Emergency Response Learnings

Audience: the Learning Clusters tagged emergency (in the default taxonomy F09a fire-fighting operations and fire-water systems, F09b emergency planning and command, F10 security and surveillance), plus control-room roles who make the first response decisions.

Engine: PMLC = python3 <skills-root>/learnings-coach/scripts/pmlc.py (this skill needs the learnings-coach skill installed alongside it). Read learnings-coach/references/learning-principles.md.

Gather

PMLC select <ws> --tag emergency --brief
PMLC select <ws> --tag emergency --top-rated
PMLC select <ws> --text "alarm|detect|isolat|MOV|valve|drill|fire.?water|pump|evacuat|command|CCTV" --brief
PMLC timeline <ws> --on incident      # writes timeline_event_day.json, the event-day sequence

Read the incident narrative for the response timeline.

Write

1. Response timeline: detection → alarm → first call → decision → isolation → ignition / escalation → control, with the gap between each step and what drove it (evidence-linked).
2. Learnings by response phase: detection and alarm; communication and escalation; authority and decision to isolate; plan coverage (was the scenario in the plan, with actions?); fire-water and fire-protection readiness; command structure; recovery. Three to five learnings per phase where the evidence supports them.
3. Plan review input: specific gaps for the emergency plan owners (scenarios, decision criteria, delegated authority, remote isolation, communications), each with evidence.
4. Drill scenarios: three drills built from this event, with injects, the decision points tested, and the time standard to measure.
5. Readiness indicators: e.g. detection-to-isolation time in drills, fire-pump availability, plan scenarios with tested isolation steps.

Output

Write the pack <run>/outputs/emergency-response-learnings.json (schema: learnings-coach/references/pack-schema.md), then:

PMLC check-pack <ws> <run>/outputs/emergency-response-learnings.json   # fix every problem it lists
PMLC pack <ws> <run>/outputs/emergency-response-learnings.json         # -> emergency-response-learnings.html (primary) and .md

Run PMLC timeline <ws> --on incident first: it writes timeline_event_day.json, the clock times on the day (each with its sources) to build the sequence from.

Sections → kinds:

1. A timeline with one event per step: time (HH:MM) where the evidence gives it, received_by = who got the alarm or call, and the gap since the previous step in text. Events on the same day without times keep your order.
2. learning items, with group = the response phase.
3. One para per gap.
4. scenario ×3: injects in situation, decision points and the time standard in prompts.
5. indicator.

Set "scope": {"tags": ["emergency"]}. Put data-quality corrections in data_quality, findings left out on purpose in exclusions, and same-condition findings in supporting_cites.

{"kind": "timeline", "events": [{"date": "YYYY-MM-DD", "time": "14:05", "text": "…", "received_by": "Control room operators",
 "cites": ["FND-…"]}]}

Learnings are for the system and the plan, not a judgement on the people who responded under pressure.