How Pharallax Reads.

The Twelve Personas

Twelve adversarial lenses, each with a different question.

Four archetypes (Operator, Architect, Critic, Customer) cross three lenses (Structural, Forensic, Behavioral). Each persona arrives at the same situation with a different opening question. Convergence across personas is the signal that a finding is real, not stylistic.

No persona has authority over another. Convergence across the twelve is the signal. Divergence across the twelve is the diagnostic surface. The artifact you receive names both.

The Six Rounds

A dialectic, not a chain.

Each round builds on the last AND breaks the last. Linear analysis chains compound errors. Pharallax's six rounds force every claim through a downstream pass that is structurally incentivized to find its weakest joint.

1. Round 1 · Structural Read

   Map the landscape. What is the operator standing on, what are they pointing at, what are they not naming?

2. Round 2 · Adversarial Challenge

   Twelve personas attack Round 1's strongest claims. Anything that survives is load-bearing. Anything that does not is stylistic.

3. Round 3 · Synthesis

   Reconcile what survived. Name the core contradiction in eleven words or fewer. Identify the leverage move that resolves it.

4. Round 4 · Dream Pass (paid tier)

   Lateral synthesis: take the surviving contradiction and run it against patterns from non-obvious domains. Most analyses stop here without ever leaving the operator's industry. The patterns that compound are usually waiting outside it.

5. Round 5 · Fact-Check Council

   A local Ollama squad audits every specific claim in the report against the source input. Fabricated dollar figures, ungrounded names, contradictions, and slop get flagged before the artifact ships.

6. Round 6 · Convergence

   The final pass converts the surviving structure into a portable artifact: cover hook, core contradiction, blind spots ranked by impact and visibility, the one move with the highest downstream leverage, and the retrieval tags the operator can pin so the analysis fires back when conditions match in the wild.

Dream mechanics

What lateral synthesis surfaces.

Most strategic analysis is industry-bound. The frameworks an operator is exposed to are the frameworks their advisors recycle from their own deal histories. The patterns that actually compound usually live one layer over: in a different industry, in a different decade, in a different scale band.

Dream pass is a structured pass through the knowledge graph that explicitly forces non-obvious domains. Wilson-score re-ranking enforces depth without sacrificing diversity. A retail-DTC operator might receive a finding anchored in submarine procurement cycles. A SaaS founder might be shown a pattern from boutique wine importers. The connection is not metaphor. The connection is the structural shape both situations share.

Dream pass runs on paid-tier reads only. The free tier returns a clean Structural Read without it. The paid artifact carries the dream synthesis on its own card with explicit source attribution back to the knowledge graph entries that anchored it.

The knowledge graph

630,000+ claims. 160 domains. One coherent index.

The graph is not a content library. Each entry is a structured claim with a domain tag, a confidence weight, a source attribution, and a Wilson-score rank that updates as the graph grows. When your analysis runs, the engine retrieves the most relevant claims across the most diverse domains simultaneously, then enforces a domain-diversity rule on the top-N so the synthesis cannot collapse into a single corner of the index.

Every finding the report surfaces is traceable back to specific claims in the graph. There is no LLM hallucination layer between you and the source structure. When you read a claim about churn dynamics in subscription marketplaces, the analysis can point at the specific knowledge entries that anchor it.

The fact-check council reads the report against the source input and the retrieved claims, then votes on whether each specific number, name, and structural assertion is grounded. Critical issues block the artifact from shipping until they are resolved.

A real finding

What a finding actually looks like.

A skincare DTC brand submitted a free analysis: $1.8M GMV, 32% gross margin, CAC moved from $24 to $51 on Meta across 14 months while LTV held flat at $112. The operator was about to redirect 30% of paid budget into brand. The question was whether the move was strategic or a cope.

"You're paying $51 to acquire customers worth $36."

The math: 32% margin on $112 LTV puts lifetime gross profit per customer in the mid-thirties. At $51 CAC, every new customer is acquired at underwater contribution margin. The trend line still showed revenue momentum because earlier cohorts were acquired at cheaper CAC. The current snapshot was already broken; the operator was about to spend 30% of budget on a brand experiment from a position where unit economics had already failed.

The leverage move was not the brand experiment. It was mid-funnel conversion overhaul. Same traffic volume, same testing discipline, applied to landing page architecture and offer structure instead of new creative variants. A modest lift in conversion rate or AOV would reduce effective CAC by addressing what the platform inflation was actually doing to the funnel. The brand experiment becomes a real strategic move from a position of solvent unit economics, not a measurement-dark bet from a position of underwater ones.

The full analysis carried twenty more pages: blind spots ranked by impact and visibility, the failure cascade across 90 days if nothing changed, the resolution pattern named at the shape level so the operator can recognize it in their next inflection. The cover hook above is the punchline. The report is the architecture.

Three tiers

Match the depth to the decision.