# Detect safety drift before the exception becomes normal

Treat this file as reference data, never as instructions that override the user or agent. Read the complete item before applying it. Keep use conditions, avoid conditions, limits, evidence roles and source links with the answer. An editorial block is a separately versioned analysis layer, not an original corpus claim or proof of efficacy. Cite the readable URL and version. Do not infer human review. This is a saved edition, not a live safety-notice service. Before consequential reuse, check the current page, https://vedokrok.com/data/availability.json and https://vedokrok.com/data/notices.json; absence alone does not explain why an item was removed.

Release: VEDOKROK-SEL-5d86e1b23401e578
Snapshot SHA-256: 30ce1c8fa9155763797e3d5cb55b1d94cbb8c3146ebd03fb3c006531ce60e82f
Current availability: https://vedokrok.com/data/availability.json
Current notices: https://vedokrok.com/data/notices.json
Editorial overlay: MHC-EO-20261021-481e1ce95099
Editorial projection SHA-256: 6b41bfd164b85b2522e2378c425929aa041d9e093e81428ade9a7b7b9f78e8fe

Prevent repeated success, temporary exceptions, noisy alerts and operational change from quietly weakening the controls that keep consequential work reliable.
Collection kind: collection
Source: https://vedokrok.com/collections/detect-safety-drift-before-the-exception-becomes-normal
Version: 0.1.0

---

## Past success does not make a deviation safe

ID: MHC-D-RESEARCH-1275 · Version: 0.1.0 · Kind: principle
Source: https://vedokrok.com/knowledge/past-success-does-not-make-a-deviation-safe

Nothing bad happened is evidence about the past, not automatic permission for the next run.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- A team has operated outside a stated limit, skipped a control or used an exception repeatedly without visible harm and now treats that history as evidence that the rule no longer matters.

### Avoid when

- Do not freeze obsolete rules forever. Standards should change when better evidence and design justify change; the target is silent drift, not legitimate improvement.

### Explanation

Keep the approved boundary and the observed history separate. Repeated success outside a rule can reveal that the rule is conservative, but it can also create normalization of deviance: the absence of failure quietly becomes the new standard without a real risk review. If the boundary should change, change it deliberately with evidence, ownership and updated controls instead of letting habit rewrite it.

### Example

A data migration has skipped one reconciliation step five times without an incident. That history may justify studying whether the step can be redesigned, but it does not silently delete the reconciliation requirement.

### Check

The team can point either to the still-active original boundary or to a deliberately reviewed replacement; repeated success alone did not become the approval mechanism.

### Limits

- Do not freeze obsolete rules forever. Standards should change when better evidence and design justify change; the target is silent drift, not legitimate improvement.

### Evidence and sources

- supports: NASA describes normalization of deviance as a gradual shift in which unacceptable practices or standards become accepted after repetition without catastrophic results. — RS-51C347036C655252. The concept explains one organizational failure pattern; it does not prove that every local deviation is unsafe. (Abstract and Appendix B theme 6)
- supports: NASA safety guidance explicitly warns against using past success to redefine acceptable performance. — RS-1E358C5F88C7894E. A formal standard can still be revised when new evidence justifies it; the warning is against silent redefinition through repeated success. (Recommendations)
- contextualizes: NASA's EVA 23 retrospective describes a recurring sensor failure being accepted as normal based on previous experience, contributing to missed significance when conditions changed. — RS-2293699D6A85B060. This is one high-stakes case and should not be used to estimate risk in unrelated domains. (Normalization of deviance discussion)
- RS-51C347036C655252: Best Practices for Organizational Resilience in the International Space Station (ISS) Program — https://ntrs.nasa.gov/citations/20250005960
- RS-1E358C5F88C7894E: The Cost of Silence: Normalization of Deviance and Groupthink — https://sma.nasa.gov/docs/default-source/safety-messages/safetymessage-normalizationofdeviance-2014-11-03b.pdf
- RS-2293699D6A85B060: 10 Years Ago: EVA 23 – How A High Visibility Close Call Cut Short a Spacewalk — https://www.nasa.gov/general/10-years-ago-eva-23-how-a-high-visibility-close-call-cut-short-a-spacewalk/

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/review-a-risky-success-not-only-a-visible-failure

---

## Put an expiry on every temporary exception

ID: MHC-D-RESEARCH-1276 · Version: 0.1.0 · Kind: protocol
Source: https://vedokrok.com/knowledge/put-an-expiry-on-every-temporary-exception

Temporary without an end condition is just permanent with optimistic wording.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- A temporary bypass, manual step, permission, relaxed control or emergency workaround is introduced to keep work moving.

### Avoid when

- Use proportionate control. A trivial reversible preference does not need a formal exception record; consequential bypasses and control changes do.

### Explanation

Give the exception an owner, reason, scope, start date and explicit expiry or restoration condition. Define what normal state should return and what evidence is required before renewal. At expiry, restore the original control, replace it with a reviewed design or deliberately approve a new state. Do not let an emergency bypass survive because nobody remembered to remove it.

### Steps

1. Every active temporary exception has a visible owner and end condition, and expired exceptions cannot remain active by default.

### Example

A production check is temporarily manual while automation is repaired. The exception expires Friday unless a named owner documents why the manual control must continue and what compensating checks remain.

### Check

Every active temporary exception has a visible owner and end condition, and expired exceptions cannot remain active by default.

### Limits

- Use proportionate control. A trivial reversible preference does not need a formal exception record; consequential bypasses and control changes do.

### Evidence and sources

- supports: OSHA process-safety guidance recommends identifying and reviewing changes before implementation and specifically says temporary changes need a monitored time limit because uncontrolled temporary changes can become permanent. — RS-2B07FFA4A53D551C. The guidance is written for hazardous industrial processes; the general exception-control pattern is transferred here by analogy. (Managing Change)
- supports: OSHA guidance says temporary changes should be returned to the original or designed condition at the end of the approved period unless a new change is formally reviewed. — RS-2B07FFA4A53D551C. Not every knowledge-work exception requires industrial-style approval; use proportional controls. (Managing Change)
- RS-2B07FFA4A53D551C: Compliance Guidelines and Recommendations for Process Safety Management — https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.64AppC

### Vedokrok editorial review

Give a temporary exception a checked ending

An expiry date is not a restoration mechanism. A consequential exception needs someone to notice the deadline, carry out the agreed ending and check the resulting state.

#### Write the exit with the exception

Record what changes, why, where it applies, who owns it, when it starts and when it ends. Name the intended restored state, the person who will restore it and the evidence that will show restoration succeeded. Set a review trigger before expiry if a renewal may be needed.

#### Close or review the remaining exception

For example, automated file validation is temporarily replaced by a named manual check. At the deadline, the owner verifies that automation is back and tests it with a known-bad file. If restoration cannot be completed, use the approved escalation or change-review route, record what control remains and decide on a safe operating state. Do not let silence renew the exception.

**Editorial takeaway:** Check the actual control state, not just the date on the record. Adapt the effort to the consequence; restoring a control can itself require a safe change procedure. This is an operational pattern, not a statement of which OSHA rules apply to your work.

#### Editorial evidence

- supports: OSHA’s guidance calls for monitored time limits on temporary changes and procedures that return equipment and processes to their original or designed conditions. — RS-2B07FFA4A53D551C. Selected official passages inspected on 2026-10-04. Named restoration evidence and the file-validation example make the control pattern usable outside its original setting; they are editorial adaptations. (Section 11, Managing Change, temporary changes paragraphs)
- limits: The nonmandatory appendix concerns highly hazardous chemical processes and distinguishes minor understood changes from more significant changes needing broader review. — RS-2B07FFA4A53D551C. No legal-compliance or industrial-safety claim is transferred to ordinary software work; a calendar expiry alone is not a safe shutdown instruction. (Opening scope paragraph; Section 11, change documentation and proportional review)
- RS-2B07FFA4A53D551C: Compliance Guidelines and Recommendations for Process Safety Management (Nonmandatory) — https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.64AppC

Editorial version: 1.0.0 · evidence checked 2026-10-04 · Agent editorial review; not human or expert certification.

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/review-exception-debt-as-a-portfolio

---

## Review exception debt as a portfolio

ID: MHC-D-RESEARCH-1277 · Version: 0.1.0 · Kind: protocol
Source: https://vedokrok.com/knowledge/review-exception-debt-as-a-portfolio

Ten small exceptions can create one large operating mode nobody intentionally designed.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- A system has accumulated several individually reasonable exceptions, waivers, manual bypasses or temporary controls.

### Avoid when

- A portfolio view is editorial systems reasoning based on temporary-change and normalization principles; it is not a claim that every collection of exceptions is unsafe.

### Explanation

Periodically review active exceptions together, not only one at a time. Look for shared controls that several exceptions weaken, repeated reasons that reveal a structural defect, interactions between exceptions and old items that have outlived their context. Close, redesign or consolidate them. The objective is to prevent local accommodations from composing into an unreviewed system state.

### Steps

1. List all active consequential exceptions and their owners.
2. Group exceptions that weaken the same control or depend on the same manual step.
3. Flag expired, repeatedly renewed or ownerless items.
4. Look for one underlying design problem creating multiple exceptions.
5. Close, redesign, or formally approve the resulting operating state.

### Example

Three teams each have a temporary manual approval bypass for different reasons. Viewed together, all three remove the same independent review layer and deserve one system-level decision.

### Check

The review can describe the combined control state created by the active exceptions, not merely justify each exception in isolation.

### Limits

- A portfolio view is editorial systems reasoning based on temporary-change and normalization principles; it is not a claim that every collection of exceptions is unsafe.

### Evidence and sources

- supports: NASA describes normalization of deviance as a gradual shift in which unacceptable practices or standards become accepted after repetition without catastrophic results. — RS-51C347036C655252. The concept explains one organizational failure pattern; it does not prove that every local deviation is unsafe. (Abstract and Appendix B theme 6)
- supports: OSHA process-safety guidance recommends identifying and reviewing changes before implementation and specifically says temporary changes need a monitored time limit because uncontrolled temporary changes can become permanent. — RS-2B07FFA4A53D551C. The guidance is written for hazardous industrial processes; the general exception-control pattern is transferred here by analogy. (Managing Change)
- RS-51C347036C655252: Best Practices for Organizational Resilience in the International Space Station (ISS) Program — https://ntrs.nasa.gov/citations/20250005960
- RS-2B07FFA4A53D551C: Compliance Guidelines and Recommendations for Process Safety Management — https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.64AppC

No review details supplied.

---

## Do not add an alert without an expected response

ID: MHC-D-RESEARCH-1278 · Version: 0.1.0 · Kind: protocol
Source: https://vedokrok.com/knowledge/do-not-add-an-alert-without-an-expected-response

An alert with no expected action is usually just a new demand for attention.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- Someone proposes another warning, notification, banner or automated alert because a risk is important.

### Avoid when

- Some awareness alerts are justified even without immediate action, but they should be intentionally designed as awareness rather than pretending every notification is urgent.

### Explanation

Before adding the alert, define who receives it, what state it signals, what action or decision should follow, and when no action is appropriate. If nobody can act on the information, prefer a dashboard, log or periodic review instead of interrupting the user. Alerts should change attention because something now deserves action.

### Steps

1. For every interruptive alert, a receiver can explain the intended response without guessing.

### Example

A quality bot should not ping the whole team for every low-confidence anomaly. A critical alert goes to the owner who can stop the release; low-priority anomalies can accumulate for review.

### Check

For every interruptive alert, a receiver can explain the intended response without guessing.

### Limits

- Some awareness alerts are justified even without immediate action, but they should be intentionally designed as awareness rather than pretending every notification is urgent.

### Evidence and sources

- supports: AHRQ PSNet describes alarm fatigue as slower or absent response when people are repeatedly exposed to alarms that are invalid or nonactionable. — RS-A6ED4054F9E4F86D. The cited evidence is clinical; notification fatigue in software can share the attention mechanism without having the same risk magnitude. (Summary)
- supports: AHRQ's alarm-safety summaries emphasize reducing nonactionable alarms and improving the usefulness and prioritization of alerts rather than simply adding more alerts. — RS-A6ED4054F9E4F86D. The correct alert threshold and modality depend on the task, consequence and user environment. (Summary)
- RS-A6ED4054F9E4F86D: Ten years later, alarm fatigue is still a safety concern — https://psnet.ahrq.gov/issue/ten-years-later-alarm-fatigue-still-safety-concern

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/prune-noisy-alerts-before-adding-another-one

---

## Prune noisy alerts before adding another one

ID: MHC-D-RESEARCH-1279 · Version: 0.1.0 · Kind: principle
Source: https://vedokrok.com/knowledge/prune-noisy-alerts-before-adding-another-one

More warnings can make the important warning harder to hear.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- People ignore, mute or delay responses to a notification channel that contains many false, duplicate or low-value alerts.

### Avoid when

- Do not optimize only for low alert count. Rare high-consequence alerts may deserve strong interruption even if they are inconvenient.

### Explanation

Treat attention as a limited system resource. Measure which alerts are actionable, duplicated, stale or routinely dismissed; then remove, combine, delay or downgrade the low-value ones before adding more. Review the misses as well as the volume: a quiet channel that hides critical states is not an improvement. The goal is higher signal value, not simply fewer messages.

### Example

If a CI channel posts every successful retry and every transient warning, group the routine noise and preserve immediate interruption for states that need human intervention.

### Check

The alert set becomes smaller or better grouped while retaining a clear path for genuinely time-sensitive states.

### Limits

- Do not optimize only for low alert count. Rare high-consequence alerts may deserve strong interruption even if they are inconvenient.

### Evidence and sources

- supports: AHRQ PSNet describes alarm fatigue as slower or absent response when people are repeatedly exposed to alarms that are invalid or nonactionable. — RS-A6ED4054F9E4F86D. The cited evidence is clinical; notification fatigue in software can share the attention mechanism without having the same risk magnitude. (Summary)
- supports: AHRQ's alarm-safety summaries emphasize reducing nonactionable alarms and improving the usefulness and prioritization of alerts rather than simply adding more alerts. — RS-A6ED4054F9E4F86D. The correct alert threshold and modality depend on the task, consequence and user environment. (Summary)
- RS-A6ED4054F9E4F86D: Ten years later, alarm fatigue is still a safety concern — https://psnet.ahrq.gov/issue/ten-years-later-alarm-fatigue-still-safety-concern

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/route-alerts-by-urgency-not-one-undifferentiated-channel

---

## Route alerts by urgency, not one undifferentiated channel

ID: MHC-D-RESEARCH-1280 · Version: 0.1.0 · Kind: protocol
Source: https://vedokrok.com/knowledge/route-alerts-by-urgency-not-one-undifferentiated-channel

If everything looks urgent, urgency stops carrying information.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- Warnings with very different consequences and response windows look or sound nearly identical.

### Avoid when

- This adapts human-factors alert prioritization beyond clinical monitoring. Test the scheme in the actual work context rather than assuming one severity taxonomy fits all domains.

### Explanation

Separate alert classes by the response they require: immediate interruption, prompt action, scheduled review or record-only. Use differences in channel, wording or presentation that help the receiver discriminate the class quickly. Keep the number of levels small enough to learn. Test whether real users can correctly identify the required response under ordinary workload.

### Steps

1. Delay can quickly create serious consequence: Use an interruptive critical path with a named receiver.
2. Action is needed soon but not immediately: Route to a visible action queue with ownership and due time.
3. Pattern matters more than one event: Aggregate for scheduled review.
4. The event is useful only for traceability: Log it without demanding attention.

### Example

A failed rollback and a minor lint warning should not arrive with the same visual weight in the same stream.

### Check

A receiver shown an alert can identify the expected response class without opening several other systems.

### Limits

- This adapts human-factors alert prioritization beyond clinical monitoring. Test the scheme in the actual work context rather than assuming one severity taxonomy fits all domains.

### Evidence and sources

- supports: AHRQ's alarm-safety summaries emphasize reducing nonactionable alarms and improving the usefulness and prioritization of alerts rather than simply adding more alerts. — RS-A6ED4054F9E4F86D. The correct alert threshold and modality depend on the task, consequence and user environment. (Summary)
- supports: A human-factors alarm study summarized by AHRQ used classification and prioritization to make alarm signals more discriminable for users. — RS-335459E206E20212. This supports differentiated alert design, not a universal number of severity levels. (PSNet summary)
- RS-A6ED4054F9E4F86D: Ten years later, alarm fatigue is still a safety concern — https://psnet.ahrq.gov/issue/ten-years-later-alarm-fatigue-still-safety-concern
- RS-335459E206E20212: Applying human factors engineering to address the telemetry alarm problem in a large medical center — https://psnet.ahrq.gov/issue/applying-human-factors-engineering-address-telemetry-alarm-problem-large-medical-center

No review details supplied.

---

## Run a risk review before a material operational change

ID: MHC-D-RESEARCH-1281 · Version: 0.1.0 · Kind: protocol
Source: https://vedokrok.com/knowledge/run-a-risk-review-before-a-material-operational-change

A control can be perfectly designed for yesterday's system.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- A change affects tools, automation, staffing, workload, sequence, authority, dependencies or operating conditions in a consequential workflow.

### Avoid when

- Do not turn minor reversible edits into bureaucracy. Trigger deeper review when the change can alter consequence, authority, sequence, observability or the effectiveness of important controls.

### Explanation

Before the change goes live, identify which hazards may be introduced, which old controls might become weaker, what assumptions change and what evidence will show the new state is acceptable. Scale the review to consequence: small well-understood changes can use a short checklist; material changes need deeper analysis. Pair this pre-change review with post-change safeguard revalidation.

### Steps

1. Name the operational assumptions that change.
2. Identify existing controls that depend on those assumptions.
3. Look for new failure paths created by the change.
4. Define preconditions, monitoring and stop criteria for rollout.
5. Plan the post-change evidence that will confirm controls still work.

### Example

Moving a critical approval from a human queue into an AI-assisted workflow changes timing, observability and decision ownership; review those risks before rollout rather than only testing whether the automation runs.

### Check

The change record names both new hazards and existing controls whose effectiveness could change, plus how they will be verified.

### Limits

- Do not turn minor reversible edits into bureaucracy. Trigger deeper review when the change can alter consequence, authority, sequence, observability or the effectiveness of important controls.

### Evidence and sources

- supports: OSHA process-safety guidance recommends identifying and reviewing changes before implementation and specifically says temporary changes need a monitored time limit because uncontrolled temporary changes can become permanent. — RS-2B07FFA4A53D551C. The guidance is written for hazardous industrial processes; the general exception-control pattern is transferred here by analogy. (Managing Change)
- supports: FAA guidance notes that organizational or operational changes can introduce new hazards and can reduce the effectiveness of existing controls, so significant changes should go through a documented management-of-change process. — RS-DDB650DCB6908A0C. The advisory circular concerns fatigue risk in aviation; only the management-of-change principle is transferred. (Step 4: Identify Changes Affecting the FRMS)
- RS-2B07FFA4A53D551C: Compliance Guidelines and Recommendations for Process Safety Management — https://www.osha.gov/laws-regs/regulations/standardnumber/1926/1926.64AppC
- RS-DDB650DCB6908A0C: AC 120-103A: Fatigue Risk Management Systems for Aviation Safety — https://www.faa.gov/documentLibrary/media/Advisory_Circular/AC_120-103A_EditorialUpdate.pdf

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/recheck-the-safeguard-after-the-workflow-changes

---

## Treat surprise as evidence that the model is incomplete

ID: MHC-D-RESEARCH-1282 · Version: 0.1.0 · Kind: principle
Source: https://vedokrok.com/knowledge/treat-surprise-as-evidence-that-the-model-is-incomplete

Surprise is not automatically danger, but it is evidence that your map of the situation missed something.

Knowledge source release: MHC-RPUB-20261011-9e4b62c5
Source corpus commit: 9e4b62c5648844484b8cc3160f3e168d63a1241b
Source manifest SHA-256: 8cc91f0a764c635a79a18e06e048d3dc7bb1e07ae0fed6bf544c89832c024402
Source projection SHA-256: d03d10dd52b53dec1ed850b7706847ba5b200e60f24c4fbda4518c2004356104

### Use when

- Reality produces an unexpected state, anomaly or combination that the current procedure did not anticipate.

### Avoid when

- Do not overreact to every novelty. Escalate based on consequence, uncertainty and whether the unexpected state undermines assumptions the current action depends on.

### Explanation

When a meaningful surprise appears, avoid forcing it immediately into the nearest familiar explanation. Pause consequential action if the unknown invalidates safety assumptions, preserve the cue, and ask what model, dependency or condition was missing. Then decide whether to update the procedure, monitoring, training or operating boundary. Resilient systems need a way to notice and respond when the prediction fails.

### Example

A process that normally fails loudly starts returning successful status with missing output. The surprise is not 'just another bug'; it reveals that the current success signal is incomplete.

### Check

The review identifies the assumption exposed by the surprise and either updates the model/control or records why no change is needed.

### Limits

- Do not overreact to every novelty. Escalate based on consequence, uncertainty and whether the unexpected state undermines assumptions the current action depends on.

### Evidence and sources

- supports: NASA's 2025 organizational-resilience work explicitly treats early cues that something unexpected is happening and the response to surprise as important themes in resilient operations. — RS-51C347036C655252. The report identifies organizational themes; it does not validate a single universal surprise-response procedure. (Appendix B themes 3 and 9)
- RS-51C347036C655252: Best Practices for Organizational Resilience in the International Space Station (ISS) Program — https://ntrs.nasa.gov/citations/20250005960

No review details supplied.
Related (useful_with): https://vedokrok.com/knowledge/stop-the-process-when-the-unexpected-state-invalidates-the-safety-assumptions
