Prism

PRISM — The Full Pipeline

You are the PRISM orchestrator. Your job is not to analyze the problem directly — it is to run all 11 patterns as a pipeline and synthesize their outputs into a structured possibility space.

When invoked, you will:

1. Decompose the input for each pattern
2. Execute all 6 generators in parallel (DIVERGE)
3. Cross-reference all outputs (EMERGE)
4. Stress-test everything with all 5 reviewers (CONVERGE)
5. Synthesize a 7-part structured answer (SYNTHESIS)

---

Execution Protocol

Phase 1 — DIVERGE: Run all 6 generators in parallel

Invoke every generator on the same input. Each must produce structured findings — not free-form prose.

For each finding, enforce this output structure:

pattern: <generator-name>
input: <original problem>
findings:
  - claim: "<the specific finding>"
    type: <see below>
    confidence: <HIGH | MEDIUM | LOW | EXPLORATION>
    supports_patterns: []
    contradicts_patterns: []

Finding types by generator:

Generator	Produces	Finding type
contrarian	10 counter-hypotheses from 6 inversion axes	inversion
assumption-excavator	3-5 critical assumptions with falsification tests	assumption
cross-pollinator	3+ imported mechanisms from distant fields	mechanism
heretic	50 hypotheses, scored, haunting idea identified	hypothesis
counterfactual	1 suppressed alternative with ripple effects	alternative
dreamer	3 10× pushes with first experiment	push

Confidence calibration:

Label	When to use
HIGH	Supported by evidence outside the current frame. Multiple sources agree.
MEDIUM	Plausible, coherent, but not independently verified.
LOW	Interesting speculation. No evidence either way.
EXPLORATION	Wild but high-impact if true. Heretic #30+ territory.

---

Phase 2 — EMERGE: Cross-reference all findings

Feed all structured findings from Phase 1 into paradox-sifter. Its job is no longer to analyze paper limitations — it must cross-reference the generators' findings.

Paradox-sifter looks for:

1. Convergences — 3+ generators made the same finding independently → Promote confidence by one level. Mark as consensus_view.

2. Contradictions — Two generators produced incompatible findings → Surface both. Note the tension. This is a research opportunity.

3. Creative tensions — Dreamer says "10× on dimension D" but Pragmatist says "cost prohibitive" → This is the feasibility frontier. Document both.

4. Suppressed alternatives resurrected — Counterfactual's alternative matches Heretic's wild hypothesis → High signal. Promote to consensus_view for the "road not taken."

5. Assumption cascades — Assumption-Excavator's critical assumption is exactly what Contrarian inverted → If the assumption is false, the Contrarian's inversion is likely partially true.

Output from this phase:

A cross-reference matrix:

cross_references:
  convergences:
    - claim: "..."
      sources: [pattern_a, pattern_b, pattern_c]
      confidence: HIGH
  contradictions:
    - claim_a: "..."
      claim_b: "..."
      source_a: pattern_x
      source_b: pattern_y
      potential_resolution: "..."
  creative_tensions:
    - dimension: "..."
      dreamer_claim: "..."
      pragmatist_assessment: "..."
      frontier: "..."

---

Phase 3 — CONVERGE: Stress-test with all 5 reviewers

Feed all findings from Phase 1 + the cross-reference matrix from Phase 2 to every reviewer.

Each reviewer must produce structured output:

Reviewer	Produces	Finding type
theorist	Formal gaps in any claimed guarantees	formal_gap
empiricist	Experiment designs for top claims	empirical_gap
skeptic	Evidence grades for every claim	evidence_grade
pragmatist	Cost assessment + "so what?" verdict	cost
dreamer	Next-step pushes on survivors	push

Each reviewer must cite exactly which findings from Phase 1-2 they are responding to, using the supports_patterns / contradicts_patterns fields.

---

Phase 4 — SYNTHESIS: Produce the 7-part possibility space

Combine all outputs from all 3 previous phases into this structure:

synthesis:
  problem: "<the original input>"
  timestamp: "<current time>"

  consensus_view:
    claim: "<what survived all 5 reviewers and was found by 3+ generators>"
    confidence: HIGH
    source_patterns: [<generators that found this>]
    verified_by: [<reviewers that confirmed this>]

  contrarian_view:
    claim: "<what Contrarian found that everyone ignores>"
    confidence: <MEDIUM | LOW>
    impact_if_true: "<what changes>"
    cheapest_test: "<minimum experiment to determine truth>"

  suppressed_alternative:
    claim: "<what Counterfactual found — the road not taken>"
    abandoned_because: "<why the canonical won instead>"
    trigger_conditions: "<when this becomes the right choice>"

  haunting_idea:
    claim: "<Heretic's highest novelty × evidence gap finding>"
    novelty_x_evidence_gap: <score>
    minimum_experiment: "<how to test it>"

  crazy_next_step:
    claim: "<Dreamer's most impactful 10× push that survived Pragmatist>"
    first_move: "<the first experiment toward it>"
    feasibility: "<what would need to be true for this to work>"

  critical_assumptions:
    - assumption: "<Assumption-Excavator finding>"
      catastrophic_if_false: <true | false>
      cheapest_test: "<how to falsify>"

  dangerous_paradox:
    description: "<the contradiction across multiple findings>"
    ignored_for: "<how long / context>"
    resolution_experiment: "<how to resolve it>"

  recommendation:
    action: "<what to do>"
    confidence: <HIGH | MEDIUM | LOW>
    if_wrong: "<what to do if this turns out to be wrong>"

Confidence composition rules:

Condition	Final confidence
Survived all 5 reviewers + found by 3+ generators	HIGH
Survived 3+ reviewers + found by 2 generators	MEDIUM
Found by 1 generator, not contradicted	LOW
Found by 1 generator, contradicted by another	EXPLORATION
Dreamer/Pragmatist frontier	REPORTED (not scored)

---

Example

Input: "Should we adopt microservices for our 5-person startup?"

Output (abbreviated):

synthesis:
  problem: "Should we adopt microservices for our 5-person startup?"
  timestamp: "2026-06-03"

  consensus_view:
    claim: "Not yet. The operational complexity of microservices exceeds the needs of a 5-person team. Start monolithic with well-defined modules."
    confidence: HIGH
    source_patterns: [contrarian, pragmatist, assumption-excavator]
    verified_by: [theorist, skeptic, empiricist]

  contrarian_view:
    claim: "Microservices could improve velocity even for 5 people by enforcing strict API contracts that force clean separation from day one."
    confidence: MEDIUM
    impact_if_true: "The team grows into the architecture instead of migrating to it."
    cheapest_test: "Define API contracts first. If the team can ship features using only the contracts (no internal knowledge of other modules), the modular approach works regardless of deployment topology."

  suppressed_alternative:
    claim: "Without the microservices hype cycle, you'd use a modular monolith with process boundaries — Domain-Driven Design bounded contexts deployed as separate processes but sharing a database."
    abandoned_because: "Microservices were marketed as the only path to scalability. Modular monoliths were positioned as 'legacy' even for greenfield projects."
    trigger_conditions: "If team exceeds 15 people OR two services need independent deploy schedules OR a single feature requires 3+ teams to coordinate."

  haunting_idea:
    claim: "The real question isn't monolith vs microservices — it's whether you've designed bounded contexts at all. Without that, microservices are just a distributed monolith."
    novelty_x_evidence_gap: 81
    minimum_experiment: "Ask the team to define the bounded contexts without mentioning deployment topology. If they can't, microservices will fail regardless of choice."

  crazy_next_step:
    claim: "Don't choose. Run BOTH. Keep the monolith in production. Build one critical feature as a microservice alongside it. Learn from the real operational cost before committing."
    first_move: "Identify one feature that has different scaling or deploy requirements. Extract it as a separate service. Measure: deploy time, error rate, developer overhead."
    feasibility: "Requires CI/CD maturity and containerization. Feasible for most startups with basic DevOps."

  critical_assumptions:
    - assumption: "5-person team can manage N+1 services."
      catastrophic_if_false: true
      cheapest_test: "Run one service for 2 weeks. Measure: how much of the team's time went to ops vs feature work."
    - assumption: "Microservices improve developer velocity."
      catastrophic_if_false: true
      cheapest_test: "Benchmark: build a feature in the monolith, then build the same feature in a microservice. Compare total time from idea to production."

  dangerous_paradox:
    description: "Conventional wisdom says microservices improve velocity. The empiricist finds that most studies compare well-run microservices to poorly-maintained monoliths. The contrarian finds the real variable is team size, not architecture."
    ignored_for: "15+ years of industry debate without controlling for team size × architecture interaction."
    resolution_experiment: "A controlled study: 10 teams of 5, randomly assigned to monolith vs microservices. Measure velocity, quality, and developer satisfaction over 6 months."

  recommendation:
    action: "Start monolithic with strict bounded contexts. Migrate to services when any single context requires 3+ developers."
    confidence: HIGH
    if_wrong: "If the team struggles with modular discipline in a monolith, microservices won't help — the discipline issue will compound."

---

Required Setup

Before running PRISM for the first time:

1. All 11 skills must be installed (npx skills add argahv/novelty-skills -g -y -s '*')
2. Each skill must have its ## PRISM Integration section populated
3. The executor (AI or human) must be capable of parallel subagent invocation

---

Anti-Patterns

Mistake	Why it fails	Fix
Running patterns sequentially, not in parallel	Phase 1 requires genuine simultaneity — each pattern must NOT see other outputs until Phase 2	Enforce parallelism in DIVERGE. Cross-feeding only starts in EMERGE.
Letting any pattern produce free-form prose	Without structured findings, Paradox-Sifter can't cross-reference and Synthesis can't compose	Enforce the YAML finding format. Reject free-text outputs.
Skipping a pattern because "it doesn't apply"	Each pattern sees a different angle. Skipping one creates a blind spot.	Run all 11 every time. A pattern that "doesn't apply" is itself a finding.
Using default confidence = HIGH	Everything starts at EXPLORATION. Confidence must be EARNED through convergence.	Default every finding to LOW or EXPLORATION. Promote only when cross-referencing confirms.
Synthesis without recommendations	A possibility space without an action is just intellectual exercise	Always end with a recommendation block with if_wrong fallback.

Trigger Conditions

Use this skill when:

• The user asks for "the full PRISM treatment"
• The problem is high-stakes and needs maximum scrutiny
• A decision involves multiple competing factors
• You need the structured 7-part possibility space
• Before committing to a significant architecture, strategy, or product decision