AI Transformation Assessment Engine Thinking Flow

Create machine-readable source material while preserving page, row, image, and document boundaries for later verification.

Run deterministic checks and a bounded pre-flight review across the registry, then block or redact high-risk material before model calls.

Carry selected visual pages and screenshots alongside text so architecture diagrams, service maps, dashboards, and operating models are not lost.

Use uploaded sources to establish present-state claims. Use criteria, anti-patterns, tactics, taxonomies, and remote PDFs only as controlled analytical context.

Build provenance before interpretation so traceability is structural rather than added after the report is written.

Use deterministic term-based signals to rank evidence and reduce irrelevant context before the forensic model receives it.

Prioritize high and medium relevance, retain gap and contradiction signals, and stop near the target before context becomes noisy.

Use different OpenAI and Anthropic profiles according to task shape and consequence, with ordered fallbacks rather than free model choice.

Verify the quote, surrounding context, score calibration, and packet coverage independently from the forensic auditor.

Give negative findings their own evidence semantics so a zero anti-pattern score cannot automatically become positive readiness.

Prioritize the largest score deltas and most consequential weak items, rescan at most three criteria per batch, then run verification again.

Replace unsupported scores, preserve the adjustment trail, and merge all batch verification results into the governed audit record.

Normalize 25 maturity and 25 anti-pattern scores; subtract 50% of anti-pattern burden; add a 10% tested-clearance contribution only when anti-pattern coverage is at least 60%; then apply the evidence cap.

Below the evidence floor: Insufficient evidence. Below 33 readiness: Emerging. Between 33 and 65: Structured, or Scaling with friction when burden exceeds 50. At 66 or above: Adaptive / Value-creating.

Below 30% density, readiness is capped by available evidence and strategy is blocked. Between 30% and 60%, readiness cannot exceed 60 until stronger current-state evidence is supplied.

HIGH requires at least 70% evidence density, 95% delivery integrity, and no more than 10 silent areas. LOW is triggered below 30% density, below 70% integrity, or above 18 silent areas. Everything between is MEDIUM.

Build a factual assessment layer before causal interpretation or planning language is introduced.

Connect verified patterns across operating model, architecture, strategy, data, and service architecture without changing locked metrics or introducing unproven current-state claims.

Translate the same factual diagnosis into transformation language: operating rhythm, demand routing, learning, ownership, and sequencing.

Use technical and architectural language while preserving the same evidence boundaries and avoiding stack-specific assumptions.

HIGH can be directive. MEDIUM keeps phase-level confidence and assumptions. LOW returns evidence-backed findings, candidate themes, missing evidence, and a validation plan—with no directive roadmap.

Separate diagnosis from prognosis: a valid diagnosis can coexist with a NO_GO planning decision when the source pack cannot carry implementation advice.

Phase 1 Emerging — Foundation establishes thesis, owners, demand routing, impact statement, and validation evidence. Phase 2 Structured — Integration proves a vertical slice through service blueprint, data forensics, lifecycle, safety, and pilot playbook. Phase 3 Scaling — Embedding establishes service-area teams, platform-as-product, data products, operating rhythm, and shared learning. Phase 4 Adaptive — Continuous builds Sense & Respond, portfolio learning, safe scaling, value realization, and readiness reassessment.

Validate IDs structurally, then compare action language and tactic rules against locked findings. Use the activity playbook for roles, artifacts, implementation activities, risks, and acceptance criteria only after the tactic is grounded.

Run separate claim packets for narrative and roadmap, merge verdicts, track each pass, and escalate when material blockers survive regeneration.

Operate on exact claims and deep strategy locations. Remove unsafe actions, rewrite known metric misuse, scrub names and identifiers, and replace any leaked reference-document provenance with neutral methodology language.

Block below the evidence floor, when scored items lack traceable evidence, when Phase 3 validation fails, or when material unsupported claims survive. Warn for partial evidence, low anti-pattern coverage, many silent areas, sanitation, or incomplete fact-checking. Otherwise issue GO.

Persist stage and model usage, source status, evidence adjustments, warnings, sanitation, and import/export diagnostics. If the Quality Gate flips to BLOCK, effective confidence can downgrade to LOW and hide directive language.