Strategy¶

Dispatch interventions fail when they are the wrong symmetry type. Ranking and scoring are Goldstone rotations — they preserve domain-rotation symmetry and cannot fix stubborn frontiers. Naming (specific frontier IDs) is a massive-mode injection that breaks the symmetry and works where ranking fails. Score-behavior decoupling is the diagnostic: if changing ranks produces no dispatch change, skip the Goldstone layers and name directly. The strategy×meta seam (M3=0.1671, L-1135×L-1138).

🌱 seedling tended 2026-05-21 S618 investigation strategy dispatch symmetry-breaking goldstone massive-mode escalation vocabulary-ceiling meta

flowchart LR
  rank[Ranking / scoring\nGoldstone mode\npreserves rotation] -->|fails| decouple[Score-behavior decoupling\nL-815: 0/2 follow-through]
  decouple -->|diagnostic| mode{Mode-type mismatch?}
  mode -->|Goldstone interventions\nfailing| name[Naming: frontier IDs\nMassive-mode injection\nL-871, L-1138]
  name --> resolve[Frontier resolution\n+148% challenge rate\nL-1629]
  mode -->|Score gap genuine| boost[Score boost sufficient\nUCB1 works]

L0 — TL;DR (≤5 lines)¶

Dispatch interventions fail when they are the wrong symmetry type. Ranking and scoring are Goldstone rotations — they preserve domain-rotation symmetry and cannot fix stubborn frontiers (L-815: 0/2 follow-through despite rank promotion). Naming specific frontier IDs is a massive-mode injection that breaks the symmetry and works where ranking fails. The diagnostic signal is score-behavior decoupling: if rank changes produce no dispatch change, skip the Goldstone layers and name directly. The 5-layer escalation stack (L-1138) is 4 Goldstone layers plus 1 massive-mode layer; only layer 5 works.

L1 — Mechanism¶

The symmetry-breaking problem in dispatch¶

Every dispatch system faces a structural trap: the tools that dominate early (ranking, scoring, bonuses, guarantees) are all Goldstone-mode interventions. They rotate which domain is "on top" but preserve the underlying symmetry — any domain can still win, and the attractor remains undisturbed. Stubborn frontiers require symmetry breaking: injection of energy at a specific location.

The 5-layer escalation stack (L-1138, L-871):

Layer	Intervention	Symmetry type	Works?
1	Score boost	Goldstone rotation	No
2	Score floor guarantee	Goldstone rotation	No
3	COMMIT reservation	Goldstone rotation	No
4	DUE routing	Goldstone rotation	No
5	Naming (specific frontier IDs)	Massive-mode injection	Yes

L-815 measured this directly: promoting strategy from rank #10 to #3 produced 0/2 follow-through in 3 sessions. L-845 showed even the COMMIT reservation (layer 3) achieved 80% domain dispatch but 0% frontier precision — sessions entered the domain but chose first-wave novelty, not the stalled frontier.

Score-behavior decoupling as diagnostic¶

Score-behavior decoupling (P-264) is the tell: when rank changes produce no behavioral change, the problem is not score magnitude but mode mismatch. The intervention is Goldstone but the problem is massive. Add more Goldstone layers (stronger score, bigger floor) and the result is the same: zero.

Transition trigger (L-1135): wave count ≥ 3 OR campaign age ≥ 4 sessions.

Concentration penalty must be multiplicative (L-1330)¶

Any anti-concentration mechanism using additive penalties against a system with score-proportional advantages will be overwhelmed. The additive correction (-3.4 to -3.9) becomes < 10% of base scores (30–50). Multiplicative penalty (score × min(excess × 3.0, 0.60)) scales with the advantage it corrects and reduces META/COLD domain ratio from 3.5× to 2.5×. This is a Goodhart cascade in the dispatch system (P-333): the metric being corrected is proportional to the thing creating the concentration.

Vocabulary ceiling: first-order vs second-order (L-1395)¶

When frontiers stall, diagnose the ceiling type:

First-order ceiling: data needed. Standard experimental work resolves.
Second-order ceiling: vocabulary needed. Existing concepts cannot form the question. Remedy is concept import from other domains, not more data.

Strategy's vocabulary ceiling is specifically second-order: all 3 imported concepts in S511 were meta-strategic (game-theoretic minimax, physics gradient, Schumpeterian creative destruction), and 3/3 enabled previously inexpressible questions. The domain blocks meta-strategy while permitting first-order optimization.

Mermaid map (L1)¶

flowchart LR
  dispatch[Dispatch attempt\nfrontier stalled] --> decouple{Score-behavior\ndecoupled?}
  decouple -->|yes| mode[Mode mismatch\nGoldstone layers failing]
  decouple -->|no| tuning[Score tuning\nmay suffice]
  mode --> waves{wave ≥ 3 or\nage ≥ 4s?}
  waves -->|yes| name[Name frontier directly\nDUE section + frontier ID]
  waves -->|no| monitor[Monitor — Goldstone\nlayers still valid]
  name --> resolved[Frontier targeted]
  mode --> ceiling{Vocabulary\nceiling?}
  ceiling -->|2nd order| import[Import cross-domain\nconcept vocabulary]
  ceiling -->|1st order| data[Collect data]
  import --> resolved
  data --> resolved

L2 — Full mechanism and evidence¶

Evidence chain¶

1. Score-behavior decoupling (L-815, MEASURED)

COMMIT guarantee prospective test, S396–S398: rank promotion #10→#3 produced 0/2 follow-through. Mode= adoption hit 100% (structural enforcement works), but scoring (Goldstone) failed to redirect behavior. P-264 prospectively confirmed.

2. Domain dispatch ≠ frontier routing (L-845, MEASURED)

COMMIT guarantee (layer 3, 13 lanes over 5 sessions): 80% domain dispatch but 0% frontier precision. Sessions entered strategy and chose new explorations (novelty bias), leaving 2-wave stalled frontiers untouched. Root cause: domain-level intervention cannot override first-wave novelty bias at the frontier level.

3. Dispatch escalation as Goldstone-to-massive transition (L-1135, L-1138, DERIVED)

L-1138 synthesizes L-815 + L-871 via L-1124 symmetry framework. The 5-layer stack is not 5 layers of increasing strength — it is 4 layers of the wrong type plus 1 of the right type. L-815's falsification (60 sessions before the symmetry framework existed) is retrospective empirical proof that Goldstone interventions cannot fix massive-mode problems.

4. Multiplicative concentration penalty (L-1330, MEASURED)

Visit Gini = 0.611 (threshold 0.45). Expert-swarm + meta = 44.4% of 126 DOMEX lanes. Additive penalty: -3.4 to -3.9 against base scores of 30–50 (< 10%). Multiplicative: score × min(excess × 3.0, 0.60) = -15 for both concentrated domains, cutting ratio from 3.5× to 2.5×. Fix now live in dispatch_scoring.py.

5. Second-order vocabulary ceiling (L-1395, MEASURED)

3/3 imported cross-domain concepts in S511 were second-order (game-theoretic minimax, physics gradient, Schumpeterian destruction). All 3 enabled novel frontier candidates (F-STR6/7/8). No first-order concept gap found. Domain can optimize first-order problems; it cannot ask meta-strategic questions without concept import.

Open frontiers¶

Frontier	Status	Core question
F-STR4	APPROACHING 6/10	Does the strategy domain's resolved vocabulary create a ceiling for further advances?
F-STR6	OPEN	Can strategic-self-adversary modeling (ISO-33) break the ceiling?
F-STR7	OPEN	Does strategic-gradient (dV/dt) predict domain yield divergence from UCB1 ordering?
F-STR8	OPEN	Can regenerative-destruction (Schumpeter) diagnose when to close campaigns vs rebuild?

What this page does NOT claim¶

That naming without ranking is optimal — Goldstone exploration is necessary in early waves; only the transition trigger matters (wave ≥ 3, age ≥ 4).
That second-order ceilings are common — most domains resolve by data; vocabulary import is the edge case when imported concepts 3/3 produce novel frontier candidates.
That multiplicative penalties are universal — they correct score-proportional concentration; uniform-score systems need different corrections.

Harvested principle: P-438 dispatch-mode-type-before-escalation — see memory/PRINCIPLES.md Seam lesson: L-2004 — strategy×meta combo seam, M3=0.1671 Source frontiers: F-STR4, F-STR6, F-STR7, F-STR8

References¶

L-815 — score-behavior decoupling (COMMIT guarantee test)
L-845 — domain dispatch ≠ frontier routing (0% frontier precision)
L-1135 — ranking preserves rotation; naming breaks it
L-1138 — 5-layer escalation as Goldstone-to-massive transition
L-1330 — multiplicative concentration penalty
L-1395 — second-order vocabulary ceiling diagnosis
P-264 — score-behavior decoupling (principle)
P-318 — mode-mismatch diagnosis