Coding Notes

The rules, conventions, decisions and open questions behind the RADPKO v2 update, gathered in one place. The Status tab tracks how far each map has been checked; this page records how the coding was done.

How maps are read

• Count unit lines, not battalion numbers: each stacked line in a box is one unit, and trailing numbers or tags such as "ETHIOPIA 1" or "INDIA 1 (-2)" are battalion names, so three "ETHIOPIA 1" lines are three units
• Combine every column at one base point: headquarters bases carry a force or sector headquarters column and a troop column at the same coordinate, and the published files add them together
• Skip command lines: HQ, Force, UNPOL and Sector lines mark commands, not contributing countries
• Follow leader lines, never proximity: on dense maps each box connects to its town by a thin line with right-angle bends; an offset box on its own line is a separate base, and boxes whose lines meet at one dot are one base
• Flag crowded headquarters knots: where several labels share a point the binding is genuinely ambiguous, so the cluster total is recorded and the rows go to human review
• Check the printed month: a report can carry an older month's map, so the map's own title is checked against the report month

How well the rules worked

• UNMISS February 2020, a sparse map: 52 of 52 base and country pairs matched the published counts exactly
• MONUSCO February 2019, the densest map-backed month: 55 of 61 pairs exact, and 61 of 61 at the cluster level; all six misses sat inside the Adikivu, Bukavu and Walungu knot, where the published files bind units to a different label than the drawn leader lines
• The full unit-count layer covers 72 of 72 months: 3,410 rows, of which 834 carry more than one unit
• Review queues: 484 rows at crowded headquarters points, 7 months flagged because the report reuses the previous month's map, and 190 rows seen on a map but absent from the coded files (logged as gaps, not added)

Conventions in the counts layer

• Staff Officer lines, which the UN reports added in 2017, count as troops
• Formed police units (FPU) count as police
• The 2005 and 2006 reports call the Côte d'Ivoire mission ONUCI; those rows are read as UNOCI
• Where a source gives no gender split, the total goes in the male column and the female column is 0
• Country names cleaned to the compile's canonical forms (5,027 fixes), with no residual non-standard names
• Serbia coded as Yugoslavia (COW 345, the successor code), kept distinct from Montenegro (341), which fixes a v1 bug
• Congo (Republic of the Congo) and DR Congo coded as distinct contributors; v1 and the published compile had merged them
• MINURCAT's Central African Republic sector, mislabeled MINUSCA in v1's 2010 data, relabeled and merged back into MINURCAT, so MINUSCA now starts in 2014
• 17 months in which the same country appears twice for a mission remain flagged per mission and are not yet merged

Abyei (UNISFA)

• All 18 Abyei deployment maps from 2018 to 2025 were coded and spliced into the staged base-level data, so UNISFA base-months now run from September 2011 to September 2025 (they had stopped in January 2018)
• Months between maps carry forward the nearest earlier coded map
• After 2018, UNISFA location rows keep troop-contributing countries only, so there is no multinational category
• Serbia contributes to UNISFA in the counts but has no coded map location
• The UN's schematic deployment maps are not drawn to scale, so outposts were placed from UNISFA's own gridded road map of the Abyei Administrative Area (its grid checked to within 0.5 km at Farouk), and a second source counts as agreeing within 3 km
• 23 outposts were accepted, covering 326 of 333 unit rows; the other 7 rows (alai) use the alal coordinate until the question below is settled

Verification so far

• Newly added base coordinates outside Abyei: 41 of 41 confirmed, with no changes recommended
• Coordinates where v1 and Geo-PKO disagree by 25 km or more: 14 proposed corrections, 2 candidate changes rejected and 9 v1 coordinates confirmed; nothing has changed in the canonical data or on Dataverse
• Blind re-reads of the three earliest Abyei maps found boxes from one column credited to several nearby town dots, and left the Force HQ binding (Abyei or Tajiel) as an open convention choice; nothing from that review has been applied
• Across these checks, accuracy came from the reading procedure (primary sources, leader-line tracing, adjudication) more than from which model did the reading, and dense maps need adjudication after any single reading

Decisions on record

• MONUSCO June 2022: the UN's reissued June map carries the May 2022 base map under a June title; it is treated as the June 2022 map (decided June 15, 2026)
• Fall 2026: research assistants blind-code v2 maps as an independent check, and their readings move cases up the Status ladder once adjudicated
• v1 and v2 are now assigned by where the data came from, not by date: a month is v1 when people coded it before 2026; for base locations this moved January to May 2020 (UNMISS and MINUSCA), January to February 2020 (MINUSMA), January 2020 (MONUSCO) and January to September 2018 (UNISFA) from v2 to v1

Open questions

• Should nyam, which sits about 15 km north of the Abyei Area boundary, be kept?
• Is alai the same place as alal, and should the two be merged?
• Should um khaer use its map label position or another coordinate?
• Should ed dahlob use the gazetteer point or the map's village dot, which sit 4.6 km apart?
• Should the gridded-map coordinates replace the older ones for dumboloya and rumamier?
• Should dal stay, although it lies just outside the Abyei Area boundary?
• Is tajiel a misspelling of tejalei, or a separate village near debab?
• Should the leu awoluum rows be matched to awoluum or leu instead of dropped?
• Is dropping the UN, UNMO and UNPOL marker rows after 2018, which leaves no multinational category, intended?
• How should Serbia's UNISFA contribution, which has no coded location, be handled?
• Should the six functional labels (about 28 rows) that have no point location be dropped?
• Which base takes the units at crowded headquarters knots when the drawn leader line and the sector headquarters convention disagree (the Adikivu, Bukavu, Walungu and Kavumu knot is the clearest case)?
• Should the flagged re-reads of the three earliest Abyei maps be applied?
• Should the 17 double-entry months be merged before release?
• Which counts panel is the release version, given that Coverage reads the final working panel while the year-by-year table reads the source-validated panel?

Changelog

• October 2, 2026: four-tab dashboard with the verification ladder; one provenance-based era rule on every tab
• July 9, 2026: Explorer and dashboard redeployed with corrected Abyei data
• July 2, 2026: three verification tracks completed
• June 21, 2026: geography view added to the coverage dashboard
• June 20, 2026: Abyei extension wired into the staged data
• June 15, 2026: unit-count layer complete for the active missions; staging compile revived
• June 10, 2026: the 40 sparse months re-read for unit counts
• June 9, 2026: both reading pilots passed