Data handover and drills

Why this matters

A pin on a map with no source, no time and no indication of whether it is confirmed can send rescue teams to the wrong place or, worse, let images showing victims’ faces and locations leak to the public. Good data handover tells recipients how far the data can be trusted, and regular drills make the team able to do it on the day.

Item status

The knowledge unit on handing data to the command post defines statuses the team must share: pending means not yet confirmed; confirmed must name the reviewer and time; closed must give a reason. Closing an item does not always mean the area is safe; it may mean it has been passed to another unit.

import csv
import io

data = """id,time,lat,lon,status,reviewer
EV01,2026-09-20T08:05+07:00,14.052,100.612,confirmed,R02
EV02,2026-09-20T08:20+07:00,14.061,100.598,pending,
EV02,2026-09-20T08:31+07:00,14.049,100.620,pending,
EV03,,14.055,100.605,pending,
EV04,2026-09-20T09:02+07:00,41.055,100.605,pending,
EV05,2026-09-20T09:15+07:00,14.058,100.611,confirmed,
"""
LAT, LON = (14.0, 14.1), (100.55, 100.65)       # assumed incident area bounds

seen, problems = set(), []
for row in csv.DictReader(io.StringIO(data)):
    rid = row["id"]
    if rid in seen:
        problems.append((rid, "duplicate id"))
    seen.add(rid)
    if not row["time"]:
        problems.append((rid, "missing time"))
    if not (LAT[0] <= float(row["lat"]) <= LAT[1] and LON[0] <= float(row["lon"]) <= LON[1]):
        problems.append((rid, "outside incident area"))
    if row["status"] == "confirmed" and not row["reviewer"]:
        problems.append((rid, "confirmed without reviewer"))
for rid, p in problems:
    print(f"{rid}: {p}")
print(f"{len(problems)} problems in {len(seen)} ids -> hold these rows, do not fill in data")

EV02: duplicate id
EV03: missing time
EV04: outside incident area
EV05: confirmed without reviewer
4 problems in 5 ids -> hold these rows, do not fill in data

Three data versions

Chapter 7 of the ICRC Handbook on Data Protection in Humanitarian Action (3rd ed., 2024), on drones and remote sensing, points out that drone imagery may disclose personal data and affect community trust. Teams should therefore separate data by recipient.

Removing names from the text alone may not be enough if the images or metadata still reveal locations. Originals are kept with a SHA-256 manifest (FIPS 180-4), as in UAT 361, and victims’ data is never uploaded to public tools just because they are free.

Drills and indicators

Tabletop drills use synthetic data with errors planted in advance, then measure how many the team finds, how many it misses and how long until the recipient understands the report. The knowledge unit on drills warns that these criteria are exercises, not team certification standards; flying skills can be measured with NIST’s standard test methods.

drills = [  # round, errors planted, found, minutes until recipient confirmed understanding
    ("D1", 8, 5, 52),
    ("D2", 8, 6, 41),
    ("D3", 8, 8, 33),
]
for name, planted, found, minutes in drills:
    print(f"{name}: detected {found}/{planted} = {found / planted:.0%}, "
          f"missed {planted - found}, report understood in {minutes} min")
first, last = drills[0], drills[-1]
print(f"time to understood report improved by {first[3] - last[3]} min "
      f"({(first[3] - last[3]) / first[3]:.0%})")

D1: detected 5/8 = 62%, missed 3, report understood in 52 min
D2: detected 6/8 = 75%, missed 2, report understood in 41 min
D3: detected 8/8 = 100%, missed 0, report understood in 33 min
time to understood report improved by 19 min (37%)

Logs and after-action review

The knowledge unit on logs and debriefing draws on the documentation, records and occurrence handling sections of CAAT’s operations manual guidance. It separates three kinds of record: events (time, source, observation, decision), defects, and improvements with an owner, review date and closing evidence. The review compares what was expected with what happened; it is not a search for someone to blame. Improvements must be checkable: “add a reviewer before items are sent” is clearer than “everyone should be more careful”.

Module lab

Common mistakes

Summary

• Pending, confirmed and closed statuses need a reviewer, time and reason
• Separate original, operational and public versions to protect victims
• Check lists before handover; hold items lacking evidence rather than filling them in
• Drill with planted errors, measure results, and improve with owners and evidence

Check your understanding

1. 10 errors were planted and the team found 7. What is the detection rate?
2. What should happen to an item marked confirmed with no reviewer?
3. Which version should have victims’ faces and locations removed?
4. A coordinate appears to be mistyped. Should the checker correct it?
5. What should a good improvement after a review include?