Code-reviewed current system · 23 July 2026

How the AI Research Council works.

One contested thesis in. Airport context, independent evidence, two distinct challenges, source verification, and a governed executive packet out.

The governed line

One question moves through a governed line.

No model researches, writes, and approves its own answer. Typed artifacts connect each handoff. Failed contracts stop the line; valid completed work resumes.

  1. 1Airport context
  2. 2Independent research
  3. 3Curate + frame
  4. 4Draft + challenge
  5. 5Edit + verify
  6. 6Produce + inspect
54 defined agents

The roster separates depth from control.

Each run seats only the lenses it needs. The manifest records what actually ran.

20
Airport Research

Economics, operations, engineering, airlines, safety, and governance.

18
Supplemental

Outside thinkers, seated only when the question benefits from the lens.

16
Process

Context, curation, writing, review, verification, design, and production.

Stage 1

Airport context comes before analysis

The Context Builder establishes the decision environment — governance, finance, capital, airlines, regulation, and operations — as a shared record every researcher may read. Context removes false assumptions without forcing consensus.

Stage 2

A parallel research swarm protects disagreement

Specialists investigate the same question at the same time, without reading one another's conclusions. Independence ends at curation — after each lens has left a separate record.

Stage 3

Creative framing keeps the story open

The Creative Director proposes truthful ways to tell the same evidence before the report's structure hardens. Creativity changes the route through the evidence, not the evidence itself.

Stage 4

Two adversaries test different failure modes

The Evidence Prosecutor hunts sources, arithmetic, causality, and counterevidence. The Airport Executive Reviewer hunts authority, funding, procurement, staffing, and peak-hour operations. The writer must answer each critic in a new draft.

Stage 5

Fresh eyes verify facts after prose is finished

The Editor cuts and tightens; the Humanizer improves voice; neither may change facts. A fresh model then opens each source, checks exact wording, and binds every reader-facing claim to its origin.

Stage 6

Release requires provenance and rendered QA

The Art Director sets the visual contract before Word or PowerPoint begins. The release gate approves the record — run manifest, evidence ledger, claim lineage, inspection receipt — not the prose alone.

Signature

Every released claim keeps a chain of custody.

Primary source → evidence ledger → reader claim → claim lineage → release gate. Every link must match: claim, footnote, evidence, source, draft.

V1 → V2

Same speed. Stronger challenge. Verified release.

V1 · Multi-agent report

Parallel briefs, one synthesis, human review. Strong analysis — less control of evidence and production.

V2 · Airport decision system

Context + lineage, two challenges, rendered QA, live telemetry. Claims and artifacts carry records, receipts, and gates.

Meet the people behind the process.

The council is defined as source-controlled agent profiles. Browse every member, read their charter, and (soon) propose changes.