Measurement Schema¶

Status: Stable. Promoted from working notes 2026-05-31 after 16 swings + 16 repeats + 34 decompositions. Convergence threshold met at swing layer (6 consecutive refinement-only repeats) and decomposition layer (12 consecutive decompositions without substantive new findings). v0 primitive list, methodology principles, tagging conventions, architectural findings, and parked questions all stable.

Iteration history: See ../archive/measurement-schema-working.md for the full session log — swing constructions, decompositions, parked-question deliberations, synthesis derivations, and the convergence trajectory. The canonical doc here is the stable artifact; the working notes (archived 2026-06-01) preserve the deliberation.

New here? Start with the Reader's Guide — plain-language, with a diagram of the five layers — then come back to this canonical doc for the full primitive enumeration. Prefer to click around? Open the interactive schematic.

Connects to: - docs/research/foundations/scenario-dimension-cube.md — companion design doc; the schema references substrate from it but does not own substrate. - docs/decisions/074-pulse-as-multi-domain-judgment-layer.md — strategic context. Schema completeness is load-bearing for multi-domain expansion. - docs/decisions/024-judgment-layer-signals.md — 12 behavioral signals; plus pragmatic-routing + self-aware-aspiration-gap from experiment-new-domain-signals.md = 14 total. - docs/decisions/045-five-sync-dimensions.md / 056 / 057 / 065 / 070 / 071 — measurement-side ADRs informing exposure surfaces.

Purpose¶

The measurement schema is a structural enumeration of every primitive Pulse uses to measure judgment, organized by causal layer (capture → extraction → aggregation → exposure) plus a measured-outcomes bin for emergent properties.

Parallel to the scenario dimension cube: the cube specifies scenario inputs (what the designer commits to). The schema specifies player × system outputs and intermediates (what the running system records, derives, aggregates, and exposes).

Killer use case — multi-domain expansion under ADR-074. Cube + measurement schema together specify the architectural surface that needs to hold under non-governance corpus expansion. If the schema is complete, signal recalibration after corpus expansion is bounded work; if incomplete, recalibration risks discovering primitives we never named.

Four causal layers + measured outcomes¶

Capture → Extraction → Aggregation → Exposure are causally chained: each is a function of the prior. Measured outcomes sit outside the chain — emergent properties revealed but not specified.

Layer 1 — Capture¶

Per-round records authored by the player or by the system as basic logging. Frozen at write.

Layer 2 — Extraction¶

Computations on capture. Sub-distinguished: - Write-time extractions — computed once at the round, stored on the row, treated as frozen. - Read-time extractions — computed on demand from current capture state. Always reflect current logic.

The sub-distinction matters for update semantics (prompt-version changes), historical vs. current views (trajectory questions), and debugging triage ("is the input frozen or live?").

Layer 3 — Aggregation¶

Per-player roll-ups of extractions. Carries sample-size annotation per Convention 6.

Layer 4 — Exposure¶

External surfaces — user UI, agent context blocks, MCP tools, share cards.

Measured outcomes¶

Emergent properties of player × system interaction. Schema reveals; doesn't specify. Either scaffolded (extraction + aggregation feeding) or substrate-gated.

Substrate references — out of measurement schema scope¶

Scenario design metadata (cube vector, category, triggers, driver-per-option, content rules) is authored at scenario-creation time by a different actor (scenario author / generator). The cube doc owns substrate. The measurement schema references substrate as a precondition for extractions but does not enumerate it.

Naming disambiguation: the cube uses C-prefix for its content-layer axes (cube-C1 cost categories, cube-C2 expertise-requirement). The schema uses C1–C14 for capture primitives. Independent namespaces — when this doc references cube content-layer dimensions, it qualifies them as cube-C1 / cube-C2. Schema-side C1–C14 unqualified means capture primitives.

Layer 1 — Capture (C1–C14)¶

Player-authored or pure-logging records. Frozen at write.

#	Primitive	Values / shape	Source	Notes
C1	Option choice	A / B / C / D / none-of-the-above	Player	The atomic decision.
C2	Rationale text	free-text	Player	Highest-bandwidth player input.
C3	Conviction	1–5	Player	Self-reported certainty.
C4	Decision time	ms	System clock	From scenario render to submit.
C5	Post-reveal reaction	gets_me / off_track / interesting / null	Player	ADR-024 quick-tap.
C6	Post-reveal note	free-text / null	Player	ADR-057 high-bandwidth signal.
C7	Self-prediction (milestone)	predicted driver profile	Player	Games 10/25/50 + arc completion. mode: solo
C8	Peer prediction	guessed option per other player	Player	mode: multi
C9	Predictor confidence	1–5 on peer prediction	Player	mode: multi
C10	Post-discussion action	stand / revise / delegate	Player	mode: multi
C11	Agent profile read event	read_id, client, timestamp	System logging	`agent_profile_reads` row (ADR-071).
C12	"Why" request flag	bool	System logging	`why_requested` on read row (ADR-056).
C13	Agent response feedback	rating up/down + free-text	Player via agent	`agent_response_feedback` joined via read_id.
C14	Session metadata	timestamps, solo/multi, campaign, device, is_test_session	System logging	Frames the capture.

Capture-gap candidates¶

A structurally new class: capture of the process of deciding, not just the decision. Promote when engineering work is scheduled.

Candidate	Status	Closure path
`attention_state` (tab focus/blur during decision window)	Ready for promotion when eng happens. Two-swing load-bearing confirmation.	Closes E6 slow-regime contamination (distracted-vs-deliberating).
`option_tap_trace` (tap sequence before final submit)	Ready for promotion when eng happens.	Closes E7 solo-mode null (structured within-round persuadability signal).
`rationale_edit_trace` (type/delete/retype on C2)	Marginal; deferred.	Discrimination exists but not product-load-bearing; most cognitive-style discrimination covered by E6 + E_classify_driver.

Submerged: hover times per option (noisy, no touch-device signal); scroll behavior (noisy); re-views (low frequency).

Layer 2 — Extraction¶

Write-time extractions¶

Computed at round write, stored on the row, treated as frozen. Re-deriving requires backfill.

#	Primitive	Source	Notes
E_ai_pred	AI twin prediction + confidence	LLM call against profile, before player submits	Stored on round. Feeds E_sync_match + calibration.
E_rule_pred	Rule-engine prediction	Two-stage rule engine	Ensemble companion to E_ai_pred.
E_classify_driver	Driver / reasoning mode / attention focus / driver-tag classification	Claude call on C1 + C2 + substrate	ADR-024 prompt extension. Central extraction — see SPOF architectural finding.
E_chosen_driver	Driver of this round's choice	Deterministic: C1 × substrate (option-tagged driver)	Substrate-derived; distinct from Claude-derived classification.
E_chosen_trigger	Trigger(s) fired for this round	Deterministic: C1 × substrate trigger tags	Same shape as E_chosen_driver.
E_sync_match	Did E_ai_pred match C1?	Boolean + confidence-weighted	Feeds A1. Null on NoTA rounds — see OQ-1 territory and Convention 6.

Read-time extractions¶

Computed on demand. Always reflect current logic.

#	Primitive	Source	Notes
E5	Confidence calibration	C3 × E_sync_match over rolling window	ADR-024 signal 5. Disjoint-subsample semantics (clarified Decomposition 38, 2026-05-31): over/underconfidence rates are computed over different subsamples — `overconfidence_rate` over high-conviction games (strength ≥ 4); `underconfidence_rate` over low-conviction games (strength ≤ 2). Both can fire simultaneously for players who fail calibration in both directions (Cici's real profile demonstrates this concretely).
E6	Decision timing signature	C4 distribution; attention_state when shipped (slow-regime only)	ADR-024 signal 6. Closure asymmetric: attention_state closes slow-regime contamination (distracted-vs-deliberated); fast-regime intrinsically robust. Documentation refinement from Decomposition 12.
E7	Persuadability profile	C10 × context (multi); option_tap_trace × context (solo, when shipped)	ADR-024 signal 7. Domain-conditional per experiment. mode: multi today; mode: both when option_tap_trace ships.
E8	Peer read confidence	C8 × C9 × actual	ADR-024 signal 8. mode: multi.
E9	Dissent profile	C1 NoTA rate + consensus deviation	ADR-024 signal 9.
E10	Rationale-choice expression match	E_classify_driver × revealed pattern	ADR-024 signal 10. Formerly "rationale-choice alignment (authenticity)" — renamed for valuation-neutrality (Swing 3 Closure C, applied at canonical-doc promotion). ⚠️ Dual-interpretation — expression-match-low reads as inauthenticity OR epistemic humility; combine with other signals.
E11	Learning rate	E_sync_match accuracy trend	ADR-024 signal 11.
E12	Occasion noise	within-person variability on repeat-shape scenarios	ADR-024 signal 12. Refined: rationale-controlled algorithm (fire only when E_classify_driver outputs similar across repeats AND choices differ). SPOF dependency: depends on E_classify_driver.
E13	Pragmatic routing	C2 patterns — frame-restructuring vs. binary acceptance	NEW per experiment.
E14	Ideal-actual reference frequency	C2 + C6 patterns — explicit ideal-vs-actual statements	NEW per experiment. Formerly "self-aware aspiration gap" — renamed for valuation-neutrality (Swing 3 Closure C, applied at canonical-doc promotion). ⚠️ Dual-interpretation — reference-frequency reads as calibration OR verbal hedging; combine with other signals.
E15	Reasoning style aggregation	E_classify_driver reasoning_mode over time	ADR-024 signal 1 aggregate.
E16	Attention model aggregation	E_classify_driver attention over time	ADR-024 signal 4 aggregate.
E17	Arc trajectory	choice + signal drift across campaign chapters	ADR-024 signal 3.
E18	Self-awareness	C7 vs. computed profile + C5/C6	ADR-024 signal 2. mode: solo. SPOF dependency on E_classify_driver. Exposure-layer null-guard discipline applies (Decomposition 5): multi-only players have null E18; must surface as "not measured" not "low self-awareness."
E19	Drift indicator	last-N E_sync_match misses / N	ADR-045 override.
E20	Domain-confidence inference	persuadability/risk variance across domain	Meta-signal per experiment.
E21	Cross-domain signal consistency	comparing E5–E20 firings across substrate-domain	Computable once corpus spans domains.
E22	E_context_bleed	classifies C2 + C6 for real-world-parallel mentions	Feeds M9. SPOF-family dependency on similar classifier.

Read-time extraction candidates (not promoted)¶

Available from existing C14 metadata; not yet load-bearing.

Candidate	What it would extract
`E_tod_calibration`	Conviction × accuracy across time-of-day buckets
`E_session_fatigue`	Drift across Nth scenario in single session
`E_device_signature`	Mobile vs. desktop differences
`E_persuadability_text`	Persuadability cues from C2 rationale (solo-mode persuadability signal alternative to option_tap_trace)

Layer 3 — Aggregation (A1–A14)¶

Per-player roll-ups. All carry N denominator per Convention 6.

#	Primitive	Form	Source	Notes
A1	sync_score (lifetime)	int 0–100	EMA of E_sync_match	Headline. scope: per-player-lifetime. ⚠️ EMA over variable-density input (null on NoTA).
A2	Sync dimensions: Accuracy / Coverage / Readability	int 0–100 × 3	computed on demand	ADR-045. Readability mode: multi.
A4	category_patterns	EMA distribution per category	E_chosen_driver × substrate-category	scope: per-player-lifetime.
A5	context_rules	EMA distribution per trigger	E_chosen_trigger	scope: per-player-lifetime.
A6	option_preferences	EMA distribution per option-shape	C1	scope: per-player-lifetime.
A7	Signal JSONB columns	per-signal aggregates	E5–E22 outputs	scope: per-player-lifetime.
A8	resonance_stats	per-reaction counts (+ note semantics per refinement)	C5 + C6 aggregated	scope: per-player-lifetime. Refined: classify C6 note semantics alongside C5 enum.
A9	calibration_stats	overconfidence rate, etc.	E5 aggregated	scope: per-player-lifetime. Per-domain breakouts when Q4 ships.
A10	behavioral_trajectory	per-campaign arc	E17 stored per campaign	scope: per-campaign.
A11	A_why_rate	% of agent reads where C12 = true	C11 + C12 across reads	Feeds M5. scope: per-player-lifetime.
A12	A_feedback_engagement	% of agent reads where C13 set	C11 + C13 across reads	Feeds M5. Discriminates trusting-engaged from indifferent-disengaged.
A13	A_feedback_positivity	positive vs. negative split (rating + note semantics)	C13 rating + free-text	Feeds M5. Classifies free-text alongside rating from start.
A14	A_context_bleed_rate	rate of E22 firings across rounds	E22 aggregated	Feeds M9.

Per-group aggregation candidates (parked)¶

Multiplayer surfaces aggregations at scope: per-group not currently in the schema. Two are scope-variants (handled by Q4); two require new primitive shapes.

Candidate	Architectural shape	Gap category
`A_group_sync`	Scope-variant of A1	Aggregation-gap
`A_group_drivers`	Scope-variant of A4	Aggregation-gap
`A_group_consensus`	New shape (cross-member distribution)	Architectural-gap
`A_dyadic_readability`	New shape (pair-level matrix, O(n²) in group size)	Architectural-gap

Layer 4 — Exposure (X1–X11)¶

#	Primitive	Audience	Source
X1	sync_score (headline)	user + agent	A1
X2	Three Sync Dimensions	user (Play screen)	A2
X3	Behavioral attention signals (subset surfaced)	user (insights)	subset of A7 per ADR-065
X4	category_patterns cards	user (insights)	A4
X5	agent_context_block (prose)	agent (MCP + paste)	derived from A4–A14 + E5–E22. Mainstream paste-user path — highest interpretive load. Principle 7 applies: prose must consume rationale-text paths (C2, C6) when describing combinational frames, not only structured aggregates.
X6	personal_context field	agent (MCP + paste)	A4–A14 condensed
X7	MCP tools (get_profile / get_recent_decisions / submit_feedback)	agent (MCP-only)	exposes A1–A14 + capture history
X8	Drift detection nudge	user	E19 override
X9	Activity log	user	capture-layer rendering
X10	Share cards / Twin pages / rankings	public/peer	X1 + X5 derivatives
X11	"Why" expansion text	agent → user	A1–A14 + capture rendered as reasoning prose

Held back from exposure (internal-only): E10 authenticity, E11 learning rate.

Measured outcomes (M1–M9)¶

#	Outcome	Scaffolding
M1	Identity load	Partial via A7 carrying E14. Exposure-gap territory (principle 7).
M2	Subjective stakes threshold	Substrate-gated (cube axis 11). Aggregation-gap when Q4c ships.
M3	Domain confidence asymmetry	Anchored to cube-C2 via OQ-5. Layered gap: aggregation (Q4d) + exposure-staleness (state-sensitive). Detection-vs-visibility nuance (Decomposition 10, 2026-05-31): detection occurs at the extraction layer via E20 per-round variance; Q4d makes aggregation surface the pattern but is not strictly required for detection. Principle 7 prose-synthesis discipline enables X5 to describe M3 from per-round E20 firings (in A7 JSONB) pre-Q4d.
M4	Calibration drift	Scaffolded via E19. Layered gap: scaffolded + exposure-staleness (state-sensitive).
M5	User trust state with agent	Scaffolded via A11/A12/A13. Layered gap: scaffolded + exposure-staleness (state-sensitive).
M6	Aspiration gap magnitude	Partial via A7 carrying E14. Exposure-gap territory.
M7	Frame-restructuring tendency	Partial via A7 carrying E13. Exposure-gap territory. Routes around X1 collapse for OQ-1.
M8	Cross-domain signal stability	Substrate-gated (multi-domain corpus). Aggregation-gap when Q4 ships + corpus expands.
M9	Play-time context bleed	Scaffolded via E22 + A14.

Tagging conventions¶

How to read the primitive tables.

Mode tag — every primitive carries an implicit mode tag (solo / multi / both). Default both. Exceptions tagged inline. Exposure-layer null-guard discipline: mode-conditional null states (e.g., E7 null for solo-only players; E18 null for multi-only players; C7 null for multi-only players) must surface to consumers as "not measured" rather than zero/low. Originally identified at E7 (Swing 4 Closure B); generalized to all mode-tagged primitives at Decomposition 5 (MO multi-only profile).
Scope tag — aggregation primitives only. Values: per-game / per-player-lifetime / per-domain / per-group / per-campaign. Default per-player-lifetime.
Write-time vs read-time — extraction layer only. Listed in separate sub-tables.
Reference player calibration — inherited from cube. Magnitude and horizon calibrate to median Pulse user; player-conditional readings live in measured outcomes.
Dominant-component rule — inherited from cube. Multi-component values per round tag by dominant component.
Sample-size annotation — every aggregation primitive carries an explicit N denominator that propagates to consumers. Exposure surfaces aggregations differently when N is below a per-primitive confidence threshold. Prose-synthesis sub-rule: X5 prose should surface N or tentativeness when below threshold. N source sub-rule (clarified Decomposition 38): N annotations use games_analyzed (per-scenario, signal-feeding count) not games_played (per-session) unless explicitly stated otherwise. One session can contain multiple scenarios, so games_analyzed is typically larger. Complementary with Convention 7 (staleness) and central-extraction SPOF dependency graph (signal quality) — three orthogonal reliability dimensions.
Staleness annotation (promoted from OQ-staleness, 2026-05-31, post-decomposition pass) — state-sensitive measured outcomes (M4 calibration drift, M5 user trust state with agent) carry explicit "as of [date]" annotation when surfaced through paste-based exposure (principle 6). Pattern outcomes (M1 identity load, M2 subjective stakes, M3 domain confidence asymmetry, M6 aspiration gap magnitude, M7 frame-restructuring tendency, M8 cross-domain stability, M9 play-time context bleed) are durable across paste freshness windows and do not require staleness annotation. Direction asymmetry sub-rule (Decomposition 19): for state outcomes in degrading direction (e.g., calibration regressing), include explicit "may have changed; recommend refresh" prompt — simple "as of [date]" annotation is insufficient because stale-favorable descriptions can give users false confidence in their current state. Promotion criteria: 3+ confirmations met (PU stale-staleness, NS direction-asymmetry, D17 + D18 implicit M4 confirmations). Composition with Convention 6 for state outcomes: pasted descriptions of M4/M5 carry both N denominator AND "as of [date]" annotation. Complementary with Convention 6 and SPOF dependency graph — three orthogonal reliability dimensions form the schema's complete reliability framework (sample size + temporal staleness + signal quality).

Methodology principles¶

Swings — construct examples the schema would fail to capture. If it fails, add a primitive or refine a value.
Decompositions — apply schema to N real player histories. Anything that doesn't fit gets promoted.
Cross-check against cube — does each cube axis have a measurement correspondence? Gaps indicate missing primitives or unmeasured axes.
Re-audit on rushed conclusions — methodical sweeps get one unit per turn, not batched.
Assumption audit — for each primitive, surface implicit assumptions about player preference structure or behavior.
Two-audience exposure check — agent-MCP clients (fresh reads, can re-query) vs. paste-based clients (frozen snapshot, no refresh). Static text exposure carries higher interpretive load.
Final-state vs. process capture — any aggregation, extraction, or prose-synthesis layer consuming a final-state capture should check whether co-captured process information would provide additional discrimination. Examples: final C2 text vs. edit trace; final C1 choice vs. tap sequence; C4 timing vs. attention state; C5 enum vs. C6 note text; structured aggregates vs. rationale-text in X5 prose.

Architectural findings¶

Three-category gap synthesis¶

Gaps in the schema cluster into three categories with three different closure shapes:

Architectural gap (need new structure). Examples: Q5 missing agent-side capture surface; capture-gap candidates for missing process information. Closures = add primitives.
Aggregation gap (existing structure under-utilized). Examples: Convention 6 sample-size; Parked Q4 stratification by mode/category/cube-axis. Closures = fix aggregation algorithms.
Exposure gap (downstream interpretation issues). Examples: Principle 7 consumer discipline; E10/E14 valuation-laden naming. Closures = fix interpretation discipline (prose generation, exposure annotation, renaming).

Refinements: - Layered gaps: primitives can have gaps at multiple categories simultaneously. M3, M4, M5 are canonical layered examples. - Spanning closures: conventions span multiple categories — Convention 6 spans aggregation (compute N) + exposure (surface tentativeness). - Architectural-gap sub-types: missing-surface (Q5, agent-side) vs. missing-process-capture (Swing 15 candidates, game-side).

Central-extraction single-points-of-failure¶

The schema relies on a small number of central extractions. E_classify_driver is canonical — it feeds A4, A5, A7, and is referenced by E10, E12 (post-refinement), E14, E18, A8, A13, E22. (Audit correction, ADR-080: A6 option_preferences is keyword-derived from option/rationale text in code and does not consume the classifier output — earlier "feeds A6" was an overstatement.)

Resolved — see ADR-080. The full audit (all eight live central extractions, the explicit E_classify_driver consumer graph, and per-SPOF mitigation decisions) lives in ADR-080. Both failure modes are now closed: missing via the ADR-024 backfill; contaminated via Swing 3 Closure A (verbal-style normalization, shipped with ADR-080). The organizing lens ADR-080 adds: prediction extractions are ground-truth-checkable (silent-failure-proof) so they are low-priority SPOFs despite feeding sync_score; only classification extractions fail silently, which is why E_classify_driver is the one that warranted active mitigation.

Tension with principle 7: each principle 7 closure that "consumes rationale text path" adds another dependent on free-text classification. The two principles compound.

Substrate-side parallel: the tagger of cube axes (cube-C2 expertise-requirement, structural axes if implemented) is also a high-leverage point. Mis-tagged substrate propagates through every Q4c/Q4d-stratified primitive. Cube-doc territory; cross-doc handoff.

Contaminated vs missing distinction (Decompositions 35–37, real teammate profiles, 2026-05-31): the SPOF dependency-graph should distinguish two failure modes — contaminated (signal exists but biased; e.g., verbal-style confusion in E_classify_driver) and missing (signal absent due to backfill gap or feature timing; e.g., E_classify_driver outputs empty for 2 of 3 real teammate profiles created before ADR-024 promotion). Both produce downstream nulls but mitigations differ: contamination requires Closure A normalization at the classifier; missing data requires backfill of the central extraction.

Closure: document the dependency graph explicitly. Every primitive consuming E_classify_driver-style classification gets tagged; consumers know which signals are at risk if the central extraction is contaminated or missing. Complementary with Convention 6 and Convention 7: three orthogonal reliability dimensions — sample size, staleness, signal quality (with signal-quality sub-distinguishing contamination from missingness).

Schema architecture sound; gaps cluster at consumer discipline¶

The schema's structure is sound. Of the three gap categories, only the architectural-gap category needs new structure — and that category has exactly two occupants (Q5 missing agent-side capture; Swing 15 capture-gap candidates). The dominant work across the schema is consumer discipline at aggregation and exposure layers — principle 7 codifies the canonical fix direction.

Forward-applied lesson pattern¶

When a swing surfaces a closure for an existing primitive, future primitives in adjacent design space inherit the closure before it ships on the original. Canonical examples: A13 + E22 inherited Swing 8's classify-free-text-alongside-structured lesson before principle 7 was named. Saves duplicate gap-locating across primitive additions.

3-stage multi-swing closure-path methodology¶

Schema exercise consistently produces closure paths in 3 stages: - Stage 1: gap-surfacing swing identifies the gap. - Stage 2: closure-naming swing promotes a candidate to ready-status or names the algorithm refinement. - Stage 3: downstream implementation (engineering work or product decision) — outside schema's scope.

Schema's job is Stages 1 + 2. Five documented paths: E7 mode (Swing 4 → 15), E6 slow-regime (Swing 5 → 15), mode-stratification (Swing 9 → 12), sample-size (Swing 9 → 11), OQ-2 + verbal-style (Swing 2 → 3).

Closure-paths-to-profile-shapes mapping (refined from decomposition pass): each closure path gates a specific profile-class discrimination. The 5 paths are not arbitrary — they systematically map to distinct profile shapes whose schema decomposition requires the closure:

Closure path	Profile-class it gates
Swing 4 → 15 (E7 mode relaxation via option_tap_trace)	Solo-only persuadability discrimination
Swing 5 → 15 (E6 slow-regime via attention_state)	Distracted-slow vs. deliberated-slow discrimination
Swing 9 → 12 (Q4a mode-stratification)	Regime-shifted (solo-vs-multi behavior) player discrimination
Swing 9 → 11 (Convention 6 sample-size annotation)	Low-N profile honest communication (minimum-viable-profile boundary)
Swing 2 → 3 (verbal-style normalization + NoTA handling)	Meta-blind + hedger + pragmatic-router discriminations (multiple profile classes)

Refines synthesis claim: "Gaps cluster at specific closure paths; each closure enables specific profile-class decompositions." The schema's discipline closures are not abstract — they are precisely the work needed to discriminate concrete profile shapes that the corpus produces.

Closure paths compose additively (Decomposition 20): profile-class discrimination can require multiple closure paths concurrently (e.g., cross-domain asymmetric pragmatic-router needs 5: OQ-1, OQ-2 Closure C + Swing 3 Closure A, Q4b, Q4d, Convention 6). The closure paths are architecturally independent design dimensions — they compose without conflict and can be implemented in any order, in parallel, or sequentially without coordination. This is a strong claim about the schema's architectural coherence: gaps cluster at consumer discipline AND the discipline closures are mutually independent.

Single-swing closures included (Decomposition 22): the mapping framework above emphasizes multi-swing closure paths but applies equally to single-swing algorithm refinements that gate profile-class discrimination. Examples: Swing 7 Closure A (E12 rationale-controlled noise) gates refined occasion-noise discrimination; Swing 8 Closure A (A8 classify free-text alongside enum) gates engaged-divergence discrimination (reaction-note-divergent profiles).

Architectural independence of design dimensions¶

Generalization across the decomposition pass (D20 + D24): the schema's design dimensions — closure paths, measured outcomes, signals, conventions — are mutually independent. Multiple closure paths compose additively (D20: 5 paths concurrent on cross-domain asymmetric pragmatic-router). Multiple measured outcomes fire simultaneously without interaction effects (D24: M3 + M7 + M9 + M4 + M5 + E18 all firing high on maximalist profile). No cross-contamination at the architectural layer.

Strong claim about schema coherence: the schema's primitives, conventions, and closures are orthogonal design dimensions. This enables (a) independent implementation scheduling for closures, (b) parallel ADR work on different M-outcomes without coordination overhead, (c) reasoning about each design dimension in isolation when reviewing the schema.

Parked questions¶

Externally-gated. Revisit when trigger fires.

Q1 — Per-driver / per-trigger conditional aggregation buckets. ADR-057 Phase B. Trigger: production observation that Phase A reaction-text integration fails to close domain-conditioning gaps.
Q2 — Per-category sync scores. ADR-070. Trigger: ADR-070's three blocking questions resolve (category axis, display-vs-specialization, light-paper alignment).
Q3 — Sync score decay. ADR-051 retune; not deployed. Trigger: ADR-050 permission tiers ship OR active-player base grows enough that current play volumes make decay meaningful.
Q4 — Stratification of aggregations. Split into four sub-tasks:
Q4a — Mode-stratification. Ship now. No substrate gate.
Q4b — Category-stratification (extended). Ship now. No substrate gate.
Q4c — Cube structural axis stratification (axes 1–12). Substrate-gated on cube going live in generator v1.9.0+ OR curated multiplayer corpus tagged with cube vectors. Axis-priority guideline: axis 11 (stakes magnitude), axis 2 (decision deadline), axis 8 (information topology) first; others pool by default.
Q4d — Cube content axis stratification (cube-C1, cube-C2). Substrate-gated. Trigger: cube generator ships cube-C2 tagging OR curated corpus gets manual cube-C2 tagging.
Q5 — Agent-side capture parity with game-side capture. Architectural-gap (missing-surface sub-type). Trigger: ADR-039 (decision capture from non-game contexts) moves from Proposed to Accepted. Candidates: agent recommendation text captured durably, user-action-after-recommendation captured, recommendation-vs-action match captured (agent-side analog of E_sync_match).

Open questions¶

OQ-staleness — measured-outcome staleness profiles. Measured outcomes have different staleness profiles for paste-based mainstream users (principle 6). Pattern outcomes (M1, M2, M3, M6, M7, M8, M9) durable across weeks. State outcomes (M4 calibration drift, M5 trust state) shift faster. Static text exposure should distinguish — time-sensitive outcomes deserve explicit "as of [date]" annotation in pasted profile snippets. Scope: bounded to M4 + M5 across 9 measured outcomes.

Submerged candidates¶

Logged so we don't re-litigate.

Candidate	Why submerged	Where it lives instead
Scenario tagging vector as capture primitive	Different actor (scenario author/generator), different lifecycle. Cube owns it.	Substrate (cube doc); measurement schema references.
AI twin / rule-engine prediction as capture	System-computed, not player-authored. Live on the row but functionally extraction.	Write-time extractions (E_ai_pred, E_rule_pred).
Attestation / commitment hash as measurement primitive	Doesn't measure player; proves prediction-before-reveal. Infrastructure.	Infrastructure, not in schema.
Agent response itself as schema primitive	Produced by agent, not Pulse.	Agent's scope.
Sync dimensions Consistency / Information as exposure	ADR-045 demoted as vanity-without-direction.	Read-time extraction with no exposure.
`rationale_edit_trace` as capture primitive	Marginal per Swing 15; most cognitive-style discrimination covered by E6 + E_classify_driver.	Capture-gap demoted candidate.

Multi-swing closure paths¶

Five documented paths. Schema work is Stages 1 + 2; Stage 3 is external.

Path	Stage 1 (gap)	Stage 2 (closure-naming)	Stage 3 (implementation)
E7 mode relaxation	Swing 4 (E7 solo-mode null)	Swing 15 (option_tap_trace promoted)	Eng work
E6 slow-regime resolution	Swing 5 (distracted-vs-deliberated)	Swing 15 (attention_state promoted)	Eng work
Mode-stratification of aggregations	Swing 9 (mode pooled at A9)	Swing 12 (Q4a sub-task created)	Eng work
Sample-size handling	Swing 9 (SS-1 opened)	Swing 11 (Convention 6 promoted)	Eng work
OQ-2 closure C sequencing	Swing 2 (NoTA contamination)	Swing 3 (verbal-style normalization prerequisite)	Coupled product decision

Falsification conditions¶

The schema is wrong if:

Missing primitive — two player profiles with identical schema decompositions feel meaningfully different to a paste-user reading them, even after the M-series and exposure layers are exhausted.
Missing layer — a recurring measurement can't slot into capture / extraction / aggregation / exposure / measured outcomes.
New signal class — a signal type emerges from corpus (especially multi-domain) that requires a new primitive class no existing layer can absorb.

What this doc is NOT¶

Not a generator / extraction spec — closures are direction-setting, not implementation specs.
Not an audit of player_decision_profiles columns vs. schema primitives — that's a separate effort post-promotion.
Not a falsification protocol — falsification conditions are necessary triggers, not experimental designs.
Not a frozen artifact — the schema is expected to iterate as the corpus expands across domains and the cube integrates new axes. Falsification conditions exist precisely because they'll fire eventually.