
Metaphor: A fixed point in a rotating frame — everything else can move because one thing does not.
Definition:
Grounding is the meta-level of choice: the values and purpose against which options are evaluated.
It is a criterion, not a state of calm and not the absence of aims.
Its strongest form is a purpose that is meaningful to the self and of consequence beyond the self.
A goal is grounded when it is self-concordant — held from interest or core value rather than from pressure.
With a grounded criterion, any method, role or template can be abandoned without loss of direction.
Without one, the most salient external agenda supplies the criterion by default.
Why it holds:
In three longitudinal studies (n = 169, 152 and 73), self-concordant goals attracted more sustained effort, effort carried the effect on attainment (β = .76), attainment raised well-being more for concordant goals (interaction β = .11), and attaining non-concordant goals did not raise well-being (self-concordance model, p. 3–6). Grade: correlational — small student samples over one semester.
A single 20–30-minute exercise linking schoolwork to contributing beyond the self raised STEM grades (d = .11; d = .21 for lower performers) in 338 students, and in 429 participants it left them solving 36% more problems than controls once a tedious task grew long (d = .32); self-oriented motives did not consistently produce the effect (Boring but Important, p. 1, 17–18, 25–26). Grade: trial — small effects, short follow-up; the intervention family has replicated unevenly.
Among 6,985 adults over 50, the lowest purpose band died at 2.43 times the rate of the highest over four years after adjustment for depression, anxiety and optimism (purpose and mortality, p. 1, 7); among 6,163 adults followed for 14 years, each SD of purpose went with 15% lower mortality regardless of age (purpose across adulthood, p. 4–5). Grade: correlational cohort; the second measure had internal consistency of only α = .36.
The construct is defined as “a stable and generalized intention to accomplish something that is at once meaningful to the self and of consequence to the world beyond the self” (defining youth purpose, p. 3); about one in five high-school and one in three college students report one (Psychology of Purpose, p. 13), and having meaning and searching for it are nearly independent (r = −.19) (Meaning in Life Questionnaire, p. 4). Grade: theoretical definition with descriptive survey data.
Grounding is not a quick exercise: a celebrated 15-minute values-affirmation effect was not significant in 77% of 1,566 analytic specifications, and the effect on its target group was null in 97% (self-affirmation reanalysis, p. 4, 7). Grade: reanalysis of a single trial — a null result.
How to build with it:
Write the purpose in two clauses: what matters to you, and whom or what it serves beyond you.
Test each current goal for concordance: would it still be pursued if no one observed or rewarded it?
Name the small set of values that will not be traded for any outcome, and treat them as fixed constraints in every plan.
When a method or role fails, return to the purpose before selecting the next method.
Build purpose by the routes associated with it — long-term mentoring, structured conversation about what matters, service (Psychology of Purpose, p. 17–19) — and treat one-off writing exercises as insufficient.
4. Threat contracts the simulated future; prepared responses restore it
Metaphor: An aperture that closes as the light grows harsher, until only the nearest object is in focus.
Definition:
Threat acts on prospection: it fills the simulated future with aversive outcomes and lowers its specificity.
Engineered emotion acts on attention: it determines which options are represented at all.
A steady agent is one whose option set is not being contracted by fear or by an external designer of salience.
Steadiness is not the absence of affect but a repertoire of responses selected in advance.
In-the-moment regulation is weak, while pre-commitment and control of inputs are strong.
Restoring the aperture is a precondition for searching the whole space.
Why it holds:
Across the instructed-regulation literature, redirecting attention had no effect on emotional outcomes (d = 0.00), suppression a small one (0.16) and reappraisal a small-to-medium one (0.36), whereas if–then plans outperformed no instruction by d = 0.91 and goal intentions alone by 0.53 (implementation intentions meta, p. 6, 24, 31–35). Grade: meta-analytic.
An estimated 31% of social-media use is driven by self-control problems (digital addiction experiment, p. 2, 6), yet ten trials of social-media abstinence found pooled well-being effects of essentially zero (Lemahieu abstinence meta, p. 7): engineered salience costs agency more than mood, and the lever is authorship of inputs rather than abstinence. Grade: trial (about 2,000 participants) and meta-analytic.
Across 136 randomised trials with 11,605 adults, mindfulness programmes reduced anxiety against no intervention (SMD −0.56), but showed no superiority over specific active alternatives such as exercise, every prediction interval crossed zero, and 89% of trials did not report measuring adverse events (Galante mindfulness meta, p. 1–2, 15). Grade: meta-analytic, preregistered; all trials at high risk of bias.
In a randomised trial of 5,049 adults, short 5–10-minute digital practice was no more likely than ordinary internet use to produce a distressing experience (OR 1.05), but 10 minutes carried more risk than 5 (OR 1.42) (digital meditation RCT, p. 1, 8); among practitioners recruited for difficult experiences, 73% reported moderate-to-severe impairment and 72% of cases arose during or after retreats (Lindahl meditation challenges, p. 15, 21). Grade: trial, and a qualitative study that does not yield prevalence rates.
Distress lowers the specificity of simulated futures in depression and schizophrenia in proportion to symptoms, because remembering and simulating share one constructive system (prospective brain, p. 1–2). Grade: correlational clinical comparison, largely from the authors’ own studies.
How to build with it:
Write if–then plans for the three situations that most reliably contract your option set.
Set rules, not moods, for what enters attention: feeds, notifications, news and the order of the first hour of the day.
When fear forecasts an outcome, record the forecast as a probability and score it later.
Use short practices that return attention to the present; treat intensive retreat practice as a serious intervention with documented risks.
Defer consequential choices made under externally manufactured urgency.
5. Identity-bound certainty terminates search
Metaphor: A search that halts at the first acceptable answer and records it as the best one.
Definition:
The form of ego that most narrows possibility is not grandiosity but the need to be right.
It binds identity to a known solution and defends that solution against better ones.
Expertise strengthens the effect, because a familiar answer arrives faster and ends the search sooner.
Loosening it means holding beliefs as revisable at a known rate, in response to evidence.
On the evidence this is a measurable habit of updating more than a mystical state.
An agent that revises keeps more of the option space live than an agent that defends.
Why it holds:
When a familiar five-move solution was available beside a shorter three-move one, expert chess players performed at roughly the level of players three standard deviations weaker; experts predicted success rates of 86%, 74% and 59% for three groups, and the observed rates were 50%, 18% and 0% (expert Einstellung, p. 1, 11–12). Grade: trial — laboratory experiment with small expert samples.
Across more than 150,000 forecasts by 743 forecasters, the strongest behavioural predictor of accuracy was the frequency of belief updating (−0.45 with error), ahead of deliberation time and measured ability (drivers of forecast accuracy, p. 7–9). Grade: correlational, within a large forecasting tournament.
Of 2,994 new business owners, 33% rated their own chance of success at 10 out of 10, against base rates showing about half of such firms survive five years (Cooper on founder odds, p. 1, 7). Grade: correlational survey.
Self-transcendent states — decreased self-salience, increased connectedness — are measurable: in a double-blind trial with an active control, about two thirds of psilocybin recipients ranked the experience among the five most meaningful of their lives (self-transcendent experience, p. 8), and ego dissolution and ego inflation load on separate, nearly orthogonal factors (ρ = −0.110) (Ego-Dissolution Inventory, p. 1, 6). Grade: trial with small, selected samples; correlational scale validation.
Contemplative practice is no guaranteed exit from self-interest: across 22 randomised trials, meditation’s effect on compassion was not significant against active controls and appeared only when the teacher co-authored the study (r = 0.42), with no effect on aggression or prejudice (Kreplin prosocial meta, p. 5). Grade: meta-analytic; 61% of included studies rated weak.
How to build with it:
When a solution arrives quickly, apply the prohibition test: if this approach were unavailable, what would be done instead?
Keep a dated log of forecasts and decisions, and count how often each belief is revised.
Separate identity from method; an identity fused with a method cannot drop the method.
Submit the hardest problem to someone whose field lacks your standard solution.
Treat awe and self-transcendent experience as a low-cost loosening of self-salience, not as a guaranteed release from ego.
6. Inherited categories fix the visible subset of the option space
Metaphor: A grid laid over the terrain by a surveyor who had other purposes.
Definition:
Most reasoning runs on categories, roles and conceptual mappings constructed earlier by others.
They are useful compressions, and they determine which options are represented at all.
A role description, a sector’s standard practice and a governing metaphor each act as such a filter.
Their stability is easily mistaken for the stability of the underlying constraints.
First-principles reasoning begins by treating a category as a choice that can be inspected.
Only then can the load-bearing constraints be separated from the conventional ones.
Why it holds:
Conceptual-metaphor theory holds that everyday thought runs on “a huge system of thousands of cross-domain mappings”, that abstract concepts such as time, causation and purpose are themselves structured by them, and that the system is “mostly unconscious, automatic, and is used with no noticeable effort” (contemporary theory of metaphor, p. 1, 39–40). Grade: theoretical — linguistic analysis; the psycholinguistic tests are cited, not presented.
On the same theory a governing metaphor highlights some options and hides others, and “novel metaphors can have the power of defining reality” (metaphors we live by, p. 7–8, 30–33). Grade: theoretical — philosophical argument without experimental data.
Memory retrieval is driven by surface similarity whereas sound inference depends on structural similarity: “the kind of similarity that most reliably led to remindings was not the kind participants most valued in making inferences” (analogy and similarity, p. 9). Grade: trial — laboratory experiments from one research programme.
Comparison operates inside shared structure: given five minutes, people listed more than twice as many alignable differences for similar pairs (M = 9.0) as for dissimilar ones (M = 3.9), so what a category groups together is what gets examined (p. 6). Grade: trial.
In retrieval experiments 88% of participants recalled a same-domain analogue and only 12% a remote-domain one (analogy and relational reasoning, p. 11); at population scale, in only seven of 46 economies did two in five or more new entrepreneurs offer products or services new to their market (GEM global report, p. 21). Grade: trial (laboratory) and correlational (descriptive survey).
How to build with it:
List the roles you occupy and, for each, three rules that have never been examined.
Describe the work without its title: which problem, for whom, by what means.
Identify the governing metaphor of your field and substitute another to see which options appear.
For each category relied upon, ask who drew it, for what purpose, and under which constraints that may no longer hold.
Retain the categories that survive inspection; drop those whose only support is precedent.
7. A first principle states what is forbidden; everything else is open to construction
Metaphor: A boundary drawn by law, with no path drawn inside it.
Definition:
A first principle is a statement about how reality works, not a description of how things are done.
At its most general, physics can be expressed as which transformations are possible and which are impossible.
Whatever no such law forbids is, in principle, open to construction.
Conventions, roles and precedents are not laws; they are earlier constructions.
Reasoning from first principles means testing each “cannot” against the constraint that actually binds.
The list of binding constraints is short, which is why the space inside it is large.
Why it holds:
Constructor theory proposes expressing every law of physics as a statement about which tasks are possible and which impossible, noting that the standard approach can rule a task out but “cannot rule them in until after a constructor for them has been designed” (constructor theory, p. 1, 4, 27). Grade: theoretical — a research programme with no new predictions yet; its author states its principles “may be false” (p. 39).
A small set of symmetries fixes the great conservation laws: invariance under translation in space gives conservation of momentum, in time conservation of energy, and under rotation conservation of angular momentum (symmetry and conservation, p. 1, 4–5). Grade: theoretical — an established mathematical theorem with full empirical support.
True limits often sit far from intuition: the only unavoidable thermodynamic cost of computation is erasure, about kT ln 2 per bit, measured at 0.72 kT against a bound of 0.693 kT (Landauer bound measured, p. 2–3, 14), while measurement and copying carry no intrinsic minimum cost (Bennett on Landauer, p. 6). Grade: trial (one colloidal system) and theoretical; a minority view disputes the derivation (Norton critique, p. 13).
Living systems operate close to the physical floor rather than a conventional one: a dividing bacterial cell releases less than three times the minimum heat physics requires for its growth rate and durability (self-replication heat, p. 3–4). Grade: theoretical — an order-of-magnitude bound computed from measured quantities.
Possibility proofs are a method in their own right: accurate self-reproduction and natural selection are possible under laws containing no design for life, provided only that digital information can be physically instantiated (constructor theory of life, p. 1, 25). Grade: theoretical — a proof inside an abstract framework, silent on probability.
How to build with it:
For every “we cannot”, name the law that forbids it: physical, logical, legal, economic — or merely customary.
Estimate the first-principles floor of cost, time or energy before accepting the prevailing rate as the limit.
Measure the gap between current practice and that floor; the widest gaps mark where possibility is concentrated.
Treat precedent as evidence that one solution exists, never as evidence that it is the only one.
Keep the binding constraints on a single page and test every new plan against it.
8. Constraints underdetermine outcomes
Metaphor: A finite alphabet and a library that can never be completed.
Definition:
First principles constrain what can exist without determining what will exist.
Each level of organisation develops properties and regularities of its own.
Chemistry is not applied physics, and a market is not applied psychology.
Knowledge of the base therefore does not allow the structure above it to be deduced; it has to be constructed.
This is how a narrow set of constraints opens an unbounded space of constructions.
The infinite part of infinite possibility is the space above the principles, not the principles themselves.
Why it holds:
Reduction is accepted and construction denied: “the ability to reduce everything to simple fundamental laws does not imply the ability to start from those laws and reconstruct the universe”, because at each level of complexity “entirely new properties appear”, with broken symmetry as the mechanism (More Is Different, p. 1–2). Grade: theoretical — argument from established physical examples, not a proof.
Far from equilibrium, which branch a system takes at a bifurcation depends on fluctuations, introducing a path-dependent element into physics and chemistry: the laws constrain the outcome without fixing it (dissipative structures, p. 11). Grade: theoretical, supported by laboratory instabilities such as Bénard convection.
A coarse-grained macro model can carry more causal information than the full micro model — 3 bits against 2.43 in one system (causal emergence, p. 7) — and 240 of 256 elementary cellular automata, including a Turing-universal rule, become predictable once coarse-grained (irreducible yet predictable, p. 3–4). Grade: theoretical — formal results on toy systems.
Evolution has been argued to be “a non-deducible construction, not an entailed deduction”, because new functions for existing parts open spaces that cannot be listed in advance (life as phase transition, p. 9–10). Grade: theoretical — a speculative preprint.
The question is open: the free-energy framework argues that its formalism “undercuts the notion of ontological emergence” (free energy principle, p. 123), and none of these results measures emergence in a firm, an institution or a life. Grade: theoretical dissent.
How to build with it:
Learn the base constraints of your field, then stop expecting them to supply the answer.
Reason at the level where the behaviour of interest is stable — cohorts, markets, institutions — rather than by summing parts.
Expect constructed systems to show properties no one designed, and look for them deliberately.
Record every case in which an existing component could serve a new function; each is a new region of the space.
Combine constraints from different levels — physical limits with human preferences, information costs with institutional rules; the fourth report in this series tests how often breakthrough businesses were built that way.
9. Novel options are recombinations of mastered structure
Metaphor: A combination lock whose dials are the structures already mastered — few dials, an unbounded number of settings.
Definition:
Novel options are constructed by recombining elements a mind already holds.
The richer and more varied the stock of mastered structure, the larger the space of combinations.
Insight comes from mapping relational structure, not surface resemblance.
The bottleneck is retrieval: the structurally relevant but distant analogue rarely comes to mind unprompted.
Possibility therefore widens with deliberate breadth and with the decomposition of concepts into parts.
Novelty stands on firmly mastered structures, never on a blank slate.
Why it holds:
The constructive episodic simulation hypothesis holds that the memory system builds future events “by extracting and recombining stored information into a simulation of a novel event” (future of memory, p. 5–7), and an experimental paradigm that shuffles person, object and place details across real memories still produces robust hippocampal activation (hippocampus and imagination, p. 9–10). Grade: correlational neuroimaging, reviewed by the hypothesis’s proponents; the cost is a measured rise in false recognition.
About 10% of students solved a classic convergence problem unaided, about 20% after reading a structurally analogous source problem, and about 75% when simply told to apply it (analogy and relational reasoning, p. 11): the analogue was available, and retrieving it was the constraint. Grade: trial — classic laboratory studies with student samples.
Analogy maps relations rather than attributes, and which relations transfer is set by systematicity — those bound together by higher-order relations such as cause (structure-mapping theory, p. 1, 4–5). Grade: theoretical, with small supporting studies reported without figures.
“The creative activity of the imagination depends directly on the richness and variety of a person’s previous experience” (Vygotsky on imagination, p. 8); consistently, founders of the fastest-growing new US firms average 45 years of age (founder age study, p. 2). Grade: theoretical, and correlational administrative data.
Blending theory concludes that new meaning arises as emergent structure in the combination of input spaces, but that novelty depends on “a background of firmly anchored and mastered mental structures” (blending, form and meaning, p. 2, 29). Grade: theoretical — built from analysed examples, not experiments.
How to build with it:
Widen the stock of structure on purpose: study, work and collaborate across fields that the current role does not require.
Master a few domains deeply; recombination needs anchored structures, not fragments.
Decompose the current concept into its component assumptions, remove one, and test what the remainder can build.
For every hard problem, ask which distant field shares its causal structure, and study that field directly.
Let an agent propose remote analogues at volume; retain the judgement of which mapping holds.
10. The adjacent possible expands locally and sublinearly
Metaphor: A frontier that advances with the explorer, so that the unexplored region grows as it is entered.
Definition:
The adjacent possible is the set of states one step from what already exists.
Each realised novelty adds new adjacent states, so the space grows as it is traversed.
Moves beyond the adjacent set rarely succeed; moves into it do.
Reinforcing the familiar slows the rate of novelty, and opening the genuinely new raises it.
Combinatorial growth can remain flat for long periods and then accelerate without warning.
Infinite possibility is reached by iteration, not by a single leap.
Why it holds:
In Wikipedia edits, social tags, texts and music listening, novelties trigger related novelties; in the model that fits all four, reinforcement of the familiar (ρ) and expansion of the new (ν) set the rate of innovation, with distinct items growing as N^(ν/ρ) while reinforcement dominates (correlated novelties, p. 4–6). Grade: correlational — large observational data sets with a fitted formal model; the space of possibilities itself is never observed.
The same data cut against an unqualified “infinite”: measured Heaps exponents of 0.45 to 0.78 mean novelty slows as reinforcement accumulates, and innovations “too far ahead of their time, i.e. not adjacent”, are naturally limited (p. 3, 9, 13). Grade: correlational.
Countries’ new exports between 1992 and 2000 were predicted by the structure of what they already exported — they moved into nearby products — and countries more complex than their income predicted grew faster (economic complexity, p. 12–13). Grade: correlational panel data; capabilities are inferred, not observed.
In the formal theory of the adjacent possible, two objects become 3, 7, 127 and then about 10³⁸ in four steps, and the transition “is not foreshadowed by any features of the curves until its onset” (TAP equation, p. 2, 9). Grade: theoretical — a toy model fitted to no data.
Iteration with explicit tests pays: across four trials and 759 startups, training founders to state theories and kill thresholds raised the probability of abandoning a weak idea by 9.8 percentage points and of exactly one radical pivot by 8.3 (scientific method replication, p. 13, 17–19). Grade: trial — replicated randomised field experiments.
How to build with it:
Each month, list what has become possible, cheap or permitted in your field during the past year.
Choose the next project to reuse most of what exists and add one genuinely new element.
Raise your own ν deliberately: one new collaborator, market or skill each quarter.
Before each step, write down the test and the threshold at which the step will be abandoned.
Read a long plateau as accumulation, not as evidence of a dead end.
11. Autonomy over goals and measures is the motivational condition of possibility
Metaphor: A measure agreed before the work, so that it functions as a contract rather than a verdict.
Definition:
A designed role supplies goals and measures written by someone else.
Imposed goals, pressured evaluation and controlling rewards reduce the intrinsic motivation on which open search depends.
When goals and measures are agreed before the work starts, by those who will be measured, the measured become their authors.
That is the core of Axiocracy, capitalism amplified for the agentic age: visible contributions, ex-ante agreement and credit in proportion to measured value, open to inspection and appeal.
Its complement is a protected sphere of life that is never accounted.
Authoring one’s own goals and measures is first-principles reasoning applied to a life.
Why it holds:
A meta-analysis found that “all expected tangible rewards made contingent on task performance do reliably undermine intrinsic motivation”, and threats, deadlines, pressured evaluation and imposed goals also reduce it while choice and self-direction increase it (Ryan and Deci 2000, p. 3). Grade: meta-analytic finding reported in a theoretical overview; the reward effect has been contested.
Across 881 samples and 443,556 people, autonomy support correlated with autonomous motivation at ρ = .44 and with controlled motivation at about zero (ρ = .01), but with objective performance at only about .20 (need supports meta, p. 13). Grade: meta-analytic, almost entirely correlational; authorship changes engagement more than immediate output.
In 119 samples of workers, autonomy satisfaction correlated with job satisfaction at ρ = .69, with burnout at −.60 and with turnover at −.38, while competence barely predicted turnover (−.07) (SDT needs at work, p. 17–18). Grade: meta-analytic, mostly cross-sectional self-report.
Goal content matters as well as goal ownership: weighting extrinsic aspirations such as wealth and image was negatively related to well-being, and attaining intrinsic aspirations was linked to well-being while attaining extrinsic ones was not (Ryan and Deci 2000, p. 8) — the empirical reason a sphere of intrinsically motivated activity should stay outside any ledger of credit. Grade: correlational.
Fit between what a role supplies and what a person needs predicts job satisfaction more strongly than fit between what it demands and what the person can do (.61 against .41) (person–environment fit, p. 29); no trial has yet tested goals and measures agreed ex ante by those measured as an intervention in its own right. Grade: meta-analytic self-report, and an acknowledged evidence gap.
How to build with it:
In the current role, draft your own goals and measures for the next quarter before it starts, and ask for them to be agreed.
Do the same for a life: three to five goals with their measures, written before the year begins and shown to two people whose judgement you trust.
Keep a personal record of contributions against agreed goals; it is portable evidence of value.
Write down what will never be scored — by you, an employer or an application — and defend that boundary.
Where an imposed measure cannot be renegotiated, decide explicitly whether to occupy that role or to leave it.
12. When agents execute, the human contribution is goal-setting, judgement and origination
Metaphor: When execution becomes cheap, the scarce input is the choice of destination.
Definition:
As agents take on more of execution, the human share of value moves upstream.
What remains scarce is choosing goals, judging what is good and originating what did not exist.
These are the capacities that grounding and first-principles reasoning train.
An economy that credits value against goals people author rewards the scarce part.
Agents can serve as coach, mirror, map-builder and red-team.
They must not author the goal, because a goal written by a model removes the grounding factor from the product.
Why it holds:
Language models could affect at least half the tasks of about 19% of US workers (GPTs are GPTs, p. 1), yet across more than four million real conversations only about 4% of occupations used AI for three-quarters or more of their tasks, and 57% of use was augmentative rather than automative (Anthropic Economic Index, p. 1–2, 7). Grade: correlational — exposure estimates and usage data, not displacement.
Among 319 knowledge workers reporting 936 tasks, confidence in the AI predicted less critical thinking (β = −0.69), while confidence in one’s own ability to evaluate its output predicted more (β = 0.31) (GenAI and critical thinking, p. 9). Grade: correlational self-report survey.
In a small four-month study, 83.3% of participants who wrote with an LLM could not quote a sentence of the essay they had just written, against 11.1% in the other groups, and fewer claimed full ownership of it (9 of 18 against 16 of 18) (cognitive debt pilot, p. 31–32). Grade: trial — n = 54, unreviewed preprint, hypothesis-generating.
Persuasion is symmetric: instructed to argue for a conspiracy, a model raised belief by 13.6 points about as effectively as debunking lowered it by 12.1, and a one-line instruction to use only accurate arguments cut that effect by about two-thirds (AI can promote conspiracies, p. 3, 5, 8). Grade: trial — four experiments, N = 3,996, preprint.
As a coach on goals people set themselves, a narrow rule-based chatbot matched human coaches on goal attainment over ten months (ηp² = .269 against .265), with a dose effect (37.6 against 17.6 points) (AI vs human coaching, p. 1, 9–10). Grade: trial — two randomised studies with students, self-scored outcomes.
How to build with it:
Write goals, charter and first draft before any agent sees them.
Instruct agents to red-team: “state the strongest case that this goal is not mine, or that this constraint is not real”.
Ask agents for first-principles cost floors and remote analogues, then verify both.
Put an instruction to use only accurate arguments into every standing prompt.
Spend the time agents release on origination: the step no one has requested.
Building from infinite possibility
The twelve principles reduce to one product — grounding × first principles × the adjacent possible — and to a sequence of practices that raises each factor. None of them waits for a new economic order.
This week — widen the default set. Take one pending decision and list every option, including those excluded in the first seconds. Sort each exclusion into constraints imposed by physics, logic, law or arithmetic and constraints imposed by norm, habit, role or preference. On the modal-cognition evidence, expect the second list to be the longer one.
This month — fix the criterion. Write the purpose in two clauses — what matters to you and what it serves beyond you — and name the values that will not be traded. Drop or renegotiate every goal that fails the concordance test. Write three if–then plans for the situations that most reliably contract the option set, and set explicit rules for what enters attention.
This quarter — locate one binding constraint. Select the most expensive constraint in your work or life and estimate its first-principles floor: its cost if only physics, law and arithmetic applied. The gap between that floor and the prevailing rate is the first opportunity. Then take one step into the adjacent possible: a project that reuses most of what exists and adds one new element, with a test and an abandonment threshold written in advance.
Inside today’s capitalism — author the measures. Markets already reward agents who set their own goals: founders, independent professionals, the employee who proposes the project rather than waiting to be assigned one. Begin there. Before each quarter, draft your own goals and measures and ask for them to be agreed. Apply three axiocratic tests to any team: can contributions be seen; were the goals and measures agreed before the work started, with a say for those measured; can the computation of rewards be inspected and appealed? Each test a team passes moves it towards an economy in which people are credited for the goals they author. Then write down the protected sphere — relationships, play, contemplation, inner development — that no ledger will ever record.
What to expect. Small, compounding effects: about a fifth of a standard deviation in a trait over a semester, d ≈ .1–.3 from purpose interventions, one well-chosen pivot in place of many random ones, and novelty that accumulates sublinearly. No honest reading of the evidence promises transformation. The claim the evidence does support is structural: an agent that holds a self-concordant criterion, tests each “cannot” against the constraint that actually binds, and keeps stepping into the adjacent set searches a larger region of the feasible space than an agent that does none of these — and that region grows with each step.
The base is narrow; the space it opens has no end. Ground in what matters, reason from what reality forbids, take the step that opens the next one.
The series
This position paper opens a series of four standalone reports, each going deeper into one factor of the argument.
The Navigator: Grounding as the Real Intelligence — how a person becomes steady, present and free of engineered emotion, and keeps their maps of the world true.
All Signal: The Mental Models of the Entrepreneur — the models that make a builder’s every move carry information, tested against the evidence, with a sequence for training them.
Your Own Measure: Success and Maximum Power After Work — how to author success in advance and measure growing capability when paid work no longer organises life.
The First Principles of Reality, and What They Can Build — an atlas of the truths about reality that breakthrough businesses exploited, organised so that new ones can be designed from it.
Source note: this position paper draws on a library of 307 primary documents assembled by ENSI across 16 research angles — randomised trials, meta-analyses, cohort studies, physics papers, formal models and theoretical texts — and on the companion reports in this series. Every figure is taken from the original publication linked inline; page references are to the document as published online. Each evidence line is graded as meta-analytic, trial, correlational or theoretical, and preprints, toy models, narrative reviews, small samples and theory without a direct test are flagged as such. Where the evidence contradicted the thesis — on belief in growth, on positive fantasy, on meditation, on how fast novelty really grows — the paper follows the evidence.