The code structure

• Those six files sit in every service’s folder. One more file sits outside them: utilities.py, which holds what all the services do the same way. The country list they all join to, the collapse from 264 regions to 250 countries, the report rendering, and the reading of the shared CSVs.
• run_<service>.py asks initialize.py for the list of steps, each step in tasks.py calls the equations in functions.py and writes its output.
• A run ends by rendering results.qmd into a standalone HTML report for that service, carrying its total and the per-country table.
• A service with a shock has a second runner for it, run_<service>_shock.py.
• The methods are work in progress.
• Three shared CSVs carry the configuration (next slide).

Automatic report by service

Terrestrial carbon’s report: its total, the same figures by income group, region and continent, the per-country table, and this map.

The state in numbers

• The library holds 22 service modules. Three of them value two services each, so together they produce 25 per-country accounts.

• Every available input is in the shared base data.
• 9 of our runs land on a committed reference or the manuscript. For the rest we could not find a reference to check against.
• Every module whose run has finished has rendered its report. Erosion is the one still running.

What the code does, service by service

• The services differ in how much work stands between input and number, which is separate from whether the number is ours: a service can run a model here and still take its coefficients from someone else.

• One of each, in order: urban cooling reads a dollar column out of a committed file, mining rescales a World Bank rent share times GDP, crop production multiplies FAOSTAT output by the CWoN land rental rate, and pollination multiplies 158 production grids by price and dependence cell by cell.
• The three process models have dynamics of their own: erosion runs InVEST SDR with sediment routing, landslides computes slope stability twice per pixel, and recreation runs a gravity model allocating visits into scored sites.
• The ladder measures work, not confidence. The account without a usable number sits at the top: stormwater has no dollar total because its price is a placeholder. Crop production, down at table arithmetic, reproduces its reference exactly.
• One label per service also hides a split. Water quality computes its dollar figure and reads the one it reports, coastal protection computes mangroves and reads corals, and wildfires computes the valuation on coefficients it did not fit. The ours column carries that, the category cannot.

What each service is

What each service is (continued)

What each service is (continued)

The status sheet

• Full per-service status, updated as work lands: the library status sheet.

Services and totals (base year 2019)

Services and totals (continued)

Services and totals (continued)

What this section is

• Every service here serves two callers from the same science: the GEP account reads a valuation, one row per country, and the macroeconomic pipeline reads a scenario shock, a table of productivity deviations.
• They are different products with different owners and different decisions, so their questions are kept apart. A question about which shock method to use does not belong beside a question about how a country’s value was measured.

erosion, as a shock

• Q: which of the three shock methods is the account’s? We compute damage (thresholded area share), service (threshold-free and magnitude-weighted per crop) and service-threshold (service restricted to a fixed severe-pixel set) side by side.
• Q: should the production run recompute the erosion shock, as the test runs already do? The scenarios file production reads names carbon and pollination as dynamic but not erosion, so erosion alone falls back to the frozen dependency table. The tracked template added erosion in August; the derived copies the runs read date from July and shadow it, because a file already present is never re-seeded.

fisheries, as a shock

• Q: which CWoN vintage produced the shock file (cwon_shocks.har)?

Turning outlines into cells

• Every service that totals a raster inside a country has to decide what a cell the border passes through counts as. Under the centre rule the cell goes to whichever country covers its exact centre. Under all-touched every country whose outline touches it claims it, and since the claims are burned into one id raster, the country rasterised last takes it. Nothing is counted twice either way; what moves is the split between neighbours.
• Q: should a habitat extent carry the fraction of the cell it covers, instead of either whole-cell rule? The centre rule loses a mangrove strip narrower than a cell, and all-touched books that same 50 m strip as a full 300 m cell. Coastal carbon now measures the fraction, which brings its mangrove extent back to the mapped one and roughly halves it. The question left for the room is whether every service that maps a thing narrower than its cells should do the same.

What a reference check can establish

• Eleven service entries rest on reproducing an author’s committed output. Where that output came from the code we ported, both sides share every assumption, so agreement establishes that the port is faithful and not that the number is right.
• Q: what should a service show beyond agreement with its author, before the account treats its number as verified?
• Q: where a check finds something that belongs to the author’s published number rather than to our port, who takes it to them, and does the account wait for their answer?

How a service measures a cell

• A service that turns a density into a total multiplies by cell area, so the cell area silently scales every number it produces.
• There is a default and there are justified departures from it. The carbon services and pollination sit on the shared pyramid in geographic coordinates and carry a per-latitude area. Erosion and ntfp work in equal-area projections because their science needs true distances and areas, and there a constant cell area is exact where a latitude formula would be wrong.
• The two fail differently. A constant cannot be applied wrongly per cell. A per-latitude area is more precise and every consumer has to remember it, which is how pollination’s earlier figure went wrong.
• Flood control sits on a projection that is neither, so its constant cell area is exact only at the equator. That is the absence of a convention rather than a third one.
• Services on different grids cannot be compared cell to cell without a reprojection, which the landward mask coastal carbon needs would require.
• Q: should a service be required to sit on the shared pyramid unless an equal-area grid is needed, and to state what its cell area rests on, the way the erosion code refuses to run on an unprojected input?

The configuration table (es_config.csv)

• The remaining services’ rows follow the same columns. Blank price-convention cells fill as each module’s convention is named. Every row is under test the moment it is edited.