Repair & Reuse · Question 02
Exception radar
Which deviations need management attention now, and which are routine noise?
Resolve the safety sign-off, verify the equipment recovery plan and decide the unapproved £8,000 option. Keep the other open records with their routine owners. Older information needs refreshing; age alone is not evidence of a new operational exception.
Fictional what-if workspace. Changes stay in this tab and reset on reload; no record is saved and no external system is changed.
Three management conversations
1 · Protect the pilot gate
SA-04 · Assurance lead. Sign-off is outstanding. Resolve before the pilot gate; a score threshold cannot clear this safety dependency.
2 · Test the delivery recovery
EQ-12 · Supply lead. A two-week equipment delay may push the whole launch to week 14. Confirm the backup and delivery date before treating recovery as credible.
3 · Preserve the spending boundary
BU-07 · Finance lead. Decide whether to approve or remove the £8,000 option through the appropriate authority. Budget headroom is not spending permission.
Why the combined schedule and cost exposure matters
The following is a planning illustration, separate from the supplied anomaly labels. It assumes all six staff do fit-out after design, the three downstream stages run sequentially, the delay is additional to the four-week equipment lead, and all six staff remain charged throughout the launch.
The record does not specify delay recovery, staffing changes, stage overlap or safety completion dates. This is an exposure scenario, not a forecast. The contingency is already included; it is not added again as spare funding.
Tune the radar, then inspect the trade-off
Score is a supplied priority signal, not a probability. The default threshold of 3 is an authored policy: it separates the three high-score deviations in this small fixture. The unresolved safety gate stays visible at every threshold.
Five highest-priority records
Open records only: safety gate first, then score descending, then newest day. This ordering does not use the fixture’s anomaly truth labels. The top five can include routine work.
What the rule catches—and misses
Evaluate the operational flag against the fixture’s immutable anomaly label. The denominator is always all seven supplied observations: three true anomalies and four non-anomalies. Six are open; one is closed. There is no sampling, hidden exclusion or claim of out-of-sample accuracy.
| Radar outcome | Fixture anomaly = true (3) | Fixture anomaly = false (4) |
|---|---|---|
| Management attention | True positives | False positives |
| Routine or closed | False negatives | True negatives |
Compare threshold policies on the same seven records
Every row retains the safety gate and closed-record exclusion. These are descriptive fixture results; picking a rule on this tiny labelled set cannot validate it for future observations.
| Threshold | Flagged | TP / FP | FN / TN | Inspect |
|---|
All records, including routine noise
Routine means local handling and monitoring under the present rule. One shift swap, one low-throughput sample and an early invoice are not evidence of a wider problem. Repeated changes, missing cover or new impact could change that assessment. No time series or control limits were supplied.
| ID | Observation | Score / truth | Current routing | Source day / age | Evidence |
|---|
Simulate a management triage brief
Turn the applied radar into an in-page list of decisions, owners and evidence-refresh needs. This creates a temporary preview only. It sends nothing, approves nothing and does not change a record.
No simulated brief yet. Adjust the controls, apply them, then generate a preview.
What this evidence can support
- Seven synthetic observations, one snapshot. Perfect classification at the default threshold is a property of these fixtures, not evidence of improved detection or production reliability.
- Policy and truth are different. A false positive against the fixture label may still deserve a routine operational check. The author’s safety override is explicit and does not change the truth label.
- Age is incomplete evidence. Moving the as-of day ages the same records; it does not simulate new observation, prove resolution or verify their freshness.
- No real authorisation. Owners and actions are supplied suggestions. Safety assurance and spending approval remain outside this workspace.
Authorship and test boundary
Model: gpt-6-astra; reasoning effort: ultra; date: 25 October 2026. This answer was produced by an independent assigned worker, using only question 02’s JSON input and the common run brief for substantive content. Tools: local file authoring and local JavaScript calculation checks; plain inline JavaScript, no dependencies, network calls, external model calls or storage. The coordinator provides review and browser checking for the run. The first candidate is preserved separately before that review.
Author checks cover the seven-record confusion counts, five-record ranking and ages, safety override, closed-record exclusion, source-age threshold and combined schedule/cost arithmetic. See Run conditions for the recorded review and browser test scope. No performance improvement over a previous answer is asserted.