Task register

This page is generated from the repository on every merge to main; changes made here are overwritten.

The register is a catalogue of 857 humanitarian mapping tasks in 49 sections (rapid mapping, health, WASH, logistics, protection, remote sensing, reporting and so on). Each task is a machine-readable contract. When you type a request, the plugin matches it against the register on your machine. No API call is made to match, and the register is never sent to the model: it is about 218 KB, so only the one matched task's contract is injected into the prompt.

What a task contains

Field	Meaning	Example
id	Section number, a dot, and a stable position	5.02
text	What the task is	"Classify damage by severity"
out	The output contract: what you should end up with	analysis
slots	Information the task cannot proceed without	aoi (area of interest), facility_type, threshold, time_range, hazard_type, sector…
tools	Suggested tool chain, validated against the live registry: acquisition first, renderer last	calculate_severity_index, apply_humanitarian_look, export_to_csv
kw	Matching vocabulary	classify, damage, severity
acc	File kinds the task can take in	vector, raster, pdf, table, image
prod	File kinds it can put out	.csv, .pdf, .html

Output contracts, and what each promises you:

out	You should end up with
layer	One or more styled QGIS layers on the canvas
layout	A composed print layout ready to export
dashboard	An HTML dashboard
report	A written report with supporting figures
analysis	Numeric results, a table and a short interpretation
dataset	An exported dataset or schema, not a rendered map
guidance	Written guidance or a procedure; no GIS output expected

How matching works

Stage 1 is pure local scoring (task_matcher.py): tokens of your request against each task's vocabulary, ignoring words that nearly every task contains ("map", "data", "area"…). A match is trusted when it clears a score (0.34) and beats the runner-up by a margin (0.08); otherwise it is ambiguous and the request is sent as typed. Failure to match is a normal outcome, never an error.

Rules that came out of hand testing:

• A request that names a tool ("Use optimal_hub_siting …") is only matched to a task that uses that tool; otherwise it goes as typed.
• A weak tie across sections (under 25 percent keyword coverage) gives no directive.
• A request that names a loaded layer is not asked "Which facility or service type?".
• A task that names one hazard (drought, extreme heat) matches only when the request names it.

What you see

1. Missing information (a slot with no safe default) is asked once, before anything is sent. Things the plugin can work out itself, such as the area of interest from open layers, are not asked.
2. The prompt preview: which task matched and how confident, what you will get, any value assumed on your behalf, and what each attached file will be read as. "yes", "send" or "go ahead" sends it; typing anything else sends your own wording with nothing added; "cancel" abandons it. Previews can be turned off in Settings.
3. The output check: after the answer, the tools that actually ran are compared with the contract, and a missing deliverable you named yourself is asked for once. See Guards.

Honest limits

The register tells the model what to aim for; it does not make the model's analysis correct. A match is a heuristic. Section 36 (engineering hydrology) has a single task. See the Humanitarian mapping task reference for the taxonomy behind it.

The 36 sections

Section	Name	Tasks
1	Base mapping and reference data	21
2	Rapid mapping and emergency mapping	25
3	Population and displacement mapping	27
4	Camp and settlement mapping	26
5	Damage and needs assessment	23
6	Infrastructure and service mapping	32
7	Health and public-health mapping	29
8	Water, sanitation, and hygiene mapping	27
9	Food security and nutrition mapping	24
10	Logistics and supply-chain mapping	27
11	Accessibility and road-network analysis	23
12	Protection and safety mapping	23
13	Conflict and security analysis	21
14	Natural-hazard and disaster-risk mapping	28
15	Climate and environmental mapping	25
16	Remote sensing and satellite-image analysis	26
17	Drone and aerial mapping	20
18	Field data collection and mobile mapping	24
19	OpenStreetMap and crowdsourced mapping	23
20	GIS database management	29
21	Spatial analysis	32
22	Data analysis and statistics	31
23	Humanitarian needs assessment	23
24	Humanitarian program monitoring	22
25a	Shelter and settlements	10
25b	Education	9
25c	Health	7
25d	Protection	6
25e	Livelihoods and recovery	7
26	Reporting and information products	32
27	Coordination and decision support	23
28	Web mapping, dashboards, and visualization	20
29	Data sharing, standards, and governance	21
30	Training and capacity building	20
31	Typical final outputs	25
36	Engineering hydrology and drainage	1
38	Humanitarian response: overview, access and needs	6
39	Engineering: terrain, corridors and plans	4
40	Urban planning: zoning, catchments and parcels	5
41	Logistics: hubs, routes and delivery risk	4
42	Agriculture: crop indices, seasons and zones	5
43	Environment: land cover, change and suitability	7
44	Public health: access, cases and indicators	8
45	Disaster risk: hazards, exposure and triggers	4
46	Utilities: network checks, faults and CAD	4
47	Transport: accessibility, speeds and weak links	4
48	Public safety: incidents, coverage and route safety	5
49	Research: provenance, workflows and metadata	4
50	Real estate: site comparison and context	5

Deliverable kinds across all 857 tasks: layer (341), analysis (306), layout (76), guidance (53), dataset (35), report (34), dashboard (12).