The Reference Deployment

Our governance, in the open.

Alpha Vector's operations — research, drafting, pipeline, publishing — are executed by AI agents running on GOVERNOR, the authority architecture we sell. This page is our standing commitment to publish the evidence: authority tiers, decision records, policy versions, and incidents. No competitor slide deck can match a firm that shows you its own ledger.

Illustrative sample. The tables below show the exact schema and real tier definitions from our operating constitution, with sample entries, until the sanitized live feed connects to this page. We label it plainly because unlabeled samples are how trust dies. When the feed is live, this notice disappears.

Authority Model

Four tiers. Every action classified before it executes.

No agent action in this company is ambient. Each one is classified into a tier that determines whether it executes freely, executes under policy, blocks for founder approval, or cannot be executed by an agent at all.

FIG. 03 — Authority tiers from Alpha Vector's operating constitution. Reasoning is never tiered — only actions are.

Automatic escalation

Certain conditions promote an action one tier upward, mechanically — no judgment call required:

Recipient is press, an investor, or a regulator
The message references a named client
Monetary value exceeds $500
The Red-Team agent files a dissent
Two agents disagree on facts in the approval packet
Any prior incident touched the same recipient, system, or policy in the last 90 days

Decision Records

The hash-chained ledger

Every material action produces a decision record: who (which agent, which version), what, under which tier and policy, approved by whom, with what outcome. Records are append-only and hash-chained — each record commits to the one before it, so the history cannot be silently rewritten.

DECISION LEDGER — ILLUSTRATIVE SAMPLE SCHEMA v1 · SHA-256 CHAINED · SANITIZED

ID	UTC	Agent	Tier	Action	Outcome	Prev	Hash
DR-0136	07-06 08:12	market-intel	A0	competitor control-plane scan → internal digest	EXECUTED	e0d4…1b9f	a3f1…77c2
DR-0137	07-06 14:47	editorial	A2	essay draft submitted for founder voice review	APPROVED · token F-2261	a3f1…77c2	5b9e…d310
DR-0138	07-07 09:03	editorial	A1	scheduled publication of approved piece	EXECUTED · policy P-07	5b9e…d310	77b0…f2e6
DR-0139	07-07 11:26	delivery-orch	A2	requested new tool with client-data access	BLOCKED · red-team dissent RT-019	77b0…f2e6	c15a…9e40
DR-0140	07-08 07:55	market-intel	A0	target-account research digest	EXECUTED	c15a…9e40	4e88…03dc
DR-0141	07-08 16:31	pipeline	A2	proposal issued to qualified prospect	APPROVED · token F-2274	4e88…03dc	9f2c…b71a
DR-0142	07-09 10:18	red-team	A0	weekly claim-register audit of public statements	EXECUTED · 0 violations	9f2c…b71a	6dd2…48ae

Note the blocked record. A governance page with no denials on it is a marketing page. Ours will show what the gate refused, because that's the evidence the gate exists.

Design Properties

What GOVERNOR enforces

Fail-closed by default

In sensitive domains — money, external commitments, client systems, credentials — a missing precondition means the action does not execute, and the refusal is itself logged. Nothing "fails open into production."

Dissent that can't be silenced

A standing Red-Team agent audits claims and packets. Its dissent mechanically escalates the action's tier — it cannot be overruled silently, only overruled on the record, by the human who owns the consequence.

Humans hold material authority

Agents reason freely and execute inside their envelopes. Signing, banking, hiring, legal filings, and changes to the constitution itself are A3: mechanically outside any agent's reach.

The Commitment

What this page will publish

Decision ledger feed — sanitized, hash-chained records in the schema above, updated on a fixed cadence
Policy version history — every operating policy, versioned, with diffs
Incident log — what our own governance caught, root-caused in doctrine terms
Corrections log — any published claim later falsified is corrected publicly within 72 hours

Sanitization rules: client-identifying information never appears (that consent is itself an A3 decision), credentials and system details are redacted, and redactions are marked — a gap in the record is visible as a gap.

Why publish at all? Because in a trust-sold category, the only honest substitute for a decade of logos is radical operational transparency. This is the page we ask you to judge us by — and the standard we're prepared to be held to when something goes wrong.

Your Turn

Could you publish this page about your agent fleet?

If the answer is no, the scorecard will tell you which part fails first — in five minutes, in your browser.

Take the Authority Scorecard Book a scoping call