Portable Runtime Evidence for Independent Accountability

As intelligent systems become embedded in decisions, workflows, institutions, and public infrastructure, accountability cannot depend entirely on provider-controlled dashboards, internal explanations, or screenshots captured after something goes wrong.

Those materials may describe an incident, but they rarely provide a portable evidentiary record of how behavior developed, which methods were applied, what the source supports, and where the resulting claims must stop.

Evidence Commons begins from a different premise: evidence of consequential computational behavior should be capable of being generated, retained, inspected, and challenged beyond the organization or system that originally produced it.

Evidence Commons is a standards-oriented public infrastructure framework through which source-bound runtime evidence can be preserved, exchanged, compared, independently inspected, and challenged across systems and institutions—without surrendering provenance, privacy controls, method identity, or explicit claim boundaries.

Its governing principle is:

Evidence should be portable enough to outlive the system that produced it and bounded enough to remain accountable to its source.

From Provider Assurance to Independent Evidence

Most current approaches to AI accountability begin with assurance.

A provider may publish an evaluation, expose a monitoring dashboard, issue an incident report, or explain how a system is intended to behave. These practices are useful, but the provider still determines:

• what is measured;

• which records are retained;

• how an incident is framed;

• which methods are disclosed;

• and what outside reviewers are permitted to inspect.

Evidence Commons does not eliminate the role of providers or institutions. It establishes an additional possibility: independently held runtime evidence derived from observable operational records.

This changes the governing question from:

What does the system or provider say occurred?

to:

What source-bound evidence exists, how was it produced, and what conclusions does it actually support?

From One Runtime to a Commons

Runtime Evidence begins with an individual operational record.

Fieldglass can qualify that source, construct a canonical runtime, compute authorized measurements, reconstruct its observable trajectory, and preserve the resulting evidence with provenance and explicit claim boundaries.

Evidence Commons extends that process beyond one investigation:

Observable Record
→ Current Evidence Run
→ Runtime Reconstruction
→ Instrument Findings
→ Certified Runtime Evidence Record
→ Runtime Evidence Passport
→ Evidence Commons

Each layer has a distinct responsibility:

• Runtime Evidence defines the source-bound evidentiary object.

• Evidence-Governed Computation governs how measurements, interpretations, and claims derive authority.

• Fieldglass makes the reconstruction and investigation process operational.

• Runtime Standards define the schemas, semantics, conformance rules, and disclosure requirements needed for interoperability.

• Evidence Commons enables preserved evidence to move across systems, organizations, and independent reviews.

Fieldglass produces the evidence-bearing artifact. Evidence Commons gives that artifact a life beyond Fieldglass.

The Portable Evidence Artifact

The central object of Evidence Commons is not a dashboard image or detached report. It is a structured evidence package whose contents preserve their relationship to the originating record and declared computational methods.

Depending on its disclosure posture, a package may contain:

• a Certified Runtime Evidence Record;

• a Runtime Evidence Passport;

• canonical runtime identity;

• source hashes or included source material;

• reconstruction frames and worldline structures;

• authorized measurements and instrument findings;

• temporal coordinates and markers;

• method, schema, and instrument versions;

• provenance and transformation lineage;

• evidence availability and missingness states;

• operator-declared context identified as such;

• claim boundaries;

• integrity and preservation metadata;

• and a declared reproduction posture.

The deterministic core must be distinguished from variable issuance information. Canonical evidence identity and computed findings may be reproducible from the same qualified source and method versions, while complete export envelopes may also contain timestamps, preservation events, operator context, or review metadata.

Those variable elements must remain identifiable and must not silently alter the canonical evidence.

Privacy Without False Reproducibility

Evidence portability does not require every source record to become public.

Operational records may contain sensitive conversations, personal information, proprietary system details, security events, regulated data, or confidential organizational context. Evidence Commons must therefore support selective disclosure and privacy-preserving participation.

However, privacy and reproducibility are not the same property.

An artifact can preserve integrity while withholding the source, but another party cannot fully recompute the evidence without access to the material from which it was derived. Evidence Commons must disclose that limitation precisely.

Disclosure postureWhat can be independently establishedMetadata onlyArtifact identity, issuance, declared methods, integrity data, and disclosure statusDerived evidence onlyPreserved findings and computations can be inspected, but not fully recomputed from sourceRedacted source packageQualified reproduction may be possible within the disclosed source scopeComplete source packageIndependent re-execution may be possible using the declared implementation and method versions

No posture should imply more authority than it provides.

Privacy may limit disclosure. It must never be used to conceal the limits of reproducibility.

Comparison Without Erasing Context

A commons makes evidence comparable, but comparison must remain bounded.

Two artifacts should not be treated as equivalent simply because they contain similarly named metrics. Meaningful comparison requires compatible:

• source qualifications;

• coordinate systems;

• schema versions;

• measurement contracts;

• evidence horizons;

• temporal marker rules;

• instrument versions;

• disclosure postures;

• and claim definitions.

Evidence Commons can support comparative investigation of regime patterns, drift, recovery, boundary formation, temporal structure, and other runtime findings—but only where the relevant methods and evidence coverage are compatible.

Aggregation does not automatically establish scientific generality. Repeated findings may reveal a pattern worth studying, but broader conclusions require appropriate datasets, controls, calibration, and independent validation.

The purpose of the Commons is not to manufacture consensus. It is to create a shared evidentiary structure through which agreement and disagreement can be inspected.

Independent Review and Challenge

Portable evidence becomes more valuable when it can be examined by people other than its original producer.

A researcher may test whether a measurement reproduces. An operator may dispute a classification. An auditor may inspect whether a claim exceeds its evidence. A standards body may identify a conformance failure. An affected institution may add relevant external context.

Evidence Commons is intended to support this process without rewriting the original artifact.

A review or challenge should preserve:

• the identity of the artifact examined;

• the version of the methods applied;

• the scope of the review;

• the evidence available to the reviewer;

• the disputed finding or claim;

• the reviewer’s conclusion;

• unresolved differences;

• and the relationship between the original and subsequent record.

The original evidence remains intact. Review creates an additional traceable layer around it.

This allows accountability to develop through inspection and challenge rather than through the replacement of one unsupported narrative with another.

The Role of Standards

Evidence Commons depends on common artifact contracts.

The Evidence Layer Standard, currently maintained as a canonical draft specification, defines how portable runtime evidence may be structured, replayed, disclosed, and governed. Related standards define runtime records, temporal compliance, classification, regimes, telemetry, evidence support, and interpretation.

Standards make it possible to determine:

• what an artifact contains;

• which version produced it;

• how its identity was established;

• whether required fields are present;

• what can be replayed or recomputed;

• which terminology is canonical;

• and where an implementation diverges.

The standard defines the artifact contract. Evidence Commons defines the accountability ecosystem in which those artifacts can circulate.

Standards conformance does not certify that every scientific interpretation is valid. It establishes that the artifact follows a declared structure and method contract.

What Exists Today

Evidence Commons is being developed from an implemented technical foundation.

Fieldglass currently provides the reference pathway for:

• browser-local source processing;

• role-aware canonicalization;

• deterministic core computation;

• runtime reconstruction;

• evidence-bound instrument findings;

• Runtime Evidence Passports;

• Certified Runtime Evidence Records;

• explicit claim boundaries;

• preservation workflows;

• and portable evidence exports.

This establishes that runtime evidence can be formed as an inspectable artifact rather than remaining confined to an analytical interface.

What the Commons Is Becoming

The broader Evidence Commons remains an emerging infrastructure program.

Its future scope may include:

• portable artifact registries;

• permissioned and public evidence collections;

• independent conformance verification;

• privacy-preserving evidence exchange;

• challenge and review records;

• compatible cross-artifact comparison;

• institutional and research repositories;

• federated preservation;

• public-interest runtime datasets;

• and long-term provenance across changing systems and implementations.

Evidence Commons should not be presented as an already completed universal network. It is an architecture, standards program, and public infrastructure direction whose first operational foundation is the evidence produced through Fieldglass.

A Civic Layer for Computational Accountability

Evidence Commons is civic because it distributes the capacity to retain and examine evidence.

A researcher can preserve a reference run.

An organization can retain evidence from an operational incident.

An auditor can inspect the derivation of a finding.

A developer can compare evidence produced before and after a system change.

A regulator can request structured artifacts rather than relying solely on screenshots or summaries.

A public-interest organization can study disclosed evidence patterns without requiring unrestricted access to every underlying record.

This does not remove institutions from accountability. It gives institutions, researchers, operators, and the public a stronger common object around which accountability can occur.

Evidence Commons is not anti-institutional. It is opposed to exclusive control over the evidence required to scrutinize consequential systems.

Claim Boundary

Evidence Commons does not automatically establish:

• objective truth;

• legal proof;

• regulatory acceptance;

• causal attribution;

• responsibility or blame;

• complete reproducibility when source material is withheld;

• comparability across incompatible methods;

• scientific validity merely because computation is deterministic;

• or public accountability merely because an artifact has been exported.

A preserved artifact may be intact, deterministic, and inspectable while its interpretation remains provisional or contested.

The defensible claim is:

Evidence Commons establishes the structures through which runtime evidence can remain portable, attributable, comparable within declared compatibility boundaries, and available for independent inspection beyond the authority of the system that originally produced it.

The Complete Arc

Evidence Commons is the public and institutional extension of the wider body of work.

Recursive Science studies intelligence in motion.
Runtime Intelligence names the behavioral organization that develops during operation.
Runtime Evidence reconstructs what the observable record supports.
Evidence-Governed Computation constrains what may be measured and claimed.
Fieldglass makes the architecture operational.
Evidence Commons preserves the resulting evidence for independent scrutiny across time, systems, and institutions.

The objective is not to replace every explanation with a metric or every institution with a public repository.

It is to ensure that consequential computational behavior can leave behind something more durable than assurance:

a source-bound, method-declared, privacy-aware, claim-governed evidence record that others can inspect, question, preserve, and—where disclosure permits—reproduce.