The problem is not another dashboard
An effect claim often ends its life as a sentence in a slide: the program improved timeliness, saved a certain amount, or reduced a disparity. The sentence moves easily. The conditions that made it defensible do not.
The intervention version may be missing. The comparison may be unclear. The outcome definition may have changed. The confidence interval may be absent. A gross value may be presented as attributable savings. The result becomes detached from the design and evidence that produced it.
An Impact Passport keeps those elements together. It is a versioned, machine-readable record attached to a specific intervention and a specific measurement contract. It states what was estimated, how it was estimated, what uncertainty remains, what value can be attributed, and what claim is permitted.
THE CORE RULE
The claim cannot travel farther than its evidence.
Three artifacts, three questions
The Impact Passport is part of a small accountability system.
Can this exact artifact operate as intended?
What evidence must exist before a causal or ROI claim?
What does the observed evidence support now?
The Measurement Contract is written before deployment or before the result is inspected. It fixes the question, estimand, population, comparison, outcome, horizon, causal graph, diagnostics, and valuation rules. The Impact Passport is issued after an analysis run and points back to that contract by version and digest.
The Control Passport follows the operational artifact. The Impact Passport follows the evidence about its effect. They should not be collapsed into one document because operational assurance and causal attribution are different claims.
Minimum conforming record
Version 0.1 requires ten groups of information.
Subject
Instrument name, version, artifact digest, organization, and operating environment.
Contract
Measurement Contract identifier, version, and digest.
Question
Population, intervention, comparison, outcome, estimand, and time horizon.
Evidence
Source identifiers, observation period, unit of analysis, counts, and missingness.
Design
Identification approach, causal graph reference, adjustment set, and rollout structure.
Diagnostics
Overlap, balance, assumptions, falsification checks, and sensitivity results.
Estimates
Method, effect, unit, interval, and status for every reported estimate.
Value
Buyer-approved inputs, gross and net value, cost, ROI interval, and break-even point.
Claim
Claim level, permitted statement, prohibited uses, limitations, and subgroup findings.
Lineage
Data, code, model, execution, approvals, supersession, and history.
Claims have levels
The template does not assume that every analysis earns causal language. The claim.level field accepts four values.
operational: a descriptive statement about what happened in the observed data.comparative: a difference between defined groups or periods without causal attribution.causal: an effect claim supported by the contract, design, diagnostics, and approvals.withheld: the planned claim is not supported and should not be made.
This field is not a badge. It is a constraint on language. A passport with a withheld claim can be more useful than a confident slide because it shows exactly what failed and what evidence would be needed next.
ROI is an attributed quantity
Impact value is not simply the number of automated tasks multiplied by an assumed labor rate. Version 0.1 separates observed operational change, estimated causal effect, buyer-approved unit value, implementation and operating cost, uncertainty, and attribution.
The template carries gross value, attributable value, net value, an ROI interval, and a break-even point. If causal identification is not adequate, the record can report operational value while prohibiting a causal ROI claim.
A compact example
passport:
id: impact-passport:instrument-x:2026q3
version: 0.1.0
status: draft
subject:
instrument_id: instrument-x
instrument_version: 0.4.2
artifact_digest: sha256:replace-with-generated-digest
measurement_contract:
id: mc:instrument-x:2026q3
version: 1.0.0
digest: sha256:replace-with-contract-digest
claim:
level: withheld
statement: No causal claim is authorized for this run.
prohibited_uses:
- Do not describe observed change as attributable savings.The downloadable template contains the full structure and implementation notes. The JSON Schema validates required fields, formats, enumerations, and the basic shape of the record. It cannot determine whether the causal design is sound. That judgment remains part of scientific and organizational review.
Versioning and refusal
A new passport is required when the intervention, contract, data window, outcome definition, model, valuation input, or approved claim changes. A previous passport should remain available and point to the record that superseded it.
Systems implementing the specification should be able to refuse issuance. Missing artifact identity, a changed outcome definition, failed overlap, an unapproved value input, or a contract digest mismatch are all valid reasons to stop. A portable record is useful only if absence and failure are visible.
Use v0.1
YAMLUsable templateComplete field structure with implementation guidance JSONValidation schemaDraft 2020-12 schema for automated checks
Version 0.1 is intentionally small enough to implement and explicit enough to criticize. We welcome field tests, proposed extensions, and examples that reveal where the specification is too vague.