Essay 05
An Institution Acts. What Can It Explain?
A recommendation becomes an action. The institution needs an account of the evidence, authority and obligations that carried it there.
Research by Mark Julius Banasihan
Research context
Specification & reference implementation · Internal implementation checks
Essay draft · The insurance record is a fictional repository example, not an observed payment or deployment result.
An insurance claim is marked for approval. The proposed payment is $8,400, payable directly to a repair facility. The record lists the evidence considered and names the manager responsible for allowing this kind of automated decision. It contains no timestamp showing that the action was carried out.
This is a fictional example published with Governed Decision Intelligence. It illustrates a distinction that matters whenever an institution acts: a recommendation, permission to proceed and a completed action each require their own evidence.
For someone waiting for the payment, those distinctions have practical consequences. An approval does not tell them whether the money has been sent. If a question arises about the decision, the institution also needs to explain what information it used, whose authority covered the action and who can address the concern.
Governed Decision Intelligence proposes a structured record for connecting those parts of a decision. The question is whether that record gives people enough information to examine what happened and determine what still needs attention.
When a recommendation becomes a responsibility
A model’s output and an institution’s action have different authors. The model may estimate, rank or recommend. The institution decides what authority that output can exercise: whether a person must approve it, whether a prior delegation covers it, and which conditions require a pause. Those arrangements determine whose judgment can affect the result.
Mark Bovens’s conceptual account of accountability locates it in a relationship: an actor must explain and justify conduct to a forum able to question, judge and potentially impose consequences. Applied here, that directs attention to who can demand an answer and what can follow from it. A generated explanation supplies only part of that relationship.
Delegation can be reasonable. Requiring a fresh human signature on every familiar, reversible action can consume attention without adding useful scrutiny. The harder question is whether the institution can recognize when an action has moved beyond the circumstances its delegation covered. A changed document, a disputed fact or a consequence that is difficult to reverse may require a different route.
To recognize that change, a reviewer needs more than the final answer. They need to know the question being decided, the material available at the time, the alternatives considered and the limits of the permission to act. They also need time and relevant expertise. A record assembled for inspection after release may help reconstruct events, while arriving too late to support an intervention that had to occur beforehand.
Governed Decision Intelligence proposes a record organized around that transition. Its Governed Decision Record brings the decision question, evidence, alternatives, uncertainty, human authority and downstream obligations into one inspectable account. It is a specification with a reference implementation. Whether that account helps people reconstruct decisions or exercise control in a working institution remains an open empirical question.
What the record preserves
The repository’s synthetic insurance example asks whether an $8,400 claim should be approved for direct payment to a repair facility. The model recommends approval with a confidence score of 0.87. The record identifies four evidence items: the claim intake form, a damage estimate, a policy record and a fraud-model output. A reader can begin to see what the recommendation rests on.
The W3C PROV model provides an established vocabulary for describing the entities, activities and agents involved in producing information. It helps distinguish a document from the process that created it and the people responsible. That is a useful reference for the provenance problem; the comparison does not establish GDI’s conformance to PROV.
The same example records reasons for caution. The repair facility is new to the network. The model has limited exposure to repair-cost inflation after 2025. Approval and deferral are set out with their stated advantages and disadvantages. These entries make the proposed decision available for scrutiny beyond its headline score. They remain assertions in a synthetic record, whose underlying system references do not identify accessible production evidence.
The example names a Senior Claims Manager as the owner of the delegation that permits automated approval. It says case-specific human review is not required under that delegation and records no completed review. That combination does not establish a missed review obligation. The execution timestamp is also empty, so the example supplies no evidence that payment occurred.
This is a useful discipline for anyone receiving an institutional explanation. “Approved,” “reviewed” and “paid” describe different events. The evidence required to establish each will differ. An institution that preserves the distinctions can identify which question it has answered and which record it still needs to obtain.
Reading the synthetic example
An authorization.
An outcome still unrecorded.
The proposed decision
Approve an $8,400 claim. Four evidence items and reasons for caution are recorded.
The authority named
A Senior Claims Manager owns the delegation. The example does not require case-specific review and records none completed.
The systems affected
The payment service and claims-status portal are identified as downstream systems.
The execution record
The timestamp is empty. The example supplies no evidence that a payment occurred.
Confidence does not settle permission
A number can make a recommendation feel ready. In the example, 0.87 sits above a stated policy threshold of 0.80. That relationship explains part of the recorded rationale. It does not establish an 87 percent chance that the decision is correct. The meaning of a confidence score depends on how it was produced and whether it has been calibrated for the task and circumstances in which it is used.
Chuan Guo and colleagues’ research on neural-network calibration demonstrated that predictive accuracy and reliable confidence estimates can diverge. Their experiments support checking calibration separately. They do not establish the meaning of the synthetic example’s 0.87, whose calibration evidence is absent from that record.
GDI distinguishes permission from the deliberation an action requires. Its reference classifier can route a permitted action toward elevated review because of the tool involved, even when the confidence score is high. A denied action receives hard escalation. These are configured rules governing a proposed action; their operation does not establish that the rules are appropriate for a particular institution.
A pause needs practical force
A required review matters only if the service responds to it. GDI’s example software can identify an action as needing further attention. The system using that result must then prevent the action from proceeding when a hold is required and preserve a record of what happened. Producing the instruction does not establish that anyone acted on it.
Suppose a question arises about the repair estimate in the fictional insurance example. A designated reviewer would need to locate the estimate used, examine the concern and decide whether payment should wait. That would be a new inquiry; the example does not record such an objection or a hold. If the institution requires a pause, its payment process must honor that instruction before money moves.
If the action has already occurred, the task changes. The institution must determine which consequences remain reversible and who can correct them. The GDI example’s payment service and claims-status portal illustrate why identifying downstream systems matters: changing one record cannot establish that every dependent process changed with it. Execution receipts, corrected entries and delivery records would need to show what happened in each relevant place.
An intact record can still be wrong
A record can remain unchanged and still contain incorrect information. A digital integrity check can help detect whether protected content has changed. It cannot establish that the original repair estimate was accurate, that a reviewer understood it or that an explanation faithfully describes how the decision was reached. Those questions require examination of the sources and the circumstances in which the record was created.
More documentation carries costs too. Records can contain sensitive material, personal information and details that are unnecessary for a particular reader. Access, retention and redaction need to serve the inquiry without exposing everything to everyone. The person affected may need a clear explanation and a correction route; an internal investigator may need additional evidence. A machine-readable record is a resource for those accounts, not a substitute for deciding what each audience must be able to understand.
The responsibilities behind the record
For leaders considering this approach, the first decision is who owns the action. A named role needs a defined scope of authority, including which decisions can proceed automatically and which concerns require someone to intervene. The person investigating a disputed payment should be able to find that authority and determine whether it covered the action taken.
The organization also needs to decide what evidence it must retain. In the insurance example, that would include the estimate used, the applicable policy and the permission under which payment could proceed. Evidence that the payment actually occurred would answer a further question. Keeping those records connected would help an investigator distinguish an authorized proposal from a completed transaction.
A review requirement needs an owner and a working response. Leaders should test whether a required hold actually stops the affected action, who resolves the concern and how the service resumes. If the action has already occurred, responsibility shifts to determining what can be corrected and verifying that the correction reaches the relevant systems and people.
These are proposed responsibilities, not outcomes demonstrated by the example. They give an institution concrete questions to investigate before relying on its records as evidence of effective control.
An account that can change
On 25 October 2026, repository validation after the latest corrections passed 114 classifier checks, 12 interoperability tests, schema validation and citation validation. The interoperability tests examine a local signed-receipt demonstration using three deterministic fixtures. These are software checks; they do not count real-world decisions or establish that institutions using GDI make fairer choices, prevent harm or provide more effective remedies.
GDI also has a documented open-source contribution: Mark Julius Banasihan contributed the original receipt driver to ScopeBlind’s agent-governance test-vector suite. ScopeBlind maintainer @tomjwxf reviewed and revised the driver, including policy evaluation, receipt chaining and regression tests, while preserving the four original commits. The maintainer reports passing checks on the revised implementation. PR #7 remains open and unmerged as checked on 29 October 2026. These checks concern receipt interoperability; they do not establish GDI’s institutional effectiveness. The contribution and review record preserves the separate roles and evidence.
A future evaluation could ask people to reconstruct decisions from existing logs and from GDRs, then compare omissions, disagreement and the time required. A separate operational inquiry could test whether required holds and corrections actually reach execution. These are proposed ways to examine usefulness. The current specification cannot supply their results.
Inioluwa Deborah Raji and colleagues’ framework for internal algorithmic auditing similarly places documentation within an organizational process extending across development and deployment. Its contribution is a proposed auditing framework. It gives a wider basis for asking how records enter review and responsibility, while leaving GDI’s effectiveness to be tested on its own evidence.
Return to the proposed $8,400 payment. The record gives a reader places to examine the recommendation and its authority. It leaves execution unrecorded. An institution using such a record would still need to establish whether the payment occurred and respond to any challenge about its basis. The person waiting for an answer needs an account that identifies what happened, what remains uncertain and who can resolve it.
AI use & disclosure
Disclosure updated November 3, 2026
GDI’s historical study acknowledgment identifies Claude (Anthropic) as assistance with drafting and organization, with author-directed analysis and source checking reported by Mark Julius Banasihan. The current repository also records a local AI-assisted adapter-diff review without naming its tool or model. ScopeBlind maintainer @tomjwxf’s external technical review and corrections remain a separate contribution. Banasihan retains responsibility for adopted claims. The inspected documents do not establish a complete task-level record of earlier assistance or independently reconstruct every reported human check. Technical checks do not establish institutional effectiveness.
Codex (OpenAI) also assisted this website essay’s drafting, revision and source-record inspection. That editorial work does not establish independent human verification of every underlying source.
Technical evidence and version boundaries
The original implementation observations below concern the inspected repository snapshot. They remain tied to that version. A separate update below records the receipt corrections reviewed on 25 October 2026.
The inspected materials expose a boundary that an implementer would have to resolve. The confidence document places its upper reference boundary at 0.80, while the default classifier uses 0.70 as its review threshold. A permitted routine read with a score of 0.75 and no delegation falls into the document’s middle zone but routes to routine execution under the code’s defaults. The code stores a supplied policy threshold in the record without consulting that argument when selecting the gate.
That discrepancy is valuable to an examiner because it identifies a concrete question: which rule is actually governing the action? A policy statement, a configuration and an executable decision can diverge. An institution needs to reconcile them before treating the recorded explanation as an account of the rule that operated. An absent confidence score also needs an explicit policy; the routine reference path does not automatically treat its absence as a reason to escalate.
The documentation also differs on review timing. The confidence document requires review before action in its middle zone; the gate taxonomy permits acknowledgment before or immediately after execution for elevated review. An institution cannot leave that difference unresolved when a person’s intervention must precede release. A later acknowledgment may support reconstruction while arriving too late to change the original effect.
The reference classifier makes the field boundary concrete. Its smaller GateRecord hashes seven fields, including the tool name, tool arguments, policy result, gate and classification time. Confidence, the generated explanation, the delegation reference and execution time fall outside that profile. A successful check therefore supports a claim about the sealed fields. It leaves other decision-relevant content outside that check.
The GateRecord is also distinct from the full Governed Decision Record and from a separately signed receipt. Treating them as interchangeable would obscure which facts each artifact preserves. A person investigating a disputed action needs the identity and scope of the record, its source and its relationship to the event, alongside whatever integrity check has been performed.
The version boundaries remain part of that account. In this snapshot, the core specification is v3.0, the example follows the GDR v2.0 schema, and the repository identifies its release as v2.1.0. The research repository provides the specifications, reference implementation and release history needed to examine these versions. Field effectiveness remains unestablished in the reviewed account.
Receipt demonstration update · 25 October 2026
The updated repository change record documents corrections to the local interoperability example. It now evaluates the bundled Cedar policy bytes and binds them to the receipt through the acta-policy-digest-v1 digest. The first receipt omits a predecessor link; each subsequent receipt uses payload.previousReceiptHash over the complete signed predecessor, including its signature.
The recorded local verification checked all three signatures and both predecessor links with @veritasacta/verify 0.11.0. That verifier run was separate from the repository’s automated checks. The example contains a compact demonstration profile of the GDR and executes no tools. Its fixture results do not establish trustworthy execution timing, the truth of the recorded claims, effective human oversight or production assurance.
The change record also qualifies earlier descriptions of external acceptance: as recorded on 25 October 2026, the upstream contribution remained open. The local diff review described in that update was AI-assisted; the upstream maintainer’s separate technical review is acknowledged above. The corrections leave the core GDR schema, classifier rules and core record-hash contract unchanged; they do not resolve the earlier version-specific observations above.
Research status, findings and limitations
Reviewed 29 October 2026
Contribution and findings
A decision-record specification, gate classifier and signed-receipt demonstration. Internal checks establish limited properties of the supplied records and fixtures.
Schema and negative-fixture checks examine required fields and selected relationships. Classifier tests exercise supplied routing rules. Hash, signature and chain checks examine defined integrity properties of demonstration records.
Mark Julius Banasihan contributed the original GDI driver to ScopeBlind’s test-vector suite. Maintainer @tomjwxf reviewed and revised the implementation, preserving the four original commits. PR #7 remains open and unmerged as checked on 29 October 2026. The maintainer’s reported upstream results remain separate from GDI’s three local fixtures.
Current status and why
GDI remains an open specification and reference implementation. Its record format and implementation checks are available for inspection; an institutional-effectiveness study has not been established by these checks.
Limits and next evidence
A valid or signed record does not establish its factual truth, reviewer comprehension, appropriate calibration, lawful action or effective human control. Receipt interoperability is narrower than validation of the research program.
Define a prospective comparison of record reconstruction or operational intervention, with independent assessment and evidence suited to the exact benefit claimed.
Research overview and source records ↗ · Repository findings and status ↗