Scenario Dimension Cube — Working Notes¶

Archived 2026-06-01 — moved from docs/research/ to docs/archive/ in the docs consolidation (ADR-083). Preserved iteration history, not current architecture. Current doc: Scenario Dimension Cube (canonical).

Status: Promoted 2026-05-30. Canonical doc now lives at ../research/foundations/scenario-dimension-cube.md. This working-notes file is preserved as iteration history — rejected hypotheses, rushed conclusions, re-attack findings, swing logs. Useful for understanding why the cube ended where it did.

Original status (pre-promotion): Active iteration. NOT a committed deliverable. Final form will live at scenario-dimension-cube.md once stable.

Opened: 2026-05-30 Origin: Session 2026-05-30 — starter pack design surfaced that scenarios drafted in a vacuum cluster in a narrow region of scenario-space. The carousel scenario came from a comedian's bit, not designer intuition — evidence that the design process was sampling from a too-small primitive set.

Connects to: - docs/research/experiments/experiment-new-domain-signals.md — proved cross-domain signals work; identified 12 governance signals + 2 new (pragmatic routing, self-aware aspiration gap) - docs/decisions/024-twelve-behavioral-signals.md — signals are OUTPUT; dimensions are INPUT axes a scenario varies - ADR-073 — starter pack design; cube informs v2 - 01_Projects/Pulse/decisions/Hmm2026-05-27-coachj-experiment-scenarios.md — 8 reference scenarios used as test cases

Purpose¶

The cube is a design tool (not an audit tool). Each axis must be independently manipulable so the generator can take a vector of values and produce a scenario.

The cube's killer use case: cross-domain consistency experiment. Hold a cube vector constant; vary only the surface domain (career / health / relational / financial / governance). If the same player gives structurally similar responses, the structure carries the signal. If not, we've found which combinations don't translate cleanly. This is the experiment-new-domain-signals finding made operational.

Two-layer model¶

The cube factors into two layers:

Structural layer — domain-agnostic axes. These are what the cross-domain experiment holds constant.

Content layer — varies by what the scenario is about. The existing governance generator prompt is largely content rules (cast composition, scan strip, voice quotes, prose constraints).

The cube sits above content; content rules realize a cube vector into a specific scenario surface.

Active structural dimensions (working list — will iterate)¶

#	Dimension	Values	Notes
1	Prior commitment	none / soft / strong	Generalizes "promise"; every scenario has a value. Confirmed clean after rigorous re-attack 2026-05-30. Flavors of strong (deathbed/oath/contract/verbal) and soft (implicit/semi-formal/hedged) are captured by C1 (formalization → adds financial/legal cost categories; ritualization → adds integrity-with-identity-weight) + content-layer texture + measured outcomes (identity load, subjective recency weighting). None = no commitment to a counterparty exists. Soft = non-binding expectation (pattern, signup, hedged response). Strong = explicit, mutually acknowledged commitment.
2	Decision deadline	seconds / minutes / hours / days / weeks / soft	How much time to deliberate. Refined 2026-05-30 — added minutes during rigorous re-attack. Seconds = pure reflex mode (<5s, no time to verbalize). Minutes = snap deliberation (5s–5min, time for one mental walk-through, brief verbal exchange). The carousel (60s) and S3 (90s) are minutes-mode, not seconds-mode. Hours: within-day deliberation, can consult someone. Days: sleep-on-it across nights. Weeks: full research and multi-party consultation, 1–8+ weeks (months deadline not added — reference set doesn't push that range, and risks confusion with months-horizon on axis 3). Soft: no specific deadline; flavors of pressure (none / social / accumulating-cost) captured by axis 12.
3	Consequence horizon	immediate / months / years / lifelong	How long effects persist from decision moment. Calibrated to reference player (35–45 working-age adult). Rushed pass added "generational" but rigorous re-attack 2026-05-30 reverted: generational is a combination (lifelong horizon + future-generations scope), not a primitive horizon value. Decades (15–25 yr) absorbed into upper edge of "years." Episodic/delayed/discontinuous patterns are content-layer texture. Tagging rule: for scenarios with active-phase + residue (e.g., 3-year relationship that ends), horizon = duration of dominant consequence; residue captured by reversibility (partial-permanent).
4	Reversibility of consequence	full / partial-recoverable / partial-permanent / none	Independent of horizon (S8 has 15yr horizon, full reversibility). Refined across two passes. Full: undo restores the original state at no additional cost (tight definition — most decisions don't qualify; S8 is one of few clean cases). Partial-recoverable: consequence diminishes to zero with time/work (S1 career: can return to industry; S5 Big Tech: can leap later). Partial-permanent: consequence persists but can be mitigated; never fully erases (S3 secret, S7 intro, carousel). None: zero mitigation path — rare; mostly deadline/window-close cases (missed grant), not action-irreversibility. Tagging rule: for scenarios with multi-component reversibility (e.g., S4 Priya: investing full, relationship partial-permanent), tag by dominant component. (Open question parked: active vs. passive recovery split within partial-recoverable.)
5	Scope of consequence	self / dyad / small group / institution / future-generations	Who is materially affected. Value "future-generations" added 2026-05-30 during axis 3 rigorous re-attack — covers people who don't exist yet (descendants, future humanity, future ecosystems). Distinct from institutional-abstract (which is a collective of currently existing people/entities).
6	Relational distance to most-affected	self / intimate / close / acquaintance / stranger / institutional-abstract	Distinct from scope. The "institutional-abstract" value (added 2026-05-30) covers cases where the most-affected party is a collective or abstraction with no specific identifiable individuals (the zoning office, the market, a regulatory body, "the company's future"). Was previously being conflated with "stranger" but plays differently — institutional patients are weighed lighter by most players.
7	Authority you hold	full / oversight / collective-vote / influence-only / none	Formal authority over outcome. Enumeration refined 2026-05-30 across three passes (rushed split partial → oversight/influence-only; rigorous re-attack confirmed; corpus decomposition surfaced collective-vote). Full: you decide, no review. Oversight: you decide, subject to review/approval (CEO recommending to board). Collective-vote: you have formal vote, share decision with peers in real time (committee member, council member, jury). Distinct play shape — coalition reasoning, single-vote leverage. Influence-only: someone else decides; you can advocate. None: someone else decides; no leverage. Collective-vote surfaced when decomposing The Marshal and The Kingdom After Silver — neither oversight (no individual decider with review) nor influence-only (player has formal vote) fit.
8	Information topology	symmetric-complete / mutual-uncertainty-resolvable / mutual-uncertainty-intrinsic / you-know-more / they-know-more	Who has what knowledge. Refined across two passes (initial rushed pass split symmetric → complete/mutual-uncertainty; rigorous re-attack split mutual-uncertainty further). Symmetric-complete: both have all relevant facts (S1: all offers explicit). Mutual-uncertainty-resolvable: both face fog, but research/consult would yield more info — option space includes "delay and research." Mutual-uncertainty-intrinsic: both face fog that no work resolves (S4 Priya: startup outcome fundamentally unpredictable). You-know-more / they-know-more: dyadic asymmetry against the most-salient counterparty. (Open question parked: by-position vs. by-effort split within asymmetric values — see parked questions.)
9	Counterparty agency	static / reactive / strategic	Other side: parameter or game-player?
10	Visibility	private / dyadic-observed / group-observed / publicly-observed / deferred-audit	Is the choice witnessed? Enumeration refined 2026-05-30 across two passes (initial rushed pass added deferred-audit; rigorous re-attack added group-observed). Group-observed (3–20 named watchers, peer-accountability dynamic — committee, hiring panel, queue at a carousel) is structurally distinct from publicly-observed (>20 anonymous crowd, performance-to-audience dynamic). Deferred-audit covers cases where decision is unobserved live but reviewed later (recordings, logs, board accountability, FOIA, surveillance). The May 27 reference set undersamples non-private values — corpus property worth flagging.
11	Stakes magnitude (NEW — 2026-05-30 swing)	trivial / meaningful / serious / life-defining	Worst-case cost on most-costly dimension. Added after constructing same-vector pair (small lie to spouse vs. lie to federal investigator) that the rest of the cube couldn't separate. Specified objectively, relative to a calibrated reference player (median Pulse user demographics — income, life stage, dependents). Subjective threshold-difference is what we measure, not what we specify.
12	Default outcome	favors-action / favors-inaction-static / favors-inaction-dynamic / forces-choice	What happens if you don't act. Promoted to primitive 2026-05-30 (was provisional). Refined later same day under rigorous re-attack — favors-inaction was conflating static and dynamic flavors. Favors-action: an active outcome happens by default (S2 lighter-path-proceeds, auto-renewal). Sub-shapes (shape-via-vote / override-only / etc.) captured by axis 7 authority. Favors-inaction-static: status quo continues, nothing accumulates (S1, S4, S5, S7, S8). Favors-inaction-dynamic: situation accumulates if you don't act — harm compounds, risk grows (S6 health markers worsen, untreated misconduct continues). Activates "how long can I delay" reasoning that static doesn't. Forces-choice: you can't escape being read as having chosen — physical or social mechanism (carousel gate, S3 Maya's reading).

Active content-layer dimensions¶

#	Dimension	Values	Notes
C1	Cost categories in tension	subset of	Which value categories pull against each other in this scenario. Load-bearing for distinguishing S1 vs S5.

Provisional / under-question¶

(Both provisional axes resolved 2026-05-30. Identity load → measured outcome. Default outcome → promoted to primitive.)

None currently. Add new candidates here as they emerge.

Submerged candidates (didn't make primary cube, kept logged)¶

Candidate	Why submerged	Where it lives instead	Date submerged
Promise weight × promise recipient	Not primitive — only fires when a promise exists. Generalized into "prior commitment" (axis 1).	Special case of axis 1 + axis 6	2026-05-30
Stakeholder count (1/2/group/system)	Counting wasn't load-bearing — the distribution and distance mattered.	Split into scope (axis 5) and relational distance (axis 6)	2026-05-30
Cost type taxonomy as axis	Not an axis — it's a taxonomy of damage. The axis is which categories are in tension.	Content-layer dimension C1	2026-05-30
Time pressure (single axis)	Was conflating two independent things.	Split into decision deadline (axis 2) and consequence horizon (axis 3)	2026-05-30
Identity load as scenario axis	Three test scenarios (solo novelist / private retirement / trolley) all showed identity load depending on the player's internalized self-image, not the scenario's structure. Not a scenario property at all.	Moved to measured outcomes / signal proxies bin. Observed in rationale richness + self-aware aspiration gap signal.	2026-05-30

Post-promotion misses (content rules the design pass didn't enumerate)¶

Items the cube model didn't surface during creation but should have. Different category from submerged candidates above — those were candidate axes that didn't survive scrutiny. These are content rules sitting alongside cube axes (siblings, not dimensions) that the design pass treated as "out of scope" and therefore never enumerated. Each one is a class of error the next design pass should check for.

Rule	Sibling to	How it surfaced	Why the model missed it	Fix
Earn the deadline — every scenario must name ONE of four external anchors that makes delay structurally impossible (counterparty event with a date / calendar fact / decaying physical or operational window / named default that fires). Generic urgency ("the team needs to decide") fails.	Axis 2 (decision deadline)	Multiplayer starter-corpus review 2026-05-30. Scenario §1 ("€1.4M in 25 minutes") read as theatrical because the deadline value was tagged on the axis but the prose didn't name WHY delay was impossible. Same gap in §2 (sundown — why?) and §3 (vote tonight — why not next week?).	The cube design pass focused on enumerating structural axes (what values a scenario commits to: which time bucket, which scope, which authority) and treated content rules as belonging to the generator prompt — a different surface, not under design. But the deadline-mechanism rule is load-bearing for whether axis 2 even means anything to the reader: tagging "minutes" without earning it produces unfooted urgency. The class of error: load-bearing content rules can hide outside the structural cube because the cube's framing puts them somewhere else.	Codified as content-rule sibling note under axis 2 in `scenario-dimension-cube.md`. Generator Principle 8 (v1.9.0) augmented with the "why not delay?" escape and the four anchor types. `planning.deadlineMechanism` field added. Case-study checklist item 5 inserted. All 8 multiplayer starter scenarios revised.

Meta-lesson for next design pass: when promoting the cube, also do an explicit pass enumerating which content rules are sibling to which axes. For each axis, ask: "what does the prose need to deliver for this axis value to land?" If the answer is "the generator prompt handles it," check whether that's actually true — Principle 8's pre-revision form covered "why not switch later?" but not "why not delay?", so the cube's reliance on the generator was incomplete.

Derived dimensions to watch¶

If the cube is internally consistent, certain combinations of primitives should reliably produce certain emergent properties. Track these to validate the cube.

(Currently empty — both prior candidates resolved 2026-05-30. Identity load went to measured outcomes, default outcome went to primitive.)

Measured outcomes / signal proxies¶

Not scenario axes — properties of the player × scenario interaction, observed in response. The cube doesn't specify these; the response reveals them.

Property	Observed in	Notes
Identity load	Rationale richness, conviction-without-articulation, explicit self-aware aspiration gap statements	High when the scenario's topic intersects what the player has internalized as self-image. Player-conditional, not scenario-conditional. Connects to the self-aware aspiration gap signal (`experiment-new-domain-signals.md`).
Subjective stakes threshold	Conviction × choice direction at calibrated objective magnitudes	Cube specifies objective magnitude; player response reveals where their subjective threshold sits. "Math is the math" at €80K = high threshold; sleepless at €5K = low.

Domain hints (not cube axes, but useful generator inputs)¶

Once a domain is chosen, certain content patterns are conventional:

Career — counterparty types: manager, recruiter, founder; typical deadline: days-to-weeks; typical cost tension: financial vs. developmental vs. relational
Health — counterparty types: doctor, partner, employer; typical deadline: weeks; expertise asymmetry common
Relational — counterparty types: intimate, close-friend, family; typical visibility: dyadic; integrity often in tension
Financial Tier 4–7 — counterparty types: market (static), advisor, spouse; typical visibility: private
Governance — boardroom — institutional scope, financial+reputational tension, strategic counterparties
Governance — sci-fi — raises stakes magnitude. CoachJ noted he chooses differently in sci-fi because lives are at stake. This is evidence that domain surface affects which cube cells are typical — sci-fi pushes magnitude up by default.
Governance — underground — TBD: needs decomposition

Test scenario decompositions¶

Three reference scenarios decomposed under the working cube. Used as ground truth for whether the cube separates scenarios that feel different to play.

Carousel (slot-5 alternate)¶

Axis	Value
Prior commitment	strong (promise to daughter)
Decision deadline	minutes (60s — snap-deliberation mode, not pure reflex)
Consequence horizon	mixed (event today, trust years)
Reversibility	partial
Scope	small group (daughter, friend, missing child)
Relational distance	intimate + close
Authority	partial (parental only)
Info topology	symmetric
Counterparty agency	static (panicked friend, 5yo daughter, missing child)
Visibility	publicly observed (queue)
Stakes magnitude	serious (child safety in worst branch)
Cost categories in tension	relational × relational × physical

S3 — Aaron's secret¶

Axis	Value
Prior commitment	strong (promised secrecy)
Decision deadline	minutes (90s — snap-deliberation mode)
Consequence horizon	years
Reversibility	none
Scope	small group (friend circle)
Relational distance	intimate + close
Authority	none
Info topology	you-know-more
Counterparty agency	strategic (Maya is reading you)
Visibility	dyadic observed
Stakes magnitude	serious (marriage)
Cost categories in tension	relational × relational × integrity

S8 — debt vs. invest¶

Axis	Value
Prior commitment	none
Decision deadline	soft
Consequence horizon	lifelong (15+ yr)
Reversibility	full
Scope	self (+ spouse)
Relational distance	self
Authority	full
Info topology	symmetric
Counterparty agency	static (market)
Visibility	private
Stakes magnitude	meaningful (€80K)
Cost categories in tension	financial × integrity (psychological-security)

Discrimination check: All three decompose distinctly. Carousel vs. S3 share commitment/deadline/scope/distance/magnitude but differ on authority, info topology, reversibility, counterparty agency, visibility. Cube earns its keep.

Near-collapse to track: S1 (career relocation) and S5 (Big Tech vs climate) decompose to nearly identical structural vectors. They separate only via C1 (cost categories in tension) — S1 has caregiving in the mix, S5 doesn't. This is why C1 stays load-bearing and shouldn't be demoted to pure content.

Swing log (attempts to break the cube)¶

Date	Construction	Result	Cube response
2026-05-30	Same-vector pair: lie to spouse (small social) vs. lie to federal investigator (perjury). Identical structural decomposition.	Broke the cube. Same player picks differently.	Added stakes magnitude as axis 11.
2026-05-30	Vary counterparty agency only: parent in hospice (static) vs. parent demanding presence (strategic), otherwise same vector.	Cube held. Scenarios feel meaningfully different.	Validates counterparty agency as load-bearing primitive.
2026-05-30	Identity-load derivation test: three scenarios (solo novelist / private retirement / trolley) constructed with varying integrity-cost / scope / visibility to test whether identity load is derived from those.	Derivation failed — but also rejected primitivity. All three showed identity load depending on whether the topic intersected the player's self-image.	Identity load reclassified as measured outcome, not scenario axis. Default outcome promoted to confirmed primitive (it's a scenario property, unlike identity load).
2026-05-30	Scope × distance independence test: three pairs constructed. (1) Marco-breakup vs Marco-40th — vary scope, hold distance close. (2) Family cabin vs hiking-club equipment — vary distance, hold scope small-group. (3) Mislead contractor vs mislead zoning office — same scope, exposed an under-enumerated value on distance.	Both axes held independent. But pair 3 surfaced that axis 6 was missing a value: "institutional-abstract" (no specific identifiable person), previously being conflated with "stranger" but playing meaningfully different.	Added "institutional-abstract" as new value on axis 6. No structural axis added or removed.
2026-05-30	Authority enumeration test: CEO-firing-CFO (own decision, board reviews) vs senior-IC-advocating-to-CEO (CEO's decision, you input). Both currently tagged "partial."	Enumeration conflation, same shape as the axis-6 finding. "Partial" was collapsing two structurally distinct cases — decide-with-oversight vs. influence-only. Cross-checked against counterparty agency, cost categories, default outcome — none captured the distinction.	Refined axis 7 enumeration from {full / partial / none} to {full / oversight / influence-only / none}. No structural axis added or removed.
2026-05-30	Systematic sweep on remaining seven axes (1, 2, 3, 4, 8, 10, 12). Each got a conflation attack (same-value pair plays differently?) and a missing-value attack (unhandled case?). Flagged as rushed — each axis got 1–2 test pairs instead of the 3+ that earlier swings had. Conclusions kept but flagged for redo.	Mixed results. Axes 1, 2, 5, 9, 11, 12 held cleanly (provisionally). Axes 3, 4, 8, 10 had enumeration gaps or conflations. Same pattern as axes 6 and 7 — structural axis list is stable, enumeration values needed sharpening.	Axis 3: added "generational" value. Axis 4: split "partial" into recoverable/permanent. Axis 8: split "symmetric" into complete/mutual-uncertainty. Axis 10: added "deferred-audit" value. All to be re-attacked individually. No structural axes added or removed.
2026-05-30	Rigorous re-attack on axis 10 (visibility). Three conflation attacks across "publicly-observed," "private," and "dyadic-observed" value-pairs. Three missing-value attacks (anonymous-observable, selective-visibility, observed-but-mute). Cross-checked against May 27 reference set.	Rushed pass got deferred-audit right but missed a needed split between small-group (3–20 watchers, named, peer-accountability) and broad-public (>20 anonymous, performance-to-crowd). Carousel scenario didn't have a clean home under the rushed enumeration.	Added group-observed as fifth value. Full enumeration: {private / dyadic-observed / group-observed / publicly-observed / deferred-audit}.
2026-05-30	Rigorous re-attack on axis 8 (information topology). Four conflation attacks (within you-know-more, mutual-uncertainty, symmetric-complete, they-know-more). Three missing-value attacks (three-party topology, layered common knowledge, time-displaced asymmetry). Cross-checked against May 27 reference set — S4 Priya was the cleanest test.	Rushed pass got symmetric vs. mutual-uncertainty split correct. Re-attack found that mutual-uncertainty itself splits cleanly into resolvable (research/consult yields more) vs. intrinsic (no work resolves the fog) — different option spaces. Also found by-position/by-effort distinction within asymmetric values; parked as open question rather than splitting now.	Refined enumeration to {symmetric-complete / mutual-uncertainty-resolvable / mutual-uncertainty-intrinsic / you-know-more / they-know-more}. Added parked question #4 (by-position/by-effort).
2026-05-30	Rigorous re-attack on axis 4 (reversibility of consequence). Four conflation attacks (within full, partial-recoverable, partial-permanent, none). Two missing-value attacks (mixed reversibility, probabilistic reversibility). Cross-checked against May 27 reference set — corpus clusters in partial-permanent, undersamples full and none.	Rushed pass got partial split correct. Re-attack found: (1) "Full" definition needed sharpening — undo restores original state at no cost, tight criterion most decisions fail. (2) "None" is rarer than implied — mostly deadline/window-close, not action-irreversibility. (3) Active vs. passive recovery within partial-recoverable parked as open question (same family as parked #4). (4) Multi-component reversibility handled via dominant-component tagging rule.	Enumeration unchanged {full / partial-recoverable / partial-permanent / none} but value definitions sharpened and tagging rule added. Added parked question #5 (active/passive recovery).
2026-05-30	Rigorous re-attack on axis 3 (consequence horizon). Five conflation attacks (within immediate, months, years, lifelong, generational). Three missing-value attacks (decades, episodic, discontinuous/delayed). Cross-checked against May 27 reference set.	First reversion of a rushed addition. "Generational" is structurally a combination (lifelong horizon + future-generations scope), not a primitive horizon value. Conceptually clean but factored incorrectly — duration belongs in horizon, who-is-affected belongs in scope.	Reverted axis 3 to {immediate / months / years / lifelong}. Added "future-generations" to axis 5 (scope) as distinct from institutional-abstract. Tagging rule added for active-phase + residue (use dominant duration; residue captured by reversibility).
2026-05-30	Rigorous re-attack on axis 1 (prior commitment). Three conflation attacks (within none, soft, strong). Four missing-value attacks (anti-commitment, multi-party conflict, retracted, conditional-firing). Cross-checked against May 27 reference set — spans enumeration evenly.	First clean confirmation of a rushed conclusion. Enumeration holds with no sharpening needed. Flavors of strong and soft are captured by C1 + content-layer + measured outcomes.	No changes to enumeration. Value definitions slightly elaborated in axis table.
2026-05-30	Rigorous re-attack on axis 2 (decision deadline). Five conflation attacks (within each value: seconds, hours, days, weeks, soft). Three missing-value attacks (already-passed, multi-stage, counterparty-controlled). Cross-checked against May 27 reference set — carousel and S3 re-tagged from seconds to minutes.	Rushed pass missed the seconds/minutes split. "Seconds" was conflating pure-reflex (<5s, no time to verbalize) with snap-deliberation (5s–5min, time for one mental walk-through) — structurally different cognitive modes. Reference set had zero pure-reflex scenarios; what I'd tagged "seconds" was actually minutes-mode.	Added minutes as new value. Refined enumeration: {seconds / minutes / hours / days / weeks / soft}. Months-deadline not added (corpus doesn't push it, risks confusion with months-horizon).
2026-05-30	Rigorous re-attack on axis 12 (default outcome). Three conflation attacks (within favors-action, favors-inaction, forces-choice). Three missing-value attacks (multi-stage, counterparty-controlled, default-not-knowable). Cross-checked against May 27 reference set — S2 was mis-tagged under rushed values (re-tagged from inaction to action since lighter path proceeds by week 6).	Rushed pass missed static vs. dynamic split inside favors-inaction. Static default (status quo continues) and dynamic default (situation accumulates: harm compounds, risk grows) activate different reasoning shapes — dynamic triggers "how long can I delay" framing that static doesn't. S6 health lifestyle is the clear dynamic case; others in the reference set are static.	Refined enumeration: {favors-action / favors-inaction-static / favors-inaction-dynamic / forces-choice}. Also surfaced a tagging-correctness fix on S2.

Parked questions (come back to)¶

Governance design cube — same or separate? Hypothesis: ONE cube, governance occupies a narrow low-entropy subspace. To validate: tag the 152 governance scenarios with cube vectors after cube stabilizes; check clustering. Likely finding: governance has blind spots in the cube it doesn't know it has.
Stakes magnitude in the existing governance generator. Strong prediction: the governance corpus has a wide stakes-magnitude range (board margin call → civilizational sci-fi extinction) but it isn't being explicitly varied per scenario — it's whatever the surface domain implies. Making it an explicit knob would let the generator produce corner cases the corpus undersamples (e.g., boardroom at life-defining magnitude; sci-fi at meaningful magnitude). Do this comparison either after the next 1–2 swings or as part of cube-stabilization — CoachJ's call on ordering.
Domain as content-layer surface vs. as cube-relevant variable. Sci-fi pushes stakes magnitude up by default; boardroom doesn't. Does that mean "domain" is a generator setting that pre-loads certain cube cells? Probably yes — but worth being explicit in the final doc.
By-position vs. by-effort within asymmetric info topology. Real distinction surfaced in axis 8 re-attack: you-know-more-because-they-couldn't-have-known (doctor) vs. you-know-more-because-they-didn't-read-the-fine-print (lazy partner) imply different disclosure duties. Currently treated as content-layer texture handled via C1 cost-categories. Promote to cube split if play testing shows C1 doesn't carry the distinction.
Active vs. passive recovery within partial-recoverable. Real distinction surfaced in axis 4 re-attack: K = publicly insulted colleague (recovery requires sustained intervention) vs. L = bad investment (recovery is passive — wait for market). The recovery-work-as-cost piece isn't cleanly captured by C1 (which tags cost-categories at decision moment, not recovery path-shape). Currently treated as content-layer / C1-adjacent. Same family as parked question #4 — both are "shape of consequence path" rather than "structural position." Promote together if either turns out to matter.
~~Identity load — primitive or derived?~~ Resolved 2026-05-30: neither. Reclassified as measured outcome.
~~Default outcome — primitive or derived?~~ Resolved 2026-05-30: confirmed primitive (axis 12).

Falsification conditions for the cube as a whole¶

The cube is wrong if:

Two scenarios with identical structural-cube vectors produce reliably different responses from the same player, even after content-layer dimensions (C1, etc.) are held identical. → Missing structural axis.
Varying a single axis while holding others constant produces no behavioral difference across players. → Axis is not load-bearing; submerge.
Cross-domain experiment: hold cube vector constant, vary domain surface. If signals don't generalize for a player who should be consistent (per their existing profile), either the cube is missing something or domain isn't the neutral surface we assumed.

Governance generator comparison (2026-05-30)¶

Mapped the cube against src/lib/ai/scenario-generator.ts SCENARIO_GENERATION_PROMPT (lines 107–438) plus validation logic (705–1001).

Cube axes the generator implicitly handles¶

Axis	How the generator handles it	Match quality
7 (authority)	Principle 14's Tier 1 / Tier 2 / Tier 3 framing matches the cube's refined enumeration {full / oversight / influence-only / none}. Tier 1 = full authority (CFO, Steward, COO). Tier 3 = pure outside advisor = influence-only. The "worldview tax" guardrail catches axis 7 collisions with load-bearing characters.	Strong match — independent convergence on the same distinction. External validation of the cube's axis 7 refinement.
6 (relational distance)	Cast block with role + stake fields renders position relative to player.	Implicit but consistent
5 (scope), 1 (prior commitment), 3 (horizon)	Scope implicit in worst-case naming. Prior commitment in `priorHistory` planning field. Horizon calibrated by Principle 13 (AI-era realism).	Partial — captured per-scenario but not enumerated
9 (counterparty agency)	Per-option pushback rule ("Vasquez resigns if you take this path") produces strategic counterparties when used	Partial — not a knob

Axis	Status in generator	Predicted effect of adding
11 (stakes magnitude)	No knob. Magnitude rides on surface domain, not deliberate parameter. Boardroom-at-life-defining and sci-fi-at-trivial are inconsistent. Prediction from earlier in session confirmed.	Expand corpus to corner cases currently undersampled.
12 (default outcome)	Partial. Decision prompt mentions deadline-default but doesn't distinguish favors-action / inaction-static / inaction-dynamic / forces-choice. Most scenarios default to inaction-static.	Dynamic-default and forces-choice scenarios become reliable rather than rare.
8 (information topology)	Mostly absent. Default is symmetric-complete. S3-style asymmetric scenarios rare in governance corpus.	Unlock the asymmetry/uncertainty corner.
10 (visibility)	Mostly absent. Private / group-observed / publicly-observed / deferred-audit not enumerated.	Unlock observation-mechanism corner.
4 (reversibility flavor)	Lock-in required (good) but recoverable vs. permanent not distinguished. Tends to default partial-permanent.	More reliable production of recoverable-consequence scenarios.
2 (decision deadline)	Scan strip's deadline row forces a value, but cognitive-mode distinction not enforced. Clusters in days-to-weeks.	Unlock minutes/hours/months edges.

What the generator has that the cube doesn't (and shouldn't)¶

Driver tagging on options (8 values) — signal/option tags, ADR-024 territory, not cube.
Trigger tagging on scenarios (10 values) — content-layer summary tags combining cube cells.
"Value Tensions to Surface" list (lines 180–196) — pre-emptively renamed 2026-05-30 from "Dimensions to Test" to avoid collision with the cube's structural "dimensions" vocabulary. The TS field dimensionsTested was renamed to tensionsSurfaced in the same pass. Field is in-memory only (not persisted), so the rename was safe. Test prompts in scripts/test-prompts/ regenerated. Original list is a value-pair list (each item is two values in opposition), distinct from cube dimensions which describe scenario-space structure.
Content/quality discipline (name-diversity rules, gender-balance enforcement, jargon test, scan strip clarity, banned voice patterns, narrative-bleed detector, lock-in-per-option, cast block coverage check) — critical, not dimensional. Cube specifies structural condition; this discipline renders the condition well into prose.
The planning block structure — namedCharacters / priorHistory / worstCase / lockInMechanism / lockInPerOption / scanStripDraft / castBlockDraft. This is the forcing function: commit to structural shape before drafting prose. This is where a cubeVector field would naturally extend. Adding it to the planning block would force the generator to commit to specific values on all 12 axes before writing.

Governance scenario decompositions (2026-05-30)¶

Pulled two random curated governance scenarios from the production corpus and decomposed under the 12-axis cube.

Scenario A — "The Bar Meeting" (boardroom): advising-CEO frame, days deadline, months horizon, partial-permanent reversibility, institution scope, close+acquaintance distance, influence-only authority, symmetric-complete info, strategic agency, publicly-observed + deferred-audit visibility, serious stakes, favors-inaction-dynamic default. Costs: relational × legal × reputational × integrity.

Scenario B — "The Preemptive Motion" (underground): advising-whistleblower frame, days deadline, years horizon, partial-permanent reversibility, institution + institutional-abstract scope, acquaintance + institutional-abstract distance, influence-only authority, you-know-more + mutual-uncertainty-intrinsic info, strategic agency, deferred-audit visibility, life-defining stakes, favors-inaction-dynamic default. Costs: integrity × legal × reputational × relational.

Findings:

Narrow-subspace hypothesis empirically confirmed. Both share Tier 3 / influence-only authority, institution scope, strategic counterparties, days deadline, partial-permanent reversibility, favors-inaction-dynamic default. Only 3 of 12 axes meaningfully differ (stakes, info topology, distance). The governance corpus appears to occupy a narrow cube subspace as predicted.
Advisor-framing is heavy in the corpus. Both samples are "you're advising X," not "you're X." This produces signals about advice-giving, not deciding. Worth flagging as a potential corpus property — the cube can express it (axis 7 = influence-only), but the concentration in this subspace is noteworthy.
Stakes magnitude varies in the corpus but is uncalibrated. Bar Meeting = serious; Preemptive Motion = life-defining. Magnitude rides on surface domain, not deliberate parameter. Empirical confirmation of the missing-knob prediction.
Information topology — partial corroboration. Boardroom example = symmetric-complete (the predicted default). Underground = you-know-more + mutual-uncertainty-intrinsic. Underground theme pushes scenarios into asymmetric territory naturally. Boardroom-governance specifically may be the symmetric-complete cluster.
The cube decomposes both scenarios cleanly. No new axes needed. External validation: the cube survives application to real corpus it wasn't built on.
Surprise — both default to favors-inaction-dynamic, not static. I predicted governance defaults to inaction-static; both samples are dynamic ("situation accumulates by the hour"). Either (a) the corpus uses dynamic more than I predicted, or (b) the random sample is non-representative. Needs more decompositions to confirm. Flagging.
Tier 3 / influence-only refinement is load-bearing in this corpus. Axis 7's rushed-pass split into oversight/influence-only directly applies to both. External validation of that refinement.

Open follow-up: decompose 2–4 more governance scenarios spanning themes (frontier, court, rebuild) and categories (resource-allocation, team-dynamics, values-culture) to confirm or revise findings 1, 4, and 6.

Round 2 decompositions (3 more scenarios — boardroom/resource-allocation, frontier/governance, court/values-culture)¶

Scenario C — "The Board Vote" (boardroom + resource-allocation): Tier 1 CEO, hours deadline, years horizon, oversight authority (recommend → board votes), institution scope, you-know-more info (Cortex conflict), strategic agency, group-observed + deferred-audit visibility, serious stakes, forces-choice default. Costs: financial × relational × integrity × reputational.

Scenario D — "The Marshal" (frontier + governance): committee member, hours deadline, months-years horizon, collective-vote authority (1 of 6), institution scope, symmetric-complete + mutual-uncertainty-intrinsic info, strategic agency, group-observed + deferred-audit visibility, serious-to-life-defining stakes, forces-choice default. Costs: physical × integrity × relational × reputational.

Scenario E — "The Kingdom After Silver" (court + values-culture): council member, weeks deadline, lifelong horizon, collective-vote authority, institution + future-generations scope ✓, symmetric-complete info, strategic agency, group-observed + publicly-observed + deferred-audit visibility, life-defining stakes (for kingdom), forces-choice default. Costs: integrity × relational × historical × reputational.

Findings updated across all 5 decompositions:

Round 1 advisor-concentration finding was a sampling artifact. Across 5 samples: 1 oversight CEO, 2 collective-vote committee, 2 influence-only advisor. The corpus spans authority types — my N=2 sample landed disproportionately on advisor framings. Correcting.
STRUCTURAL FINDING: "collective-vote" added to axis 7. Both Marshal and Kingdom put the player as 1 of N voters with formal decision power in real time. Neither oversight nor influence-only fits — coalition reasoning and single-vote leverage are a distinct play shape. Axis 7 refined to {full / oversight / collective-vote / influence-only / none}. Third refinement of axis 7 across the session.
Default outcome distribution revised. 2 dynamic / 3 forces-choice across 5 samples. Forces-choice is more common than initially predicted — scenarios with imminent scheduled events (board meetings, committee sessions, agenda items) use scheduling as a forcing function frequently.
Information topology — modal default holds, but variation exists. 3 symmetric-complete, 2 you-know-more, 1 mutual-uncertainty. Boardroom-governance defaults symmetric-complete; underground + competitive-corporate (Board Vote with Cortex conflict) push asymmetric. Hypothesis revised.
Stakes magnitude spans the cube. 2 serious, 1 borderline life-defining, 2 life-defining. Corpus covers magnitudes — without explicit knob, it tracks the surface domain's natural weight.
Future-generations scope value validated by real corpus. Kingdom decomposes cleanly with the value added during axis 3 re-attack. External validation.
"Player has authority or stake" rule holds in all 5. Even advisor scenarios cast the player's recommendation as load-bearing. The new design rule is being followed.
The cube remains stable under real-corpus application. Only one new value (collective-vote on axis 7) needed across 5 decompositions. Convergence continues.

Round 3 decompositions (3 more — rebuild/governance, boardroom/team-dynamics, underground/team-dynamics)¶

Completing world coverage (rebuild added) and category coverage (team-dynamics added).

Scenario F — "The Constitutional Vote" (rebuild + governance): ballot commission member, days deadline, lifelong horizon, collective-vote authority, institution + future-generations scope, close + institutional-abstract distance, symmetric-complete info, strategic agency (all 3 named figures campaigning), group-observed + publicly-observed + deferred-audit visibility, life-defining stakes, forces-choice default. Costs: integrity × relational × historical × reputational.

Scenario G — "The Three Questions" (boardroom + team-dynamics): Margot's legal advisor, strong commitment (NDA), weeks deadline, years horizon, partial-permanent reversibility, small group scope (first non-institution sample), close distance, influence-only authority, you-know-more info (NDA), strategic agency, dyadic-observed + deferred-audit visibility, serious stakes, favors-inaction-dynamic default. Costs: integrity × relational × legal × reputational.

Scenario H — "The Timing Anomaly" (underground + team-dynamics): cell member, hours deadline, years horizon, partial-permanent reversibility, small group + institution scope, close distance, collective-vote authority, mutual-uncertainty-intrinsic info ✓ (is the pattern real?), strategic + static-or-strategic-on-monitors agency, private + deferred-audit visibility, serious-to-life-defining stakes, forces-choice default. Costs: operational × developmental × strategic.

Cross-corpus findings (n=8 decomposed scenarios)¶

Authority distribution: 0 full / 1 oversight / 4 collective-vote / 3 influence-only / 0 none. Collective-vote is the modal authority pattern at 50%. The cube value addition is strongly validated. "Full" is essentially absent from governance corpus — it lives in personal-domain scenarios instead (May 27 S8). Cube spans both; corpus uses a subset.

Default outcome distribution: 5 forces-choice / 3 favors-inaction-dynamic / 0 favors-inaction-static / 0 favors-action. CoachJ's deliberate-design observation empirically confirmed. The corpus aggressively closes the "do nothing" escape hatch. Static defaults are absent from this sample entirely.

Information topology distribution: 4 symmetric-complete / 3 you-know-more / 2 mutual-uncertainty-intrinsic (some overlap). Symmetric-complete is modal at 50% but NOT the universal default I initially predicted from N=2. Asymmetric scenarios exist meaningfully (Board Vote with Cortex conflict, Preemptive Motion, Three Questions NDA, Timing Anomaly surveillance).

Stakes magnitude: 4 serious / 4 life-defining / 0 meaningful / 0 trivial. Corpus spans upper half of the magnitude range; lower half absent.

Decision deadline: 0 seconds / 0 minutes / 3 hours / 3 days / 2 weeks / 0 soft. Tight deadlines dominate — consistent with forces-choice mechanism.

Scope: 7 institution / 1 small group (Three Questions) / 2 future-generations (overlap with institution). Personal/self/dyad scope absent. Heavy institution skew is a corpus property.

Visibility: 5 group-observed / 6 deferred-audit / 3 publicly-observed / 1 dyadic / 1 private (multiple-value scenarios common). Public/group/deferred is heavy.

Structural findings¶

Cube converged. Across 8 decompositions, only one new value (collective-vote) surfaced. No new structural axes. The 12-axis cube + content axis + 2 measured outcomes is stable under real-corpus application.
All re-attack refinements pay off in the corpus. Group-observed (Marshal queue analog), deferred-audit (NDA, court records), mutual-uncertainty-intrinsic (Timing Anomaly, S4 Priya), institutional-abstract (Veriton/Helion), future-generations (Kingdom, Constitutional Vote), collective-vote (4 samples). Every value added during rigorous re-attack has a clean home in real corpus.
Corpus subspace mapped. Governance corpus lives in: institution scope / collective-vote or influence-only authority / serious-to-life-defining stakes / forces-choice or inaction-dynamic default / strategic counterparty agency / group+deferred visibility. That's a narrow but consistent corner of the cube. Personal domain scenarios live in a different corner (May 27 set). Cube spans both.

Content-rule implications for generator integration¶

Collective-vote scenarios should make peer dynamics explicit. Three of four collective-vote samples (Marshal, Kingdom, Timing Anomaly) under-use the peer-dynamics surface area — only Constitutional Vote names multiple campaigners with positions. Generator rule: when axis 7 = collective-vote, name ≥2 other voters with stated leanings; options should include the cost of how each choice reads to those peers (coalition risk, being-seen-as-X risk). This addresses CoachJ's flag that collective-vote is currently more circumstantial than load-bearing.
Forces-choice should remain the default for new scenarios. Corpus empirically validates the design intent. Static defaults should only be used deliberately (when the scenario's signal IS the player's tolerance for inaction).
Information topology should be an explicit knob. Corpus produces asymmetric scenarios but not deliberately — they emerge from surface domain (NDA settings, whistleblower settings, surveillance settings). Adding it as an explicit knob would let the generator dial it independent of theme.
Stakes magnitude should be an explicit knob. Same pattern — corpus spans serious-to-life-defining but not deliberately. Lower-magnitude scenarios (meaningful, trivial) are absent and might unlock undersampled signal regions.
Default outcome surface should extend beyond forces-choice + inaction-dynamic. Corpus has zero static-default and zero favors-action samples across n=12. Each default-value measures a different signal: forces-choice (values under constraint), inaction-static (tolerance for ambiguity), inaction-dynamic (urgency calibration), favors-action (interventionism). The no-escape-hatch principle should be preserved as default-default but other defaults deliberately included for signal coverage. CoachJ accepted this framing 2026-05-30.

Round 4 decompositions (4 more — boardroom/values-culture, frontier/team-dynamics, court/governance, underground/values-culture)¶

Filling in 4 more (theme × category) combinations of the 20-cell grid. Now at 12 of 20.

Scenario I — "The Visit" (boardroom + values-culture): ED, minutes deadline (first in corpus), years horizon, partial-permanent reversibility, institution scope, close + acquaintance distance, full authority (first in corpus), you-know-more info, strategic agency, group-observed + deferred-audit visibility, serious stakes, forces-choice default. Costs: integrity × relational × financial × reputational.

Scenario J — "The Three Tables" (frontier + team-dynamics): advisor to ship captain, soft deadline, lifelong horizon (11–23 yr journey), partial-permanent reversibility, institution + future-generations scope, close + institutional-abstract distance, influence-only authority, symmetric-complete info, strategic agency, publicly-observed + deferred-audit visibility, life-defining stakes, favors-inaction-dynamic default. Costs: relational × integrity × historical × strategic.

Scenario K — "The River Boundary" (court + governance): guild arbitration council, soft deadline, lifelong horizon, partial-permanent reversibility, institution + future-generations scope, acquaintance + institutional-abstract distance, collective-vote authority, you-know-more info (Colthart's finding), strategic agency (Farren logging), group + publicly-observed + deferred-audit visibility, life-defining stakes, favors-inaction-dynamic default. Costs: integrity × historical × relational × institutional.

Scenario L — "The Record That Got Built" (underground + values-culture): advising Nora 14 months later, soft deadline, years horizon, partial-permanent reversibility, small group + institution scope, close + acquaintance + institutional-abstract distance, influence-only authority, you-know-more info, reactive agency, deferred-audit + private visibility, serious stakes, forces-choice-borderline default. Costs: integrity × developmental × historical × reputational.

Cross-corpus findings revised (n=12)¶

Authority distribution: 1 full (Visit) / 1 oversight / 5 collective-vote / 5 influence-only / 0 none. Full authority appears, contrary to earlier n=8 finding. Rare but present. Collective-vote and influence-only remain the modal patterns.

Default outcome distribution: 7 forces-choice / 5 inaction-dynamic / 0 static / 0 favors-action. The static + favors-action gap is robust at n=12. Confirms the deliberate-design pattern AND the recommendation to extend.

Information topology distribution REVISED: 5 symmetric-complete / 6 you-know-more / 2 mutual-uncertainty-intrinsic (overlap). You-know-more is now slightly more common than symmetric-complete. The "governance defaults to symmetric-complete" hypothesis that motivated the [[project-it-depends-traces-to-information-topology]] memory is wrong as a corpus property — the corpus produces asymmetric scenarios systematically when the player has insider expertise (NDA, whistleblower, professional-advisor framings). The "it depends" complaint may have been more specific to scenarios where the cube would tag symmetric-complete but the scenario surface failed to explicitly say so.

Stakes magnitude: 5 serious / 2 borderline / 5 life-defining / 0 meaningful / 0 trivial. Corpus spans upper half of magnitude range; lower half absent.

Decision deadline: 1 minutes (Visit) / 3 hours / 3 days / 2 weeks / 3 soft. Full range minutes-to-soft used. Earlier "clusters in days-weeks" was sampling artifact.

Scope: 9 institution / 3 small group / 4 future-generations (overlap) / 0 personal-self-dyad. Personal scope genuinely absent from governance corpus.

Structural findings (final)¶

Cube has fully converged. 12 decompositions, 5 worlds, 4 categories: zero new structural axes, zero new values needed beyond collective-vote (added in round 2). Every cube value enumerated has a clean home in real corpus.
Every re-attack refinement validated by real corpus. Group-observed, institutional-abstract, future-generations, collective-vote, mutual-uncertainty-intrinsic, deferred-audit, minutes, full authority, you-know-more all surface in samples. The slow-down has been thoroughly vindicated empirically.
The Visit is the structurally complete sample. Every axis fires cleanly and the cube fully explains what makes the scenario work: full authority + minutes deadline + group-observed + serious stakes + forces-choice + you-know-more.
Corpus subspace map confirmed but wider than initial prediction. Governance corpus lives in: institution-or-small-group scope / collective-vote or influence-only or occasional-full or oversight authority / serious-to-life-defining stakes / forces-choice or inaction-dynamic default / strategic counterparty agency / group+deferred+occasional-public visibility / 5:6 symmetric-vs-asymmetric info topology / minutes-through-soft deadlines. Wider corner than n=2 suggested but still narrower than the cube's full space.

Round 5 decompositions (4 more — frontier/resource-allocation, court/team-dynamics, rebuild/values-culture, underground/resource-allocation)¶

Now at 16 of 20 (theme × category) cells. Coverage: all 5 worlds, all 4 categories sampled at least twice.

Scenario M — "The Eira Signal" (frontier + resource-allocation): council, hours deadline (6h), months-years horizon, partial-permanent reversibility, institution scope, acquaintance + institutional-abstract distance, collective-vote authority, symmetric-complete info, strategic + static agency, group-observed + deferred-audit visibility, life-defining stakes, forces-choice default. Costs: physical × resource × strategic.

Scenario N — "The Seating Chart" (court + team-dynamics): council member at succession banquet, hours deadline, years horizon, partial-permanent reversibility, institution scope, acquaintance + institutional-abstract distance, collective-vote authority, symmetric-complete info, strategic agency (3 claimants + Harren), group-observed + deferred-audit visibility, life-defining stakes, forces-choice default. Costs: financial × political × relational × strategic.

Scenario O — "The Runner from Fenn's Crossing" (rebuild + values-culture): advisor to caravan, minutes-hours deadline, months-years horizon, partial-permanent reversibility, institution + small group scope, close + institutional-abstract distance, influence-only authority, mutual-uncertainty-resolvable info (depot inventory estimated; option D resolves), reactive + strategic agency, group-observed + deferred-audit visibility, life-defining + serious stakes, forces-choice default. Costs: physical × relational × resource × integrity.

Scenario P — "The Subpoena" (underground + resource-allocation): network member, days deadline (72h), years + lifelong horizon, partial-permanent + none-for-some-options reversibility, institution + small group + future-generations scope, close + institutional-abstract distance, collective-vote authority, you-know-more info, strategic agency, private + deferred-audit visibility, life-defining stakes, forces-choice default. Costs: physical × strategic × legal × integrity.

Cross-corpus findings (n=16)¶

Authority distribution: 1 full / 1 oversight / 8 collective-vote (50%) / 6 influence-only / 0 none. Collective-vote firmly modal at the larger sample.

Default outcome distribution: 11 forces-choice (69%) / 5 inaction-dynamic / 0 static / 0 favors-action. Forces-choice strongly dominant. Static and favors-action genuinely absent.

Information topology distribution: 7 symmetric-complete / 7 you-know-more / 4 mutual-uncertainty (overlap). Essentially balanced. Hypothesis revision holds.

Stakes magnitude: 5 serious / 3 borderline / 8 life-defining. Life-defining is now majority.

Decision deadline: 1 minutes / 1 minutes-hours / 5 hours / 4 days / 2 weeks / 3 soft. Full range sampled.

Scope: institution dominant; future-generations in 5+ samples; personal/self/dyad absent.

Cube convergence holds. No new structural axes or values across 16 decompositions. Same convergence signal as n=8 and n=12.

Round 6 — grid complete (4 more — frontier/values-culture, court/resource-allocation, rebuild/resource-allocation, rebuild/team-dynamics)¶

All 20 (theme × category) cells now sampled at least once.

Scenario Q — "The Sun on the Wall" (frontier + values-culture): advisor, hours deadline, years horizon, partial-permanent reversibility, institution scope, close + institutional-abstract distance, influence-only authority, symmetric-complete info, strategic agency, publicly-observed + deferred-audit visibility, life-defining stakes, forces-choice default.

Scenario R — "The Closed Session" (court + resource-allocation): kingdom council member, weeks deadline (90d), years horizon, partial-permanent reversibility, institution scope, acquaintance + institutional-abstract distance, collective-vote authority, you-know-more info (closed session), strategic agency, private + deferred-audit visibility, life-defining stakes, forces-choice default.

Scenario S — "The Split Road" (rebuild + resource-allocation): advisor to caravan, hours deadline (2h), months horizon, partial-permanent reversibility, small group scope, close distance, influence-only authority, mutual-uncertainty-resolvable info, static agency (terrain), dyadic + group-observed visibility, life-defining + serious stakes, forces-choice default.

Scenario T — "One Rule for Them" (rebuild + team-dynamics): council member, hours-days deadline, years horizon, partial-permanent reversibility, institution scope, close + institutional-abstract distance, collective-vote authority, symmetric-complete info, strategic agency, group-observed + publicly-observed visibility, life-defining stakes, forces-choice default.

Final cross-corpus findings (n=20, full grid coverage)¶

Authority distribution: 5% full / 5% oversight / 50% collective-vote / 40% influence-only / 0% none. Shape stable across n=4, n=8, n=12, n=16, n=20. Collective-vote at 50% is robust.

Default outcome distribution: 75% forces-choice / 25% inaction-dynamic / 0% static / 0% favors-action. The static + favors-action gap is now confirmed at full grid coverage. Empirically robust — corpus genuinely does not produce these.

Information topology distribution: 45% symmetric-complete / 40% you-know-more / 25% mutual-uncertainty (overlap). Roughly balanced. Hypothesis revision firmly holds — corpus is not symmetric-complete-by-default. Asymmetric topology emerges systematically when player has insider expertise (NDA, whistleblower, advisor with privileged knowledge, closed session).

Stakes magnitude: ~25% serious / ~15% borderline / ~60% life-defining. Heavily life-defining-skewed. Meaningful and trivial absent.

Decision deadline: 1 minutes (5%) / 7 hours (35%) / 5 days (25%) / 3 weeks (15%) / 4 soft (20%). Full range used, hours-mode most common.

Scope: institution dominant (~75%); small group (~15%); future-generations appears in 25%+ when constitutional/precedent themes activate; personal/self/dyad genuinely absent.

Cube convergence — final¶

After 12 swings, governance-generator comparison, and 20 corpus decompositions spanning all 20 (theme × category) cells:

Zero new structural axes added since the very first swing (stakes magnitude).
One value added since the rigorous re-attack pass (collective-vote on axis 7, surfaced at n=4).
Every cube value enumerated has a clean home in real corpus. None vestigial.
The cube cleanly decomposes every sampled scenario regardless of theme, category, or domain.

The cube has converged. Ready for promotion to scenario-dimension-cube.md as the canonical doc.

Round 7 — within-cell variance check (n=29, +9 scenarios across 3 cells)¶

Pulled 3 additional scenarios per cell from boardroom/governance, frontier/governance, underground/governance to test whether single-sample-per-cell tagging misses within-cell diversity.

Boardroom + governance (n=4): Bar Meeting + Cardiac Unit + Forensic Report + Pause. Authority varies (3 infl-only, 1 oversight). Default varies (2 forces, 2 dynamic). Info varies (2 symmetric, 2 asymmetric). Stakes vary (serious to life-defining). Most diverse cell — wide cube signature.

Frontier + governance (n=4): Marshal + Recount Decision + Buyout Ballot + Sensor Drift. Authority varies (2 collective-vote, 2 infl-only). Default clusters (all forces-choice). Info varies. Stakes cluster (all life-defining). Scope clusters (all institution + future-generations). Intermediate clustering.

Underground + governance (n=4): Preemptive Motion + Quarterly Close + Calloway's Condition + Della Foss. Authority clusters (3 infl-only, 1 collective-vote). Default clusters (3 forces-choice, 1 dynamic). Info clusters (all you-know-more, with uncertainty overlay on 2). Stakes cluster (all life-defining). Scope clusters (institution + small group). Tightest cell — narrow cube signature; "whistleblower formula" with insider-expertise, life-defining stakes, forces-choice, you-know-more.

Variance finding — cells cluster at different tightness¶

Different (theme × category) cells have different cube-signature tightness: - Underground + governance: tight signature, repeatable formula across scenarios - Frontier + governance: intermediate — clusters on stakes/scope (sci-fi defaults), varies on authority and info - Boardroom + governance: wide signature, scenarios span much more of the cube

Actionable implication for corpus design: Boardroom can be deliberately diversified along most axes (corpus shows the variance is producible). Underground has a tight signature that may be harder to break out of — if the corpus wants underground scenarios with symmetric-complete info or with collective-vote authority, the existing patterns may need to be deliberately broken rather than randomly sampled.

Re-audit check (per [[feedback-changes-trigger-reaudit]]): Did the 9 new decompositions change any prior findings? Updated authority distribution at n=29: 1 full / 2 oversight / 13 collective-vote (45%) / 13 influence-only (45%) / 0 none. Collective-vote dropped from 50% to 45%; influence-only rose to 45%. Both still essentially balanced; no structural finding invalidated. Default outcome stays at ~75% forces-choice with 0% static / 0% favors-action. Info topology still ~balanced. No prior finding invalidated. No re-audit triggered.

Information-topology hypothesis revision (2026-05-30)¶

The session opened with a strong hypothesis: governance corpus defaults to symmetric-complete information topology, and the "it depends" complaint traces to this missing constraint. After n=12 decompositions, this hypothesis is revised:

The corpus does produce asymmetric scenarios systematically (6 of 12) when the player is positioned as someone with insider expertise (NDA recipient, whistleblower, professional advisor).
The "it depends" complaint may have been about a subset of scenarios where the cube would tag symmetric-complete but the surface failed to make the constraint explicit. Not all governance scenarios.
The original memory ([[project-it-depends-traces-to-information-topology]]) holds for those specific cases but should not be read as "all governance defaults to symmetric-complete."
The integration recommendation still holds: making information topology an explicit knob would let the generator dial it deliberately rather than letting it emerge from theme.

Headline findings¶

The cube validates against reality. Generator was built without a dimensional model, just intuition and iteration. Independent emergence of Tier 1/2/3 (= axis 7) in the generator confirms the cube's refinement to {full / oversight / influence-only / none}. Strong external validation.
The cube adds four explicit knobs the generator doesn't currently have: stakes magnitude, default outcome (especially dynamic), information topology, visibility. These are the blind spots — currently-undersampled corners.
Cube and generator are complementary layers. Cube = structural condition. Generator = render the condition well. Integration: cube fills planning block; generator content rules realize it into scenario surface.
Empirical confirmation that governance occupies a narrow cube subspace. Generator defaults (symmetric-complete information, favors-inaction-static, private/dyadic visibility, partial-permanent reversibility) describe one corner. The 152 governance scenarios likely cluster there — the audit-after-cube-stabilization parked question becomes the way to verify this.

What this doc is NOT yet¶

Not the committed deliverable (scenario-dimension-cube.md) — that gets written when the cube has survived multiple swings and a small-corpus decomposition pass.
Not a generator spec — the generator prompt informed by the cube comes later.
Not an audit of the existing corpus — that's the governance-cube parked question.