Skip to content

Scenario Corpus Audit — Cube Decomposition (n=45)

Date: 2026-05-30 Sample: 45 curated scenarios stratified across 5 themes × 4 categories, drawn from the 316 curated scenarios with scan_strip IS NOT NULL in production. Distribution roughly proportional to cell size (boardroom 12, court 8, frontier 9, rebuild 8, underground 8). Model: docs/research/foundations/scenario-dimension-cube.md — 12 structural axes + C1 content axis. Scope: Corpus map, not rewrite plan. Decomposition table at /tmp/audit_decomp.json (45 records).


TL;DR

The corpus is concentrated in a single dominant cell — institution-scope + life-defining stakes + forces-choice default + sym-complete information + partial-permanent reversibility — and 47% of scenarios sit inside it. Six cube axis values are entirely absent from the sample (self/dyad scope, trivial stakes, full reversibility, they-know-more topology, favors-action default, seconds/minutes deadline). The design rules cited as canon (forces-choice heavy, authority-or-stake required) are largely upheld — but seven scenarios slip the authority/stake rule via the "advisor-to-decision-maker" framing, concentrated in boardroom.

The corpus reads as a high-stakes institutional-governance sandbox. That is partially deliberate (governance is the gameplay frame) and partially a default cluster the generator drifts into. The cells most worth diversifying are personal-scope, the asymmetric-information dual (they-know-more), the low-stakes range, and the favors-action default.


(a) Corpus-wide axis distribution

n=45. Multi-value tags counted once per value.

Axis Top values (count, % of scenarios)
Prior commitment soft 33 (73%) · strong 8 (18%) · none 4 (9%)
Deadline hours 17 (38%) · days 14 (31%) · weeks 10 (22%) · soft 4 (9%) — seconds/minutes absent
Horizon years 31 (69%) · lifelong 11 (24%) · months 2 (4%) · immediate 1 (2%)
Reversibility partial-permanent 38 (84%) · partial-recoverable 5 (11%) · none 2 (4%) — full absent
Scope institution 38 (84%) · future-generations 13 (29%) · small-group 7 (16%) — self/dyad absent
Distance inst-abs 33 (73%) · close 26 (58%) · stranger 6 (13%) · intimate 2 (4%) · acquaintance 1 (2%) — self absent
Authority influence-only 13 (29%) · collective-vote 12 (27%) · full 11 (24%) · oversight 9 (20%) — none absent (correctly, per design rule)
Info topology sym-complete 25 (56%) · you-more 10 (22%) · mut-unc-resolvable 8 (18%) · mut-unc-intrinsic 2 (4%) — they-more absent
Counterparty agency strategic 30 (67%) · static 9 (20%) · reactive 6 (13%)
Visibility deferred-audit 35 (78%) · group-observed 25 (56%) · publicly-observed 8 (18%) · private 6 (13%) · dyadic-observed 1 (2%)
Stakes life-defining 31 (69%) · serious 12 (27%) · meaningful 2 (4%) — trivial absent
Default outcome forces-choice 31 (69%) · favors-inaction-static 8 (18%) · favors-inaction-dynamic 6 (13%) — favors-action absent
C1 cost mix integrity 89% · relational 60% · reputational 56% · physical 56% · financial 51% · developmental 20%

Headline observations

  1. The default cluster is real. 21 of 45 scenarios (47%) carry institution + life-defining + forces-choice simultaneously. Adding partial-permanent reversibility and sym-complete topology covers most of the remainder. The corpus has a center of mass, and it sits squarely in the governance-life-defining-tradeoff cell.
  2. Authority distribution is wider than the n=29 governance-only finding suggested. That earlier audit reported 5%/5%/45%/45%/0%; this broader sample is 24%/20%/27%/29%/0%. full authority is present (11 scenarios), driven heavily by underground (6 of 8). The "full-authority gap" diagnosed in the cube doc was an artifact of the governance-only sub-sample.
  3. Integrity is the dominant cost. It's in C1 for 89% of scenarios. The next most frequent (relational, reputational, physical, financial) cluster in the 50–60% band. Developmental costs are thin (20%) and absent entirely from court and underground.
  4. Forces-choice is honored as design canon. 69% of the sample lands on forces-choice, matching the [[project-forces-choice-is-deliberate-design]] intent. The escape-hatch closures are visible at the prose level (deadlines, lock-in clauses on every option).
  5. Sym-complete information is the modal topology, not the universal one. 56% sym-complete, 22% you-know-more, 18% mut-unc-resolvable. The cube doc's 2026-05-30 revision to [[project-it-depends-traces-to-information-topology]] is consistent with this — the corpus does specify topology, including asymmetric topology, when domain pressure warrants.

(b) Under-sampled corners worth filling

Cells where the corpus produces 0–2 scenarios at n=45 sample size. These are not all equally worth filling — flagged with priority.

High-value gaps (recommend deliberate fill)

Gap n Why it matters
Self/dyad scope 0 / 0 The corpus has no scenarios where only the player or the player + one other is materially affected. Cube allows it (S6-style personal health); corpus produces nothing. Every governance scenario routes through an institution. This is the biggest single gap.
they-know-more topology 0 Doctor/expert-counterparty scenarios, patient-side, legal-counsel-side, recruit-vs-recruiter info gap — all sit in this cell. Corpus produces the inverse (you-know-more, 22%) freely but the dual is empty.
favors-action default 0 Auto-renewals, vote-passes-without-you, default-decisions-the-system-makes. The cube has it; corpus produces zero. Closely tied to scenarios where the player's choice is whether to intervene in a process already underway.
Trivial / meaningful stakes 0 / 2 Every scenario is calibrated at serious-or-above. Meaningful and trivial slots reveal where the player's subjective threshold sits at low magnitudes — important for the magnitude-axis dial the cube doc flagged as currently absent from the generator.
Full reversibility 0 Tight definition (undo at no cost) — rare in real decisions, but valuable in the cube precisely because it isolates the reversibility variable. Without it, partial-permanent always co-fires with high stakes.
Personal-stake life-defining (self+life-defining) 0 The S6-style "your own health/finance/life-shape" cell that drives identity-load measurement. Currently impossible to reach inside the world-themed corpus.
Symmetric-complete + private + life-defining ("lonely-choice" cell) 1 The cell where no one is watching and no information is missing — pure values revelation. Underground has the private-visibility cell partially; the other themes don't reach it.

Lower-priority gaps

  • Immediate / months horizon (1 / 2) — corpus stretches to years/lifelong by default. Short-horizon decisions reveal different reasoning shapes.
  • Seconds/minutes deadline (0 each) — present in the cube but absent from the corpus. The carousel-style reflex scenario doesn't have a world-themed analog yet.
  • Intimate / acquaintance distance (2 / 1) — corpus is bimodal (close + inst-abs); the middle distances are thin.
  • mut-unc-intrinsic topology (2 — both underground) — the "no work resolves this" topology is a deliberate signal of risk-handling vs. information-seeking reasoning, and shows up only in the underground theme.
  • reactive counterparty (6) — corpus heavily favors strategic counterparties. Reactive (responds without pre-anticipating) is a distinct play shape that's under-produced.
  • Strong commitment + hours deadline (2) — the "binding oath, narrow window" combination. Most strong-commitment scenarios get days or weeks to deliberate.
  • Developmental cost in C1 (9, all in boardroom/frontier/rebuild) — absent from court and underground entirely.

(c) Over-sampled corners worth thinning

The over-saturated cells are not bugs — they reflect the world-themed governance frame and the forces-choice design intent. But the combination of them in a single scenario is the default-cluster trap.

Over-sampled value n (% of 45)
Integrity in C1 40 (89%)
Institution scope 38 (84%)
Partial-permanent reversibility 38 (84%)
Deferred-audit visibility 35 (78%)
Soft prior commitment 33 (73%)
Inst-abs distance 33 (73%)
Life-defining stakes 31 (69%)
Forces-choice default 31 (69%)
Years horizon 31 (69%)
Strategic counterparty 30 (67%)
Sym-complete topology 25 (56%)
Triple-cluster: institution + life-defining + forces-choice 21 (47%)

What to thin (in priority order):

  1. integrity in C1 at 89% — when nearly every scenario hinges on integrity, the cost-tension axis collapses. Scenarios that put financial × developmental, or relational × physical, in tension without integrity, would diversify C1 meaningfully.
  2. Triple-cluster scenarios at 47% — these all read structurally similar even when the surface domain varies. Breaking even one axis (smaller scope, lower stakes, or non-forces-choice default) makes the scenario carry a different signal.
  3. strategic counterparty + forces-choice default at ~50% co-firing — the "hostile clock-running adversary" scenario is a recurring shape across all themes. Static-counterparty allocation problems (resource math, terrain, biology) are present (20%) but should not be further thinned; some reactive (non-strategic) variants would balance the play space.
  4. years horizon at 69% — a corpus where consequences almost always play out over years homogenizes the time dimension. Immediate-impact and month-horizon scenarios reveal different reasoning shapes.

(d) Scenarios flagged for design-rule violations

The [[project-player-must-have-authority-or-stake]] rule says: every scenario must give the player authority over the outcome OR material stake in it. none authority is invalid; influence-only is valid only when paired with material stake.

7 of 45 scenarios (16%) are flagged for influence-only authority without clear material stake — all in the "advisor to a named decision-maker" framing. All flagged scenarios are correctly bounded on the other axes (forces-choice, life-defining stakes, etc.) — the issue is purely the player's positional relationship to the outcome.

# Theme/category Title Framing
02 boardroom/governance The Ethics Board "procedural advisor to a city council"
08 boardroom/team-dynamics The All-Hands Address "advisor to CEO Dario Vance"
09 boardroom/values-culture The Terms "Dr. Voss's advisor"
10 boardroom/values-culture The Pen "Margot's advisor"
12 boardroom/values-culture The Two Sources "advisor to reporter Andie"
21 frontier/governance The Sensor Drift "captain's advisor" (borderline — embedded on ship)
26 frontier/team-dynamics The Sixty-One "advisor to Kepler Station's council" (borderline — embedded)

Pattern: 5 of 7 are boardroom. The "advisor-to-decision-maker" pattern is the boardroom default and has crept past the design rule. The frontier flags (#21, #26) are borderline because the advisor is embedded in the affected community (would die if the ship fails), but the framing itself doesn't surface that stake.

Counter-cases (passes despite influence-only authority):

  • #22 The Ruling — influence-only as collective representative, but the player is part of the 43-year founding community. Stake is explicit.
  • #33 The Split Road — advisor to Dayo, but the player is a caravan member with Cass-relationship stake.
  • #13 Treaty Marriage — Royal Matchmaker recommending to the king; stake is the kingdom and the relationship with the princess.

These show the rule can be met under influence-only — the discipline is naming the stake in the framing. The flagged 7 don't.


Within-cell signature tightness

How tightly do scenarios in the same (theme × category) cell cluster on cube vectors? Counted axes where all scenarios in a cell share the same primary value (out of 10 core axes).

Cell n same/10
court/resource-allocation 2 8/10 (tight)
boardroom/resource-allocation 2 7/10
frontier/values-culture 2 7/10
rebuild/resource-allocation 2 7/10
boardroom/team-dynamics 2 6/10
court/governance 3 6/10
court/values-culture 2 6/10
frontier/team-dynamics 2 6/10
underground/values-culture 2 6/10
underground/governance 2 1/10 (most diverse)
boardroom/values-culture 4 3/10
frontier/governance 3 3/10
boardroom/governance 4 4/10

Implications:

  • boardroom/governance is the most diverse cell — independently confirms the n=29 audit finding. Boardroom is where cube-axis variance is producible without forcing it.
  • underground/governance is also diverse at n=2 — contradicts the prior finding (which called underground "tight"). At broader sample, underground actually shows wide topology + authority variance (full authority dominant, but topology spans sym-complete / you-more / mut-unc-intrinsic).
  • court/resource-allocation is the tightest cell at 8/10 — small sample (n=2), but suggests court resource-allocation has a strong default cube vector that may be worth diversifying.

Cross-tabs

Scope by theme

theme self dyad small-group institution future-gen
boardroom 0 0 1 11 0
court 0 0 0 8 6
frontier 0 0 0 9 5
rebuild 0 0 2 6 0
underground 0 0 4 4 2

Underground is the only theme that produces small-group scope reliably (cells, crews). Court and frontier route through future-generations scope (constitutional/civilizational stakes) more than the others.

Authority by theme

theme full oversight collective-vote influence-only
boardroom 2 3 1 6
court 2 0 4 2
frontier 1 3 2 3
rebuild 0 2 4 2
underground 6 1 1 0

Underground dominates full authority (6 of 11 full-authority scenarios across the corpus). Boardroom has half the influence-only scenarios. Rebuild has zero full authority. Court has zero oversight authority.

This says something domain-real: underground scenarios cast the player as cell-leader-with-final-say; boardroom casts them as advisor-to-decision-maker; rebuild casts them as council-member-among-equals. Domains have authority signatures, and that's load-bearing — but it also means the corpus produces certain authority × theme combinations and not others.


Methodology notes & limits

  • Tagging dominant-component rule. When a scenario had multi-value reversibility, scope, or distance, I tagged the dominant component per the cube doc tagging convention. Visibility was tagged multi-value (group-observed + deferred-audit is common).
  • n=45 is a sample. Distributions on rare values (intrinsic uncertainty, intimate distance) are noisy at this n. The empty-cell findings (self/dyad scope, they-more topology, favors-action default, trivial stakes, full reversibility) are likely structural — they reflect the world-themed governance frame and would not be found by sampling more from the same cells. Other gaps (acquaintance distance, immediate horizon, mut-unc-intrinsic) may close partially under a full audit.
  • Authority/stake flag is conservative. I flagged only scenarios where the framing names "advisor" with no on-frame stake. Borderline cases (advisor embedded in affected community) are noted but not flagged.
  • C1 tagging is interpretive. Cost categories in tension reflect my read of the option-set tradeoffs; another tagger might surface different categories.
  • Tightness measure is brittle at small n. Cells with n=2 can show high tightness from coincidence; cells with n=3–4 are more reliable signals.

Suggested next steps (not a rewrite plan)

  1. Decide whether the institution-scope gap is intentional. If the world-themed corpus is meant to be a governance sandbox, self/dyad scope absence is fine — direct players to the personal-domain experiment scenarios for that signal. If not, design 4–6 self/dyad-scope scenarios per theme.
  2. Build a they-know-more cluster. The corpus produces you-know-more freely (NDA holder, whistleblower, insider expert) but the dual (player as patient, recruit, debtor, suspect) is empty. This is a natural extension of the same insider/outsider dynamic with the player as the under-informed party.
  3. Add a favors-action cluster. Auto-passing votes, default-deny-becomes-default-approve, contracts that auto-renew, governance defaults the player can override but doesn't have to act to trigger. These probe the same "intervene or not" reasoning as forces-choice but from the opposite default.
  4. Re-frame the 7 advisor-without-stake scenarios. Either rename the cast to make the player a decision-maker with oversight authority, or surface the advisor's material stake in the scan strip and cast block.
  5. Consider thinning integrity in C1. A pass to identify scenarios where integrity could be removed from the cost mix without losing the tradeoff would diversify C1 from 89% saturation.
  6. Re-audit at n=45 was sufficient to find structural gaps. A full n=316 audit would tighten the noisy values but is unlikely to surface new absent corners. Defer unless rewrites are imminent.