Skip to content
The problem

Three teams. One metric.
Three different numbers.

Sales team
Monthly recurring revenue
$0K
Finance team
Monthly recurring revenue
$0K
Product team
Monthly recurring revenue
$0K
0%

of US companies don't trust the data feeding their AI agents.

2024 enterprise data governance survey

What if the number came with a receipt?

A machine-readable contract that says "this number is trustworthy, and here's the proof."

Section 2

Anatomy of a trust contract

A trust contract is a machine-readable artifact that tells you (and your AI agents) whether a metric can be trusted right now. It carries the signals that vouch for a number — lineage, freshness, ownership, and more — and scores the measurable ones into a single grade. Toggle the factors below to see how trust degrades.

Try turning off "Data freshness" below and watch the grade crater.

Monthly Recurring Revenue

Total recurring subscription revenue for the current month at Meridian Analytics.

102d ago
Owner: Sarah Chen
A93/100
Trust Score
Trust factors (toggle to see impact)
Data Freshness
23/25 pts
Governance Status
20/20 pts
Source Reliability
14/15 pts
Owner Accountability
9/10 pts
Data Consistency
8/10 pts
Agent thresholds:Auto-act90Escalate70Halt50
Beyond the contract

Consumption confidence

The factors above are static signals. They describe the metric itself: its lineage, owner, freshness, definition. The trust layer also tracks a feedback signal called consumption confidence: how confidently real humans and agents actually act on the score. When a definition changes, CC resets proportionally. A cosmetic tweak barely moves it; a formula rewrite zeros it out.

See how CC responds to change

Freshness, lineage, and ownership are three of the signals a trust contract carries. The next section lays out the full set — and what each combination of signals lets you do.

What makes a metric trustworthy

Six signals travel with every number. What you can do depends on which ones hold.

A governed metric carries a trust contract — six signals that say where it came from, when it was last computed, who owns it, and how confidently it's been used. Each signal you have unlocks a different move. Stack them, and the number becomes safe to act on — for a person or an agent.

Lineagewhere it came fromFreshnesswhen it was computedOwnershipwho's accountableConfidencehow confidently it's consumedSpot-checktrace a live numberFlag & escalatecurrent, and ownedDefer to itowned & provenReport itthe MVP threeMonitorTrusted actionact without asking “is this right?”audit onlyhearsay

* One signal alone can mislead. A widely-used number feels trustworthy, but popularity is not provenance — the platform calls that false trust, which is why usage never appears as a lobe here. Only the signals that vouch for the number itself move it toward the center.

The trust contract — six signals that travel with every metric

Five describe the number; one is feedback from how it's consumed. The MVP ships the first three — enough for a basic reliability check. The diagram maps those three plus consumption confidence.

Lineage · static
where the number came from
Freshness · static
when it was last computed
Ownership · static
who to contact if it looks wrong
Discoverability · static
can it be found in the catalog
Usage · static
how widely it's relied on
Consumption confidence · feedback
how confidently it's consumed

Diagram structure inspired by David McCandless, “What Makes a Good Visualization?”

Section 3

When trust breaks

Trust doesn't fail in isolation. One broken link degrades every metric downstream. Trigger a failure to see it propagate.

Click a failure scenario below to see trust degrade across the metric chain.

Section 4

The same contract, two consumers

A human reads the trust grade and decides whether to act. An AI agent reads the same contract and decides autonomously. Drag the slider to see how trust level changes behavior.

Analyst dashboard
Metric
Monthly recurring revenue
Value
$4.2M
Trust grade
A
Score
92/100
Decision

"Grade A. I can confidently include this in the board deck."

Reviewed by Sarah Chen
Agent API response
{
  "metric": "MRR",
  "value": "$4.2M",
  "trust_score": 92,
  "grade": "A",
  "action": "auto_execute"
}
Auto-execute

Agent will proceed autonomously. No human review needed.

Drag the slider below 70 to see the agent halt. Above 90, it acts on its own.

Trust Score92
0 (Halt)5070 (Escalate)90 (Auto-act)
Section 5

Building the trust layer

Three gates every organization needs to pass before its data can power autonomous decisions.

Where is your organization? Select a gate.

Three layers of agent governance

Most organizations have layer one. Almost nobody has layer three.

Select a layer to see an example.

"If ESPN can agree on LeBron's stats across every broadcast,your company can agree on MRR."

The ESPN test for data trust

The trust layer isn't a product. It's a thesis about how organizations should structure the relationship between data, decisions, and autonomy.

Read the full thesis on Substack →

Part of: Stage 02 · The Trust Contract

Back to the map

A layer is an idea until it has an artifact. The artifact is the trust contract: a machine-readable envelope that travels with every metric, carrying the signals a human and an agent can both read to know whether to act, caveat, or stop.

Read the essay