Skip to content
the truth layer

A publication on metric integrity

Notes on metrics, governance, and the trust gap.

Every enterprise data platform has a layer for the plumbing and a layer for the definitions. Almost none have a layer for confidence.

The truth layer is a machine-readable confidence layer that sits between your data platform and your decisions, scoring whether a metric can be trusted right now, by humans and AI agents alike.

SubscribeRead sample →

Built and written by Justin Nixon. Twenty years building governance infrastructure that makes enterprise data trustworthy enough to act on.

Explore

See the map →

More to explore

Severity classifier

See how the trust layer classifies definition changes and resets consumption confidence proportionally.

The agent lifecycle

Six stages of an AI agent consuming a governed metric, from query embedding to calibration. Interactive.

Watch trust evolve

Two event types, one trust signal. Trigger severity changes and agent decisions on the same metric and watch the unified consumption confidence absorb each one.

Copilot vs the contract

Same cell, same instant. Copilot says PASS. The truth layer says FAIL. See why.

Three places calibration fails

Same readings, different verdict. Toggle the weighting and the aggregation. Watch a trust score flip from Act to Escalate without a single signal moving.

The falsification engine

Click into three real claims from the trust contract design session. Each shows the question that landed, what collapsed, and the amendment that shipped.

The jurisdiction boundary

One dial: how much does the trust layer read? Drag it up and watch detection flatline while exposure rockets. Reading the one governed cell and stopping is the optimum.

Keep score

A grade is a prediction. Run the loop, grade a metric, let the agent act, and watch how often "safe" actually held. Keep score and the claimed 80 falls to meet reality. Turn it off and it stays 80 forever.

The write-back problem

The moment an agent acts on a number and the action goes wrong, the reflex resets the number’s trust. But the number may have been fine, and the agent may have grabbed the wrong one. Toggle the attribution guard and watch the loss find the right ledger, or watch a good number crater for a mistake it never made.

The cost of distrust

Distrust is a line item, paid every week on numbers that were right. Enter your own teams, meetings, and shadow copies, watch an annualized cost assemble from four components, then drag one coverage slider and watch the bridge take it away.

Field note

“Trust is built by humans, consumed by machines.”

Justin Nixon, from the commute notes

Writing

All essays →

More essays

The Jurisdiction Boundary

A trust layer that reads more does not catch more. It only exposes more. Where its authority to read should stop, and why the optimum is a single governed cell.

The Truth Layer, End to End

One governed metric, traced from definition to decision. Watch the confidence signal travel every layer and survive the trip into a spreadsheet.

The Falsification Engine

Ten hard questions asked against one trust contract design. What survived, what collapsed, and the amendments that shipped because of it.

The Calibration Gap: Why Your Trust Score Was Weighted by Feel

Five signals get combined, the composite ships, the agent acts on it. Almost no architecture audits whether the weighting was set by outcome or by feel.

The Excel gravity well: why the biggest consumption surface is the hardest to govern

Every metric you ship ends up pasted into a workbook within a week. The most-consumed data surface in the enterprise is the one your governance layer can’t reach.

Three weeks, three amendments: a method for stress-testing your own architecture

A senior finance analyst gives notice on a Wednesday. By Friday, three of her workbooks have stopped behaving. Nobody changed the formulas or the data.

Semantic drift and the layer nobody built

A metric owner changes one field on a Tuesday afternoon. Swaps a trailing 12-month average for a rolling 6-month window. Commits the change. Goes to lunch.

The eval gap: why agent observability isn't enough

It's the 3 AM page that never fires.

Why your AI agent needs a trust contract

AI agents are making decisions with your data. Trust contracts give them the confidence signals to know when to act and when to stop.

The truth layer crisis

What's missing between your data platform and trusted decisions.