NEO5: The Asymmetric Nexus Protocol

Why this was built

In April 2024, a conviction formed: Palantir was being mispriced as a defense contractor when it was actually becoming enterprise AI infrastructure. That call predated any system — it was instinct, pattern recognition from 20 years of watching tech business models, and a willingness to hold through volatility.

The conviction held. It was acted on. It worked.

But the experience exposed a gap. The right call had been made through instinct that could not be inspected, repeated, or handed to the next decision. There was no way to know whether the reasoning was right or the outcome was right. Those are different things.

NEO5 was built to answer that question. Not to find the next Palantir — to build the judgment infrastructure that could validate an asymmetric call before capital moved.

What was built

The Asymmetric Nexus Protocol — NEO5. A multi-agent investment decision system. 157+ documented sprints, live at neo5.thessenchill.com.

The architecture evolved in two phases.

Phase 1: The agent swarm

Nine named agents, each with a distinct role in the reasoning chain:

Data backbone: SimFin API for fundamentals, Stooq for technicals, an archive-driven signal pipeline. Each sprint produced a documented proof note.

Phase 2: The Capital Gate

As the system matured, the focus shifted from generating verdicts to enforcing the use of verdicts. The Capital Gate introduced a formal permission ledger — no capital movement without a logged decision.

Every decision attempt passes through five permission levels:

The gate blocks an ADD if: valuation is unknown, evidence is incomplete, or the operator cannot state what changed since the last decision. Override attempts are logged as decision debt — they don’t elevate permission.

What the system produced

The bake-off (April 2026)

A structured head-to-head: NEO5 versus a human analyst, same ticker, same time window.

Ticker: SMCI.

NEO5 won. The analyst’s “something here” is not a decision. NEO5’s “fundamentals stale” is.

The Delta-4 behavior proof (April 2026)

Primary pick: AXON over RKLB.

The reasoning was explicit: 60% software-grade margins for AXON versus 38% gross margin for RKLB. Government demand grounded with specifics — 205 awards, $226.5M in 12 months.

NEO5 did not surface a guess. It surfaced a reason, in writing, before capital moved.

The gap — version 1

The system ran. The agents deliberated. The verdicts landed.

Then came the proof log.

The Capital Behavior Proof Log was built to track one thing: documented instances where the system changed what the operator did with capital. It had space for ten entries.

All ten rows were empty.

The system was not wrong. The theater was wrong. The build had been for the satisfaction of having built something rigorous — not for the decision the system was supposed to make better.

The gap — version 2

The system was rebuilt. The Capital Gate replaced the proof log as the primary decision surface. Seven real decisions were logged through the gate.

All seven returned L1: Wait / Research Only.

The Consequence screen reported:

Operator Followed Gate: 0.

Not zero overrides. Zero confirmations that behavior changed. The system tracked the decision and stopped there. “Capital Saved: $11,500” is aspirational accounting — the gate said no, but it has no record of whether the operator honored it.

The system got smarter. The theater got smarter with it.

The consequence contract

Written into the system. One rule:

Create one documented capital behavior change.

What counts: a blocked chase, a reduced size forced by valuation burden, an exit triggered by a broken thesis, a decision memo written before moving capital.

What does not count:

More documentation. More prompts. More architecture. More dashboards. More analysis that does not affect capital behavior.

Contract status: OPEN.

What this proves — and what it does not

A system can be technically correct and still not change what the operator does. The gap between “the system is right” and “the operator acted on it” is not a product problem. It is a behavior problem, and behavior problems are not solved by better dashboards, better signals, or better architecture.

The consequence contract made the gap visible. Most builders never write the contract, so the gap stays invisible and gets misdiagnosed as a product problem — forever.

What this does not prove:

• The conviction that started this predates NEO5 by eight months. NEO5 did not find the original asymmetric call. It was built to replicate the judgment that made it — and to validate the next one.
• The SMCI bake-off victory does not prove NEO5 would have produced a closed consequence entry. Winning a memo comparison is not the same as changing a capital act.
• A different operator with different habits and capital constraints might have closed the contract.

What the project produced beyond itself

NEO5 the investment system did not close its consequence contract. NEO5 the building process produced real, reusable infrastructure:

The consequence contract — a diagnostic instrument. Write it before you build the system. If it cannot close, you are building for the satisfaction of building, not for the decision you said the system was for.

The Capital Gate and permission ledger — a formal gate pattern now applied to other decision domains. The permission taxonomy (L0–L4) is reusable wherever a decision requires both evidence and behavioral accountability.

The proof-versus-summary discipline — every agent produces a summary; proof is reviewable evidence that can be checked without trusting the chat. This distinction became the foundation of subsequent work.

The behavioral finance interrupts — FOMO, action bias, sympathy chasing, disposition effect — named and routed at the point of decision, not diagnosed after the fact.

The byproducts outlasted the original system’s open contract.

Current status

The site stays live. The repo is active. The consequence contract is open.

The field “Operator Followed Gate” currently reads: 0.

This is not a confession. It is the experiment. The contract has a date. When it closes — in either direction — the update publishes here.

Related: Dashboard Theater — the named pattern this receipt proves.