Pieta · Knowledge Refinery

Pieta

Grounded Divergence

We force AI models to disagree, then patent whatever survives contact with reality. 97.9% of what we generate is structurally sound and novel to every database we check it against.

97.9% Survival rate
when system eats its own output
1,800+ Validated artifacts
generated to date
$4.14 Total LLM spend
per 25-book batch
13 Patent-ready
mechanism blueprints
REQUEST TECHNICAL BRIEFING Explore the 4 products →
What We Produce

Four offerings. One engine. Zero fluff.

01
Patent-Ready Blueprints
Mechanism blueprints grounded against real patent neighbours, packaged for provisional filing. 13 candidates in active pipeline.
02
Anti-Collapse Training Data
Synthetic data for LLM labs that resists model collapse. Gen 1 → Gen 2 tested and proven at scale with 97.9% survival.
03
Physics Validation
Doctorate-level structural audit of new concepts. Every candidate checked against 791 EPO patents before it ships.
04
Cross-Domain Hypotheses
Enterprise R&D fuel: non-obvious combinations across 32 domains, delivered with full audit lineage and provenance.
Technical Briefing

See the full picture.

The complete Pieta overview — architecture, validation data, unit economics, and partnership terms — is available to qualified Enterprise R&D and LLM lab teams. Leave your email and we'll send the briefing within 24 hours.

No spam. No lists sold. We reply personally within 24 hours.

Pieta Knowledge Refinery

Pre-seed overview · July 2026 · Confidential

01 / 12 Grounded Divergence

Pieta forces AI models to disagree, then patents whatever survives contact with reality. 97.9% of what we generate is structurally sound and novel to every database we check it against.

Patent-pending 3-Pool Refinery · 4-LLM triangulation with live fallback · 8,431 patent-derived atoms from 791 real EPO patents grounding the substrate.

25 books
Latest validated run
13 doctorates
Deep research volume
$4.14
Total cost, entire batch
2,253
Live concurrent LLM calls
02 / 12 Why Now

Every model trained on the same web is converging on the same answers. The industry calls it model collapse: AI trained on AI output slowly loses the tails of the distribution — the rare, non-obvious, structurally novel connections that actually matter.

More compute does not fix it. It accelerates it.

Our answer: We already tested it — and held. Every LLM eventually degrades when trained on its own synthetic output. Our Gen 1 → Gen 2 test (99.0% → 97.9% survival) is real, repeatable evidence we resist it.

03 / 12 The Moat

0% factual match means nothing on its own. It survives the Gate, or it burns.

Stage Match Rate What It Means
Standard AI 100% Retrieved something already written. Confident, fluent — and unpatentable.
Pieta, Raw Output 0% Unknown. Can mean hallucinated noise, or genuine structural novelty.
Reality Check Gate 3 burned Out of 142 audited. Checked against 8,431 patent-derived atoms from 791 real EPO patents.
Grounded, Patentable 97.9% 139 of 142 survive as real material. Structurally sound. Novel to every indexed database.
04 / 12 The Engine

Pieta does not ask one model to be creative. It forces a collision between three separate knowledge pools, makes four independent models disagree about what it means, and grounds only what survives against 791 real patents.

The Pipeline: Draw → Triangulate → Diverge → Ground → Audit. Every step is fail-closed and fully traceable.

"Time-triggered validity checking of distributed state variables" collided with "Epistemic procedural fidelity" and "Red Layer Constitution" — three unrelated domains, one mechanism, grounded in real patent neighbours, scored 9/9 on book and PhD rigor.

Full architecture available under NDA to qualified partners.

05 / 12 The Substrate

The Universal Library v3.0 locks 32 domains · 7 meta-families · 20 sub-families. 791 real EPO patents atomized against the live OPS API. This is the ground truth every Pieta hypothesis is measured against — not synthetic data.

32 domains
Locked taxonomy structure
791 patents
Real EPO patents, live API
3 pools
Legacy, Universal, Synthetic
100%
Atom utilization before repeat
06 / 12 Proof — Scaled Validation

Gen 2 Survival Rate: 97.9% (139/142)
Gen 1 Baseline: 99.0% (4,692/4,740)
Total Output: 25 Books · 13 Doctorates · 60,903 words
Total Cost: $4.14 · 2,253 live LLM calls · 10 fallback activations

The Anti-Collapse Proof: For Gen 2, we fed Pieta's own synthetic, AI-generated atoms back into the 3-Pool Refinery. The survival rate only dropped from 99.0% to 97.9%. The system consumed its own output without degrading into Habsburg AI.

07 / 12 Resilience

Tested under live fire, not in a demo environment.

Timeline: Q1 2025 concept → Q2–Q3 2025 taxonomy + triangulation → Q4 2025 CTTM + Gate → Q1 2026 791 patents, 4,692 passed → July 2026 3-Pool + Gen 2 validation

Proven Live: 2,253 concurrent calls · 14 bugs fixed live · Zero data loss · Recovered from two real infra incidents

Next 90 Days: Scale to 1,500 patents · First licensed engine pilot · First IP Foundry into provisional filing · Real atomic locking on manifest store

08 / 12 Dual-Layer Model

Product 01 — The SaaS Engine: Disagreement Intelligence. License the 3-Pool Refinery to Enterprise R&D and LLM companies. $50K setup + $25K/mo subscription.

Product 02 — The Monopoly: The IP Foundry. We run Pieta on ourselves. Every grounded mechanism is ours to patent and license into energy, defense, and pharma. Customers rent the engine; we keep the inventions.

09 / 12 Disagreement Intelligence

Rent the engine that tested itself against model collapse — and held.

A hosted deployment of the 3-Pool Refinery, tuned to the customer's own domain taxonomy — generating disagreement-forced, gate-audited synthetic data their own models can train on without collapsing.

Target Customers: LLM Labs (anti-collapse training data) · Enterprise R&D (non-obvious tech combinations) · M&A Diligence (structural blind spots) · Government Research Labs (patent-proven gaps)

10 / 12 IP Foundry

We do not sell prior art. We invent the future of it.

We keep running the refinery internally, on our own 791-patent substrate. Every mechanism that survives the Reality Check Gate — structurally sound, free of contradiction, and novel to every indexed database — is a candidate we own outright.

Licensing Targets: Energy (material-science / systems-engineering collisions) · Defense (non-obvious sensing/control/materials combinations) · Pharma (structural mechanism candidates grounded against real patent neighbours)

Model: Royalty-bearing exclusive and non-exclusive licenses on Pieta-originated, patent-filed mechanisms.

11 / 12 Defensibility

The moat is the system, not just the output.

Layer 01 — IP: Patent-pending workflow (Ref. 2026/07141). Routing, orchestration, and CTTM grounding architecture filed.

Layer 02 — Substrate: 32-domain taxonomy + 791 EPO patents atomizing against live OPS API today.

Layer 03 — Provenance: Reality Check Gate. 97.9% survival at scale, fail-closed.

Layer 04 — Track Record: 14 bugs, zero data loss. Proven operational resilience under real, concurrent live-fire load.

The substrate compounds — by the time you copy the patent, the library has moved and the IP Foundry has already filed the next one.

12 / 12 The Ask

Pre-seed capital to convert proven at scale into paying at scale.

① Scale the substrate to 1,500+ patents · ② Productize the Disagreement Intelligence engine · ③ Land 2–3 anchor engine customers · ④ File the first IP Foundry mechanisms

"The data wall is real. We already tested ourselves against it. 97.9% survival. $4.14 per batch. Zero data loss. That is the pitch."

Contact: David A. B. Van der Walt · Sole Founder · david@pietarien.com