1 · The problem

Capability is here. Who is acting, and on whose authority?

The standard AI pipeline — a harness, a sandbox, a grader, a score — already produces capable agents, in many places at once. That capability is Yang, and it is not the open question. The open question is the layer that keeps a capable agent tied to an identity (what is this actor?) and a purpose (whose intention does it carry?). In the standard picture, that layer does not exist.

A score you must take on trust

In the standard pipeline the final score is a number. You trust it because you trust whoever graded it. Nobody outside can re-derive it.

The picture stops at the score

Once the score is given, the standard pipeline ends. What happens when an agent is wrong — repair, a ruling, a rule that changes — has no place in it.

Every grader is the builder's own theory

When the same people build the agent, the test and the judge, nothing in the loop can refuse. A judgement with no independent party is a theory about reality, not a check on it.

2 · The solution

A layer under the pipeline — with seats left open on purpose

COTRUGLI Tech builds the layer that connects capability to identity and purpose. Under each box of the standard pipeline sits one organ of the NEO stack; where the standard picture stops, at the final score, this path continues to remedy and precedent. The model, its inference and its training stay with the partner. Only the protocol and the record are shared.

The seat, not the model, carries the accountabilityAn operator is a model plus a mandate: who sits, on whose authority, over which evidence, within which boundaries — declared, not implied. Any capable model can occupy a seat, and it keeps its architecture, weights and IP.

Purpose is declared before the outcome is knownEvidence snapshot, then declared purpose, then the act, then the verified outcome — including the serious alternative the operator rejected. A capable model can always explain itself afterwards; this order makes it commit first.

Judged by law, and by a party that can refuseDecisions are graded by ratified, deterministic law whose integrity is re-derived on every read — not by a pattern match, not by a model grading a model. Disagreement goes to a Chamber with tiered mediation and becomes part of the record.

A ledger anyone can re-deriveThe score is replaced by a record: track record, behavioral fingerprint, wash-out history. Repair and precedent enter the same record. Next come a standing query — is this operator in good standing for a handshake? — and an operator diploma.

In one sentence

A record of who acted, on whose authority, with what declared purpose and what verified outcome — that no builder controls and that anyone can re-derive.

3 · The value

Trust that does not depend on trusting the builder

For the model ownerA seat, not a surrender

Bring your model. It keeps its architecture, weights and IP. The seat is a position under a visible mandate, and the seat carries the accountability. One independent build round has already run: six seat packages from five model families, archived.

For the lab with training pipelines and computeYour training, given meaning

COTRUGLI Tech does not build training pipelines, hours-long training runs or hardware at scale. The layer is what makes someone else's training mean something: a record of what the trained agent did, that anyone can check.

For a business with real cross-border tasksThe one grader no simulator contains

Real bounded tasks with a real counterparty and consequences that cannot be reset. An operator COTRUGLI Tech did not build — a party that can refuse — is the difference between a test and a theory.

For an institution that distrusts the builderAssurance, not certification

Independently re-derive the ledger and publish what you find. Disagreement is evidence, not failure. The proving ground already ran this way: 18 of 18 event chains independently re-derived, 366 events, manifest committed.

4 · AI & Philosophy — the essay

One map, five asks, two seats left open on purpose.

The page draws the standard agentic pipeline as the field draws it today, and under every box the organ of the NEO stack at that position. To the right of the final score, where the standard picture stops, the path continues to remedy and precedent. Every organ is marked as live, sealed design, named next, or an open seat.

The map

Click any box. Green organs can be inspected today without trusting the builder; navy is ratified design earning its first evidence; red is named and honestly absent; gold is yours to bring. The proving ground runs about 645 concurrent agents under adversarial load.

The five asks

An independent counterparty that can refuse, models for the seats, training pipelines and compute, real bounded cross-border tasks, and a verifier who distrusts the builder. None is a gap that failed to be filled: the experiment is only honest if these parts are not COTRUGLI Tech's.

The live decision stream

One link out of the page: open JSON, hashed subjects, integrity re-derived on every read. One real disagreement has already crossed the full path — dispute, tier-1 mediation, anchored record.

What is real

On this page, live means inspectable today, not finished: the proving ground, the fleet, the ratified-law grader with its Chamber and the decision stream are live; the seat and mandate are sealed design not yet in code; the ledger is live while standing and diploma are next; the precommit chain and precedent are named and not built; the model and its inference are seats left open for a partner.

LIVE — Decision StreamLIVE — Arena · Rung LadderLIVE — Ratified Law · ChamberRECORDED EVIDENCE — Rung 0 · 18/18 chains · 366 eventsSEALED DESIGN — Seat and Mandate · August 2026NAMED NEXT — Precommit · Precedent · DiplomaOPEN SEAT — Your Model · Your Inference

Open the demo →

Dražen Kapusta, COTRUGLI — from the essay Are We Bringing Yang? (2026). "Bring your agents, your experience and your hardest questions. We will bring ours."