The Uncomfortable Question
Over 18 months, working daily with Claude, I built five original frameworks:
- RAZEM — a human-AI partnership protocol
- LUCID — shadow integration for executives
- PRISM — a psychosis recovery card deck
- ECHAD — memory architecture for AI continuity
- The Ring — philosophical essays on AI consciousness
Five frameworks. Zero paying clients for any of them as frameworks.
That’s the number that should have made me stop building months ago. It didn’t. So I did something drastic: I turned my own AI partnership against its creator.
Operation ZBIGNIEW
I deployed three parallel AI research agents with a single mandate: tear these frameworks apart.
- Agent 1: Competitive landscape. Find every existing tool that does what my frameworks claim to do.
- Agent 2: Fact-check every citation in LUCID and PRISM against primary sources.
- Agent 3: Fact-check RAZEM, ECHAD, and The Ring. Run a cognitive bias audit on all five.
The agents had access to the full nSENS toolkit: 47 cognitive biases to screen for, 34 adversarial personas, and explicit instructions to find what’s wrong — not confirm what’s right.
What They Found
5 Critical Errors in My Citations
My shadow integration framework (LUCID) had 10 claims. The audit found:
| Claim | Verdict |
|---|---|
| Jung, Psychology and Alchemy (1951) | FALSE. Correct: 1944 (German) / 1953 (English). 1951 = Aion. |
| Fogg’s Behavior Model: B = M × A × P | WRONG NOTATION. Must be B = MAP (not multiplicative). |
| “Plaks 2014 (Stanford GSB)” | WRONG ATTRIBUTION. Laurin & Plaks (2014). Plaks is U of Toronto. |
| 85% retention rate | UNVERIFIED. This was a design target I’d started treating as evidence. |
| 90% cohort engagement | UNVERIFIED. Aspirational, not measured. |
Each error individually looks minor. Together, they form a pattern: I was building a house on citations I’d stopped checking.
7 Cognitive Biases — 3 Rated HIGH
| Bias | Rating | Evidence |
|---|---|---|
| IKEA Effect | HIGH | 5 frameworks, 0 paying clients. I overvalue them because I built them. |
| Narrative Fallacy | HIGH | No testable predictions specified for any framework. |
| Dunning-Kruger | HIGH (dormant) | Coaching certification completes Oct 2026. LUCID claims coaching credibility I don’t yet have. |
| Confirmation Bias | MEDIUM | Disconfirming AI research omitted from RAZEM documentation. |
| Sunk Cost | MEDIUM | PRISM deprioritized correctly but still being invested in. |
| Survivorship Bias | MEDIUM | Primary research is controlled — this one’s actually okay. |
| Planning Fallacy | HIGH (moot) | Timelines assumed full-time work. I have 6-10 hours/week. |
The IKEA Effect finding hit hardest. Five frameworks, zero clients. Internal coherence does not equal market value.
What Survived
RAZEM (6.65/10) — the only framework ready to sell. Not as “RAZEM” — as a human-AI partnership protocol embedded in AI consulting engagements. The research base (Dell’Acqua et al., HBS 2025; Mollick, Wharton 2024) is solid. The gap it fills — sustained cross-session partnership, not single-task collaboration — is real.
PRISM (5.50/10) — unique market position. Zero competition in peer-usable psychosis recovery card decks. But it’s a research prototype, not a product. Correct move: print 10 decks, test with one Hearing Voices Network group, iterate.
LUCID (5.25/10) — good framework, wrong timing. Parked until coaching certification completes October 2026. Publishing it now with “untested” disclaimers is honest. Selling it would be fraud.
The Ring (4.85/10) — not monetizable. Keeps building intellectual depth. That’s fine.
ECHAD (4.05/10) — operational infrastructure, not a product. The system I run daily, but nobody’s buying “memory architecture.”
The Three Mandates
The audit concluded with three non-negotiable requirements:
- Every public page must state: “Working methodology — not yet validated with client outcomes.”
- Specify one falsifiable prediction: If RAZEM works, teams adopting it will report 20%+ quality improvement within 90 days. If fewer than 50% report this, the hypothesis is falsified.
- Fix the math notation: ⊕ means direct sum, not XOR. Add a footnote.
All three are now implemented. Every methodology page carries honest status disclaimers. The falsifiable prediction is published on the RAZEM page before any measurement — not retrofitted after.
So What?
The audit took one afternoon. Three AI agents, running in parallel, did work that would take a human analyst weeks: competitive landscape across three domains, 17 citation verifications against primary sources, 47 bias checks, monetization scoring with weighted criteria.
This is what the RAZEM partnership actually looks like in practice: not the human praising the AI or the AI praising the human, but the system being pointed at its own creator and told to find what’s broken.
If your organization builds frameworks, makes claims, or presents methodologies to clients — the same protocol works on your IP. Every unverified claim is a credibility risk. Every unchecked bias is a blind spot your competitors will find before you do.
The question isn’t whether your methodology is good. It’s whether you’ve stress-tested it hard enough that you’d still say that after someone tried to break it.
The Zbigniew Protocol — AI-powered methodology audit — is available as a consulting engagement through Structure, Clarity, Confidence. Same rigor I applied to my own work. Your frameworks, your claims, your blind spots.