What Should a Judgment Deferral Log Look Like? — The Structure That Audit Trails Cannot Record – 궁리연구소

Abstract

When a judgment is deferred, most systems do not record that deferral. Audit trails are designed to capture actions, and deferral is not an action — it is a state. This document defines the eight minimum components a judgment deferral log must contain and describes the structural difference between audit trails and deferral logs. It compares the Columbia space shuttle disaster (2003), where an unrecorded deferral contributed to catastrophic failure, with medical Active Surveillance protocols, where structured deferral recording has demonstrated equivalent survival outcomes to immediate treatment. The document declares the structural bifurcation between deferral (Suspend) and delegation (Delegate), and addresses only deferral logs.

Keywords: Judgment Deferral Log, Judgment Gap, Audit Trail, Record Specification, Deferral-Delegation Bifurcation, Accountability, Pre-Judgment Validation, Latent Failure, Active Surveillance, Columbia Disaster, Near-Miss Reporting, EU AI Act

This document does not provide conclusions or recommendations. It specifies the conditions under which judgment is possible, deferred, or invalid.

Definitions

Term	Definition	Source
Judgment Deferral Log	A structured record of the moment, reason, unmet conditions, and accountability state when judgment is deferred	Gungri Research Judgment Theory
Judgment Gap	The temporal interval between the declaration of deferral and the resumption of judgment	Gungri Research Judgment Theory
Audit Trail	A system that records decision actions (approval, rejection, modification) along with the responsible party and timestamp	General usage
Deferral (Suspend)	Halting judgment due to unmet conditions while the judgment subject remains unchanged	Gungri Research Judgment Theory
Delegation (Delegate)	Transferring judgment ownership to an external party	Gungri Research Judgment Theory
Sealed Information	Information that existed at the moment of judgment but was intentionally excluded from the judgment scope. Example: unverified intelligence regarding a competitor acquisition that was obtained at the decision point but excluded due to unconfirmed source reliability, or a regulatory draft intentionally excluded from the judgment basis because legal review had not been completed	Gungri Research Judgment Theory
Latent Failure	Unrecorded states that accumulate within a system until they form the conditions for an accident	Reason (1997)

A judgment was deferred. The person responsible decided: “I cannot make this decision right now.” That decision may have been entirely rational. But six months later, when the audit team reviews the case, the record contains nothing. No approval. No rejection. No deferral. From the system’s perspective, nothing happened at that moment.

This is the recording problem of judgment deferral. Deferral does not mean judgment has disappeared. It means judgment has been suspended because one or more conditions for sound judgment were not met. But most recording systems are designed to capture actions — and deferral is not an action. It is a state. An unrecorded deferral is, in retrospect, indistinguishable from the absence of judgment entirely.

What an audit trail records is clear. Who approved. When they approved. Which outcome was selected. This is an action record, and it is essential for organizational accountability.

But there are things an audit trail does not record. The fact that a judgment was deferred. The reason for deferral. Which conditions were unmet at the moment of deferral. Who held responsibility during the deferral period. What conditions would need to be satisfied before judgment could resume.

An audit trail records what happened. A judgment deferral log records why something has not happened yet. The objects of recording are different, the timing of recording is different, and the information recoverable in post-hoc review is different.

This is not a theoretical distinction.

The 2003 Columbia space shuttle disaster demonstrates what happens when a judgment deferral goes unrecorded. Eighty-two seconds after launch, a piece of insulating foam separated from the external fuel tank and struck the shuttle’s left wing. Engineers could not determine whether this had caused structural damage. A judgment was made that further analysis was needed — this was, in structural terms, a judgment deferral.

But this deferral was never registered as such in the formal recording system. The unmet condition — the absence of high-resolution satellite imaging data — was not explicitly documented. The resumption criterion — reassessment after obtaining imaging data — was never formalized. When management requested a status report after orbital insertion, the record reflected a state equivalent to “no concerns identified.” The deferral was treated as absence.

The Columbia Accident Investigation Board (CAIB) report identified this structural failure explicitly. The report described a process by which the organization normalized deviation signals, and identified as a core problem the fact that engineers’ concerns were not maintained as visible states within the formal decision-making pathway. The audit trail was complete. Every meeting was recorded. Every email was preserved. But no structure existed to record “judgment has been deferred” as an independent item.

Seven crew members died. The records recovered after the disaster showed who attended which meetings. But whether a state of “we cannot yet judge” was declared in those meetings — and if so, under what conditions that deferral was overridden without resolution — had to be reconstructed after the fact. It required an investigation, not a record review.

A counter-example also exists. There are domains where structured judgment deferral is recorded and tracked within the system.

In medicine, Active Surveillance is a protocol for low-risk prostate cancer in which immediate treatment is deferred in favor of regular monitoring. This is not “choosing not to treat.” It is structurally deferring the treatment judgment.

The protocol works because deferral is designed into the recording framework. Diagnostic indicators at the point of deferral — PSA levels, Gleason scores, biopsy results — are recorded. The reason for deferral is specified: “Current tumor progression rate does not meet the risk-benefit threshold for immediate intervention.” Resumption criteria are defined in advance: treatment resumes if PSA doubling time falls below three years, if the Gleason score increases, or if MRI findings change. A monitoring schedule for the deferral period — PSA testing every three to six months, biopsy every twelve to twenty-four months — is established.

The ProtecT trial (Hamdy et al., 2016) randomly assigned 1,643 men with localized prostate cancer to active monitoring, surgery, or radiotherapy and followed them for ten years. The result is notable: prostate-cancer-specific mortality at ten years was approximately 1 percent across all three groups, with no statistically significant difference. Structured deferral did not worsen survival outcomes compared to immediate treatment.

The critical point is this: the same outcome — “no treatment administered” — is interpreted entirely differently depending on the recording structure. Under an active surveillance protocol, “no treatment” is a recorded judgment. Without such a protocol, “no treatment” due to missed follow-up is an unrecorded absence. In medical malpractice litigation, the legal consequences of these two situations are fundamentally different.

For a judgment deferral to be recorded in a state that permits post-hoc review, a minimum set of components is required.

Timestamp — the starting point of the judgment gap. Reason — the specific basis for deferral. Unmet Conditions — which of the conditions required for sound judgment were not satisfied. Scope — the boundaries of the judgment subject at the moment of deferral. Available Information — the inventory of information that had been secured at the time of deferral. Sealed Information — information that existed at the time of deferral but was intentionally excluded, along with the reason for exclusion. Accountability State — the responsible party and their authority status at the moment of deferral. Resumption Criteria — the threshold conditions that must be met before judgment can resume.

Of these eight components, an audit trail can capture only one: the timestamp. The remaining seven fall outside the design scope of audit trails. (This component structure is based on a proprietary judgment condition analysis framework developed by Gungri Research. The detailed structure of that framework is not included in this publication.)

A structural bifurcation needs to be made explicit here. There are two fundamentally different situations in which a judgment is not executed.

Deferral (Suspend) means the judgment subject does not change. It is a declaration: “I cannot judge this right now.” One or more conditions for sound judgment are unmet, and the same subject will resume judgment when conditions are satisfied. Ownership of the judgment remains with the person who deferred it.

Delegation means the judgment subject transfers. It is a declaration: “I am handing this judgment to another party.” Ownership of the judgment moves to an external entity — a superior, a committee, a specialist, or an algorithm. The reasons for delegation vary. Some delegation is healthy — driven by the need for specialized expertise. Some is pathological — driven by avoidance of responsibility. Some is coerced — imposed by power structures.

These two situations require fundamentally different recording structures. A deferral log needs to record which conditions are unmet and what the resumption criteria are. A delegation log additionally requires: to whom the judgment was transferred, the structural reason for the transfer, and the conditions under which ownership would return to the original subject. This document addresses deferral logs. Delegation requires a structurally different recording specification and is a separate subject of analysis.

The period between the moment judgment is deferred and the moment it resumes is called the Judgment Gap. This interval is a blind spot in the record.

Four problems arise within the gap. New information enters during the deferral period but is not recorded. The responsible party changes during the deferral period but the transition point is not recorded. The moment deferral conditions are resolved does not coincide with the moment judgment actually resumes. And judgment is forced through before the deferral conditions have been resolved.

What occurred in the Columbia disaster was precisely the fourth situation. The deferral — further analysis needed — was overridden before it was resolved. Because the deferral itself was not recorded, whether an override even occurred could not be identified after the fact. Reason (1997) calls such structures latent failures: not individual incidents, but unrecorded states that accumulate within a system until they form the conditions for an accident.

Why does this matter now? Three reasons.

First, existing systems are not designed to record “not judging.” Most decision-making systems record actions — approvals, rejections, modifications. Because deferral is a state rather than an action, it is structurally excluded from the recording scope. This is not a defect in any individual system. It is a categorical limitation in how recording systems are designed.

Second, the spread of AI-assisted decision-making has made structured deferral recording more urgent. When an AI system produces a recommendation and a human defers — “I cannot decide this now” — the absence of a deferral record means the system cannot distinguish between a human who ignored the AI recommendation, a human who never reviewed it, and a human who consciously deferred judgment. The “effective human oversight” required by Article 14 of the EU AI Act should encompass not only approvals but also deferrals and rejections. However, specific implementation criteria for this have not yet been established.

Third, the accumulation of deferral logs is a precondition for organizational learning. When deferrals occur repeatedly under the same judgment conditions, the pattern points not to individual capability problems but to systemic design failures in the conditions themselves. This pattern is identifiable only when deferrals are recorded. Aviation’s Near-Miss Reporting systems operate on this principle. Barach and Small (2000), in an analysis applying aviation near-miss reporting to medical settings, showed that the key mechanism contributing to accident rate reduction was not “recording incidents” but “recording states that did not yet become incidents.” A judgment deferral log applies the same principle to the domain of decision-making.

One question cuts across all three reasons. Can your organization’s recording system currently distinguish between ‘judgment failure’ and ‘judgment deferral’? If it cannot, the recording system is treating deferral as indistinguishable from failure.

The accountability structure is worth stating plainly. The party ultimately responsible is the decision-maker who resumes and executes the judgment.

But when a judgment deferral log exists, additional facts become verifiable. Who held responsibility at the point of deferral. Whether responsibility was transferred during the deferral period. Whether the resumption criteria were satisfied at the point of resumption. Without a log, none of these three facts can be confirmed. Accountability assignment remains possible, but the evidentiary basis for that assignment does not exist.

When an AI system recommends deferral or signals that deferral conditions have been resolved, the relationship between that system’s output and the human’s decision to resume judgment is also a subject for the log. Without this record, AI-prompted resumption and independent human resumption cannot be distinguished.

A judgment rendered atop an unrecorded deferral is, from the recording system’s perspective, indistinguishable from a judgment executed after confirming that all conditions were met. The responsibility for this indistinguishability lies not with the individual who executed the judgment, but with the organization that failed to design a deferral recording structure.

About Gungri Research Lab

This essay presents a record specification analysis from the Gungri Judgment Theory research program. The full research document — What Should a Judgment Deferral Log Look Like? — The Structure That Audit Trails Cannot Record (GRL-TV-001-EN) — contains the complete deferral log component definitions, audit trail comparison analysis, judgment gap recording problems, deferral-delegation bifurcation structure, accountability analysis, and full citations.

Related Literature:

Columbia Accident Investigation Board. (2003). Report of the Columbia Accident Investigation Board, Volume I. NASA.
Hamdy, F. C., et al. (2016). 10-Year Outcomes after Monitoring, Surgery, or Radiotherapy for Localized Prostate Cancer. New England Journal of Medicine, 375(15), 1415–1424.
Reason, J. (1997). Managing the Risks of Organizational Accidents. Ashgate.
Barach, P., & Small, S. D. (2000). Reporting and preventing medical mishaps: lessons from non-medical near miss reporting systems. BMJ, 320(7237), 759–763.
European Parliament & Council. (2024). Regulation (EU) 2024/1689 (AI Act). Article 14.
International Organization for Standardization. (2023). ISO 42001:2023 — Artificial intelligence — Management system.

Limitations

This document does not provide operational guidelines for specific industries or organizations.
This document does not address the technical implementation of judgment deferral logs (software design, database schemas, etc.).
This document does not address the recording specification for delegation (Delegate). Delegation is a separate subject of analysis.
This document does not provide interpretations of specific laws or regulations.

FAQ

Q1. How does a judgment deferral log differ from an audit trail?

An audit trail records actions (who approved what, when). A judgment deferral log records states (why judgment has not yet occurred). The recording object, timing, and recoverable information differ structurally.

Q2. What are the minimum components of a judgment deferral log?

Eight components: timestamp, reason for deferral, unmet conditions, judgment scope, available information, sealed information, accountability state, and resumption criteria. Of these, only the timestamp can be captured by existing audit trails.

Q3. What is the difference between deferral (Suspend) and delegation (Delegate)?

Deferral means the judgment subject does not change — “I cannot judge this now,” with the same person resuming when conditions are met. Delegation means the judgment subject transfers to an external party. The two require fundamentally different recording structures. This document addresses only deferral logs.

Q4. What might have been different in the Columbia disaster if a judgment deferral log had existed?

If the deferral had been registered in the formal recording system, the state “further analysis needed” would have been visible as an unresolved item when the re-entry decision was made. The CAIB report identified the failure to maintain engineers’ concerns as visible states in the formal decision pathway as a core structural failure.

Q5. Why is the Judgment Gap problematic?

During the interval between deferral and resumption, new information may enter, the responsible party may change, or judgment may be forced through before deferral conditions are resolved. Without a deferral log, none of these events can be identified after the fact.

Terminology Attribution

The following terms are proprietary concepts defined within the Gungri Research Judgment Theory framework: Judgment Deferral (HOLD), Judgment Gap, Judgment Deferral Log, Sealed Information, Judgment State (READY / HOLD / NOT READY), Pre-Judgment Validation.

Citation Format

Gungri Research Lab. (2026). “What Should a Judgment Deferral Log Look Like? — The Structure That Audit Trails Cannot Record” GRL-TV-001-EN.

License

This document is distributed under the CC BY-NC-ND 4.0 International License. Non-commercial sharing of the original text is permitted. Modifications and derivative works are not allowed.

이 글은 결론이나 판단을 제공하지 않으며, 판단이 가능한 조건과 유예 상태를 구조적으로 설명한다.

This document does not provide conclusions or recommendations. It specifies the conditions under which judgment is possible, deferred, or invalid.