Repository navigation
Branches data and environment
In one sentence: This page pins down which code and which data produced every number in this wiki as of 15 September 2026, so that you never quote a result without knowing which generation it belongs to.
Where this sits: Before everything. The pipeline (GeoDMS network → Arrow OD → Julia LP → λ-sweep → rounding → scenarios → deck, see The pipeline in plain words) is the same on paper for every generation of the data; this page says which data and which code produced the numbers on Pharmacy results September 2026 and in the deck. Every claim here is about state, so each names the command or file that proves it.
Issue #53 is the cautionary tale: for two months the deck's headline was "southern Italy stands far above the field", and the cause was a pharmacy list holding 12 991 of Italy's ~20 600 pharmacies (§3). A ranking is only as good as the provenance of its inputs. Second, the repository, the GitHub remote and Maarten's working copy are three different states (§1–2), and the wiki can only document the public one.
ServiceAccess is the pharmacy branch. Its head, locally and on GitHub, is e25416b of 10 September 2026 ("#52 refine-only walks the grid LP-only until a bracket can close"): git ls-remote origin ServiceAccess and git log -1 ServiceAccess agree, and every permalink in this wiki is pinned to it. It holds the GeoDMS configuration under cfg/ (GeoDMS OD matrix and choice set); at the root the live Julia path — settings.jl, lp_run.jl, lambda_sweep_simplex.jl, collect_descriptives.jl, export_s1s2_locations.jl and the run_* runners (Running a sweep end to end) — next to the legacy lambda_sweep.jl (the first, interior-point driver; The lambda sweep and the Pareto frontier), greedy.jl and the abandoned catchment-cap scripts cap_scenario.jl, s1_cap.jl, s2_cap.jl (Alternatives not pursued); and doc/, the deck pipeline with its committed data (deck_data.json, agg_data.json, frontier_metrics_interp*, pharmacy_descriptives*.csv, the .pptx; Deck and charts pipeline).
logs/, doc/charts/ and git.txt are git-ignored, which matters in §5; scratch/ is git-ignored too, but 25 files were added by hand (git ls-files scratch). The branch forked from main at b765147 (20 October 2025; git merge-base ServiceAccess origin/main) and is 240 commits ahead (git rev-list --count origin/main..ServiceAccess).
origin/main is at d9f1f17 (6 August 2026). git log --oneline ServiceAccess..origin/main lists ten commits ServiceAccess does not have — nine by Chris Jacobs-Crisioni, one by Maarten: 5495304 (10 Nov 2025, Maarten, .gitignore ignores .vs); 60a5c63 and the merge 1eb1619 (15 Jan 2026, catchments, settlement hierarchies, functional rural areas); 32cd4f8 (17 Feb, Euclidean distances); aec3cca (20 Feb, stored country split — where the per-landbody rule first appears on main, git log -S IsConnected_seldomain origin/main); aac1da3 (20 Feb, GUI button Step_0_Countries_split); 8252039 (24 Feb, "Cycles fix from NetworkModel_world", 257 lines in Templates.dms); 94d59ba (27 Mar, POI access); db8f6e9 (24 Apr, compressed vs uncompressed impedance check); d9f1f17 (6 Aug, city centroids).
Nine of the ten are GeoDMS configuration and one is a .gitignore line; none touches the Julia side. git diff ServiceAccess origin/main --stat -- cfg: 25 files, +1 627 / −1 986 lines; Templates.dms alone differs by 598 lines. The local main in C:/ProjDir/NetworkModel_EU is stale (still b765147), so always compare against origin/main.
"Keep the largest strongly-connected road component per landbody (one polygon of Country_Split) rather than per study area" is the rule that gave islands their road networks back; what it is, why it was needed and what it moved (ITG: 983 169 unreachable residents, 84.6 % of its baseline cost, before the fix; #13, #49) is on Scope rules and data gaps, and the configured items are on GeoDMS OD matrix and choice set. What matters here is that the rule exists on both branches but got there twice:
- On
mainwithaec3cca(20 February 2026): the landbodies are computed once withgeos_split_union_polygon, written to%NetworkModelDataDir%/Regions/Country_split.fssand read back (origin/main:cfg/main/SourceData/RegionalUnits.dmsaround lines 136–148). Chris's remark on #13 — "storeCountry_split_storebefore running from now on" — refers to that implementation. - On
ServiceAccesswith79cb58b(31 August 2026), committed by Chris directly on this branch — not a merge, not a cherry-pick (git branch --contains 79cb58blists onlyServiceAccess; +267 lines over five files). Here the landbodies are computed in memory every run (Country_Split,RegionalUnits.dmsaround line 117); there is no.fssstore (grep -rn Country_split_store cfgfinds nothing), so the "store it first" instruction does not apply.
The selection rule itself is identical on both branches: Roads_isConnected (Templates.dms around line 93) reads Strong_connectivity/IsConnected_seldomain (Main_in_seldomain around line 775, IsConnected_seldomain around lines 779 and 796); the Undirected_connectivity container around line 727 is dead code (GeoDMS OD matrix and choice set). 79cb58b also changed StudyArea_default from 'Netherlands' to 'ITG' (ModelParameters.dms around line 23); batches always export STUDY_AREA, so only GUI sessions notice.
Nothing has been merged in either direction since October 2025. ServiceAccess lacks the stored country split, the cycles fix, POI access, the compression check and the city centroids. main lacks everything pharmacy-related: the Pharmacies container; ModelClient, the client set actually modelled (GeoDMS OD matrix and choice set); the NUTS export column; the .mmd read-holder changes for GeoDMS 20.19 (d80622d); the #50 fix 7086928; Allocate_T and the read-back of Julia's S1/S2 results (ReadSweepResults_T). If you edit cfg/, say which branch you are on in every commit and comment (a Templates.dms line number means different things on the two), and expect a merge to conflict in Templates.dms, RegionalUnits.dms, Analyses.dms and ModelParameters.dms, with the two landbody implementations reconciled by hand. The 8252039 cycles fix has never been run against the pharmacy ODs; whether it changes any is unknown. Results are reproducible only from ServiceAccess.
Issues #52, #53 and #44 (and the superseding comments on #45 and #48) quote commit hashes that are not on the remote: on 15 September git cat-file -t 216ac77 (and the other four) fails with "Not a valid object name". What each contains is known only from the issue texts. S1 and S2, the λ grid and refine-only mode are explained on Scenarios S1 S2 S3.
| Hash | Per | Contains |
|---|---|---|
6aaab27 |
#52 | The refine-only walk climbs from the pinned S1 point towards S2 in ×1.3 steps instead of taking the next grid step (Denmark had spent 1 883 s on one bracket endpoint) |
92ab4e4 |
#52 | The same ×1.3 climb towards S1, once the LP's open count is within ×2 of today's |
82d84b4 |
#52 | No basis restore after the last scenario is pinned |
86e242e |
#52 | A timed-out climb step ends the walk instead of being retried |
216ac77 |
#52, #53, #44 (#45/#48 zips) |
The 11–12 September regeneration: Italy on the OECD list, Hungary new, Finland back, S1/S2 pinned in 87 of 88 sweeps; deck_data.json, README, the 76-slide deck, doc/check_s1s2.py, the #45 zip s1s2_open_locations_20260912.zip (44 areas) and the #48 zip issue48_frontier_data_20260912.zip
|
Consequence: this wiki documents e25416b. The results page and the committed doc/deck_data.json are the 3 September generation (de5b178, regenerated on the post-landbody results; agg_data.json a8bd280, 4 September) as committed at the 10 September head — the 10 September commits only touched charts, README and slides. deck_data.json has 42 entries (14 countries + 28 NUTS-1 regions), of which the aggregate uses 41 because country-level Poland is dropped (Metrics aggregation and ranking); adding Hungary and Finland gives the 43 of 12 September. In this generation Italy is on the ESPON list and S1/S2 in the exports still read off the nearest grid point.
The 12 September state exists in issue #53's closing comment and nowhere else until Maarten pushes: 43 disjoint areas, 51 350 pharmacy cells, S1 −23 % total travel at today's pharmacy count, S2 −30 % pharmacies at today's travel (LINEAR), the Nordic and Baltic areas leading the ranking. The results page's −26 % / −34 % (41 areas, 3 September data) and #53's −23 % / −30 % (43 areas, 12 September) are different data generations, not a disagreement; never mix them. The FRI re-sweep of #54 (the scheduled task NM_fri54 on Maarten's machine) is a further, later generation.
The twenty countries of the Country enum in cfg/main/SourceData/Locations.dms are read the same way at e25416b: a parquet %NetworkModelDataDir%/Locations/pharmacies/<code>_pharmacies.parquet with lat/lon columns, through the template CountryDataT, which builds the geometry from lat/lon and reprojects to LAEA (around lines 366–405). CountryDataShpT (around line 411) reads a point shapefile instead (Italy's ESPON file, 10 July–10 September). AvailableCountry (around line 389) lists the countries whose file is present; the EU union is built over those only, so a missing file never breaks the pipeline. Adding a country is Recipe A of Running a sweep end to end (2aeb6c0 and 8963d27 are the templates).
Counts are n_pharmacies / n_pharmacy_cells from the committed doc/pharmacy_descriptives.csv (last regenerated dd2c157, 3 September, post-landbody) unless stated; a "cell" is a 1 km² grid cell holding at least one pharmacy, the unit the model counts on the Existing side (GeoDMS OD matrix and choice set, Glossary). Each file's reference date sits in its ref_date column (#44, #53); the files are not in the repository, so the dates cannot be verified from this checkout.
| Country (code) | Source and delivery commit | Pharmacies / cells | Notes |
|---|---|---|---|
| Austria (aut), Belgium (bel), Czechia (cze), Denmark (dnk), Estonia (est) | OECD list shared by Taina Pacheco, 0b36dcb 12 Jun 2026 (Chris) |
1 467 / 1 131 · 4 860 / 2 963 · 2 668 / 1 627 · 536 / 466 · 491 / 292 | — |
| France (fra) | parquet since 94f88db 24 May 2026 (first four: FR, IT, NL, SE; origin not recorded); whether it was refreshed on 12 June is not recorded (0b36dcb only edits the enum) |
20 698 / 13 863 | swept as 13 NUTS-1 areas |
| Ireland (irl), Lithuania (ltu), Latvia (lva), Norway (nor), Slovenia (svn) | 0b36dcb |
1 652 / 865 · 1 253 / 582 · 818 / 496 · 1 057 / 758 · 355 / 306 | — |
| Iceland (isl) | 0b36dcb |
none |
No study area: Iceland lies outside the Ardeco 2021 population grid, so it has no clients and no pharmacy OD; a GeoDMS run hangs rather than errors (#44; Running a sweep end to end). Dropped from every batch on 5 Jul (d0ae5ef; make_descriptive_table.bat lines 34–35) |
| Italy (ita) | three generations — see below | 12 991 / 8 794 (ESPON) | the results page and deck_data.json at e25416b reflect the ESPON list |
| Luxembourg (lux) | 0b36dcb |
105 / 86 | pilot area for the July changes |
| Netherlands (nld) | parquet since 94f88db; 12 June refresh not recorded. Jip's apothekersregister20251120_cleaned_Geocoded.csv (8ebc6aa) is still declared as unit Apotheken (Locations.dms around line 155) but is not what Pharmacies reads; whether the parquet derives from it is not recorded |
1 992 / 1 617 | the only hard-coded raw count in Julia (RAW_PHARMACY_COUNTS, lambda_sweep_simplex.jl line 9); it feeds the informational n-raw log column and changes no result (Log lines and output files) |
| Poland (pol) | 0b36dcb |
13 791 / 7 030 | swept as country and as 7 NUTS-1; country row excluded from the aggregate |
| Portugal (prt) | 0b36dcb |
2 801 / 1 893 |
No records on the Azores (PT200) and Madeira (PT300) although the population grid has residents there (doc/replan_connectivity_fix.md §6); handled by the NUTS exclusion rule 872b388 (#49, Scope rules and data gaps), so Portugal's figures are mainland only |
| Sweden (swe) | parquet since 94f88db; 12 June refresh not recorded |
1 448 / 1 088 | swept as 3 NUTS-1 areas |
| Added September 2026 — not in the committed CSV | |||
| Finland (fin) |
0b36dcb, removed the same day in c11fd38 (#44: a GeoParquet with a real geometry column broke the union on GeoDMS 20.0), reinstated 2aeb6c0 10 Sep on 20.19 |
797 / 720 (#44) | no code fix needed: the 20.19 reader ignores the geometry column and the template's lat/lon geometry wins |
| Hungary (hun) | OECD delivery 7 Sep 2026: National Center for Public Health and Pharmacy list, geocoded by the OECD; records whose name starts with "DR." removed, raising the reported accuracy — presumably of the geocoding — from 76 % to 90 % (neither is defined in #53); wired in 29fcf7f 10 Sep |
3 094 / 1 865 (#53) | typology Type 1 / HIGH (8963d27) |
| Generation | Source | Count | Commit | Reflected in |
|---|---|---|---|---|
| 1 | June parquet, origin not recorded | 20 637 pharmacies in the study area / 12 991 cells in the descriptives of d0ae5ef (5 Jul; 20 735 in the first run 707c48b, 22 Jun) |
94f88db / 0b36dcb
|
the June–July decks |
| 2 |
ESPON 2021 point shapefile, via CountryDataShpT
|
12 991 / 8 794 |
6a208f0 10 Jul |
the results page and deck_data.json at e25416b; the "southern Italy" headline |
| 3 | Ministry of Health list geolocated by the OECD, delivered 7 Sep | 20 621 in the study area; residents per pharmacy 4 520 → 2 848 |
29fcf7f 10 Sep; results in 216ac77 (unpushed) |
#53's closing comment: ITF 1 951 → 4 955 pharmacies, ITG mean travel 4.9 → 2.0 min |
Why generation 2 replaced generation 1 is not recorded (#46 says only "Data included and read"); the June file is archived as ita_pharmacies_20260622_unchecked.parquet. One number needs a flag: the ESPON pharmacy count of 12 991 — quoted by Home, Travel cost functions, Metrics aggregation and ranking, FAQ and the Locations.dms comment around line 376 — equals the June list's cell count in d0ae5ef exactly. Whether that is coincidence or a sign that the ESPON file carried one point per cell of the same list is not recorded and stays open (the dd2c157 descriptives, with 2 022 ESPON cells holding more than one pharmacy, argue against the simple reading; neither file is in this checkout). Two consequences: the committed descriptives at e25416b are pre-OECD-Italy and without Hungary and Finland (collect_descriptives.jl and make_descriptive_table.bat list both, but their rows exist only on Maarten's machine); and the enum had only four countries until 12 June 2026, so any result dated before that is Netherlands/France/Italy/Sweden only.
| Input | Vintage | Where it is read | Notes |
|---|---|---|---|
| TomTom road network |
2021 extract: NW2021_SP_streets.mmd, TomTom_Compiled_Themes_2021_Junctions.mmd
|
cfg/main/SourceData/Infrastructure/TomTom.dms around lines 40–58 |
Junctions are clipped to the country bounding box for a country run but to the exact NUTS-1 polygon for a NUTS-1 run (around lines 53–55), so border effects differ between the two kinds of study area (GeoDMS OD matrix and choice set) |
| Population grid |
Ardeco 2021, 1 km: pop_2021_1km_ARDECO.tif
|
Locations.dms around line 69; PopVersie := '2021_Ardeco' (ModelParameters.dms around line 19) |
Client = every populated cell (Client := 'population', 033dd32); since f0acbc9 (1 Sep) the LP and the baseline use ModelClient, the cells with pharmacy-coverage data, on both sides (GeoDMS OD matrix and choice set) |
| NUTS and country polygons |
Nuts0.shp, Nuts3.shp, Nuts1_derived.shp (union of NUTS-3, pre-generated by NUTS1/WRITE_geometry); countries CNTR_RG_01M_2020_3035.shp
|
cfg/main/SourceData/RegionalUnits.dms around lines 3–40 and 86 |
The NUTS year is not recorded in the configuration or the file names; probably the 2021 edition (inference; unverified). The region-exclusion rule of #49 needs only the code prefixes (settings.jl, excluded_nuts_regions / in_excluded_region; Scope rules and data gaps) |
| DEGURBA | DGUR_LEVEL2_GRD_1KM_2021.tif |
Locations.dms around line 82; Chris's degurba_avg in Analyses.dms around line 338 |
Not used for the deck colouring any more: the DEGURBA-style class of e75c73c was replaced by the policy typology in 559dad9 (17 Jul) (Metrics aggregation and ranking) |
All of these live under %NetworkModelDataDir%, which defaults to %SourceDataDir%/NetworkModel_EU (cfg/main.dms around line 29) — a GeoDMS registry setting on each machine.
Take C:/ProjDir/NetworkModel_EU on this machine as the test case (checked 15 September 2026).
It can read all the code and every committed result — doc/deck_data.json (42 areas × LINEAR and LOGISTIC, Travel cost functions), agg_data.json, frontier_metrics_interp*, pharmacy_descriptives*.csv, issue48_*.csv and the decks — and run steps 2–5 of doc/README_deck.md (charts, build_deck.mjs, metrics, ranked lists) from those files without any log, so the 3 September deck can be re-rendered. build_deck.mjs COUNTRY_SET (line 488) lacks Hungary and Finland at e25416b, so a deck rebuilt from a 12 September deck_data.json with this script would list both among the NUTS-1 regions (Deck and charts pipeline; whether 216ac77 fixed it is unknowable until pushed). It can also run the toy examples on the method pages, and the LP on any small set of the three Arrow files it reads (OD rows, clients, candidates; Log lines and output files) that you make yourself.
It cannot:
- Run step 1 of the deck pipeline (
build_deck_data.pyparseslogs/sweep_*.logandlogs/refine_*.log):logs/is git-ignored and absent here. - Run any sweep, baseline, descriptives run or S1/S2 export: the Arrow inputs and Julia outputs live under
C:\LocalData\networkmodel_eu\{ExistingPharmacies,NewPharmacies,Pharmacies_Descriptives}\on Maarten's machine, whose repository is atE:\prj\JRC\NetworkModel_EU(everyrun_*.ps1starts withSet-Locationthere;export_s1s2_locations.jlline 18 defaultsS1S2_OUTunder it). HereC:\LocalData\networkmodel_eu\holds only school-era exports of February–April 2026 (Finland_*,Netherlands_*,Romania_od.arrow,ExistingSchools/,NewSchools/; the.xmlsidecars say GeoDms 19.0.0 (February) or 19.2.0 (April) and Lola Dekhuijzen's run filec:/users/louiselauradekhuijze/…/cfg/main.dms) — noExistingPharmacies/, noNewPharmacies/. - Rebuild a network or OD: that needs the TomTom, population, NUTS and pharmacy files on the
%SourceDataDir%share plus GeoDMS 20.19.1.m. - Reproduce Maarten's package versions: there is no
Project.tomlorManifest.toml, so the Julia environment is unpinned.
Versions actually seen (see also Running a sweep end to end, Installation of Julia):
| Component | Version | Evidence |
|---|---|---|
| Julia | 1.12.5 |
Installation of Julia; julia is not on PATH on this machine |
| HiGHS | 1.13.1 (git hash 1d267d97c) |
"Running HiGHS 1.13.1" in every sweep log |
| Julia packages |
Arrow, JuMP, HiGHS, Random
|
using line of settings.jl; versions unknown — ask Maarten for ] status
|
| GeoDMS | 20.19.1.m, installed build |
GEODMS_EXE default in run_pharmacy_pipeline.bat line 76 and make_descriptive_table.bat line 24; d80622d adapted the .mmd read holders to the 20.19 contract, so older engines will not read them. The school pipeline run_oeso_pipeline.bat stays on 19.5.0 |
| Python, Node, PowerPoint | Python 3 + matplotlib (no requirements.txt); pptxgenjs 4.0.1 (doc/package.json); PowerPoint via COM (doc/merge_deck.ps1) |
doc/ scripts |
| Julia data root | C:\LocalData\networkmodel_eu |
settings.jl lines 3–5; must equal GeoDMS's %LocalDataProjDir%
|
A frontier in the deck is not necessarily one run: since #52, build_deck_data.py (around line 166) takes the frontier rows from the last full sweep log and, when a newer logs/refine_<area>_<FUNC>.log exists, the S1/S2 summary and any bisection rows from the refine-only run (merge rule: Deck and charts pipeline). When the two runs were solved on different inputs (July OD vs the 2 September rebuild, or a subsampled vs a full candidate set), their points do not lie on one curve: FRI's joined curve has a kink where the refine points meet the July curve — issue #54, a provenance artefact, not a concave dent of the frontier (The lambda sweep and the Pareto frontier).
In the table, movers are the areas whose OD row count changed after the landbody fix (ITG +60 %, PL8 +28 %, Portugal +11 %, plus SE2/SE3 after the #50 fix); candidate-subsampled means only every k-th candidate cell was offered to the LP (LOCATION_SELECTION_FACTOR=k), keeping the cells of existing pharmacies (PROTECT_BASELINE, "baseline-protected"; Scope rules and data gaps). Provenance per area, as far as known on 15 September:
| Group | Areas | Full sweep | OD under the full sweep | September refine |
|---|---|---|---|---|
| Italian NUTS-1 (5) | ITC, ITF, ITG, ITH, ITI | September, after 29fcf7f (S1/S2 pinned inside the sweep); ITG had also been re-swept 2–3 Sep as a mover before the OECD list superseded that run |
2 Sep rebuild + OECD list (network rebuilt: FinalSet carries the pharmacy connector links, #53) |
none needed |
| New countries (2) | Hungary, Finland | September | current data | none needed |
| Movers (4) | PL8, Portugal, SE2, SE3 | 2–3 Sep (run_resweep_batch.ps1, after 7086928 fixed the #50 crash for SE2/SE3 and Norway) |
2 Sep rebuild | refine-only, 11–12 Sep |
| FRI | FRI | July (22 Jul, scratch/run_fri_protect.ps1 per #54 — not in the repository): 6 765 of 17 141 candidates offered (factor 3, baseline-protected), interior-point solver rather than warm-started simplex, λ grid stopped at w = 0.5 |
pre-landbody OD (85 existing-side rows and 8 candidates differ from the rebuild) | refine-only 11 Sep on the full 17 133 candidates → the two-LP curve of #54; being fully re-swept (NM_fri54), unpushed |
| Marginal OD drift (16) | Estonia, FRC, FRD, FRE, FRF, FRG, FRH, FRJ, FRK, FRL, FRM, PL2, PL4, PL5, PL6, SE1 | July, full candidate set | July OD differs by a few dozen to (SE1) 3 500 of 1.6 M rows from the 2 Sep rebuild; the 2 Sep check (scratch/diff_rebuild.py) compared Arrow file sizes and baseline-block counts against a 1 % threshold, not row contents, so this went unnoticed |
refine-only 11–12 Sep on the rebuilt OD → kinks (the step where refine points join the July curve) ≤ 1.5 % (FRH); S1/S2 pinned on current data. Re-sweeping all 16 is ~32 runs / ~20 h — awaiting Chris's decision (#54) |
| The rest (16) | Netherlands, Luxembourg, Latvia, Slovenia, Lithuania, Ireland, Norway, Denmark, Austria, Czechia, Belgium, Poland, FR1, FRB, PL7, PL9 | July, full candidate set | July OD matched the 2 Sep rebuild in the #54 check (inference from "16 more areas"; per-area output on Maarten's machine) | refine-only 11–12 Sep; country-level Poland LINEAR is the one sweep whose S1/S2 are still the nearest λ-grid rows rather than bisected points — each solve near S1 takes 3–15 h (#52; Scenarios S1 S2 S3) |
Which July log an area's frontier comes from is visible only in logs/ on Maarten's machine. All 41 areas were re-swept under soft coverage by 4ec47a9 (16 July, "deck and derived data regenerated for all 41 regions"). After the baseline-protection commits f167da2/f6edc57 (21 July) only a subset was re-run: the Round-2 dense grid on 11 areas, high-λ extensions for Belgium, FRI, Portugal, ITF and PL8, and the baseline-protected FRI run (cb21c6c, 22 July).
The PL8 question. doc/issue48_README.md line 67 says FRI and PL8 were candidate-subsampled; README (around line 332) and Pharmacy service access (around line 118) say only "the largest areas are candidate-subsampled"; issue #54 says "FRI was the one area whose July sweep used the candidate subsample". The July PL8 sweep was subsampled: the message of 4cee238 (24 July) records a "PL8 baseline-protected subsample", the source of issue48_README.md's "FRI and PL8". #54 is consistent with it only because PL8's July log was replaced by the 2–3 September full re-sweep (run_resweep_batch.ps1 sets no LOCATION_SELECTION_FACTOR; no .ps1 in the repository does, and the settings.jl default is 1). Confirming that the September PL8 log is full-set needs its M= / protected … kept … (factor=…) header in logs/sweep_PL8_LINEAR.log on Maarten's machine; until then, read "the largest areas are candidate-subsampled" in README and the wiki as a July statement.
-
Repository. The wiki is its own git repository,
https://github.com/ObjectVision/NetworkModel_EU.wiki.git, cloned atC:/ProjDir/_Tools/NetworkModel_EU.wiki; HEAD on 15 September is1d16cc7(10 September 2026). Its history: the school-era pages (Home and Service allocation procedure created by Jip on 19 November 2025, Installation of Julia by Maarten on 9 March 2026, edited by Maarten, Jip and Lola until 31 March 2026), then Maarten's two commits of 10 September (c2f7813,1d16cc7), which wrote Home, Pharmacy service access, Lambda sweep method and Pharmacy results September 2026 as a condensed copy of that morning's README (1a78bc1,08de764). Pages are pushed by whoever edits them: GitHub's web editor commits as the editing user, a local clone as the git user. -
Images. The chart scripts write to
doc/charts/(git-ignored); ten PNGs (four of which the README embeds) were copied by hand todoc/img/(last touched1ed57f2, 10 September) and from there to the wiki'simages/. Copying is manual, so the two drift: at1d16cc7eight of the ten wiki images are byte-identical todoc/img/, butlambda_axis_AGGREGATE.pngandlambda_axis_Netherlands.pngare older renders (md5sumdiffers; they lack the log axis and level labels offe263d1/1ed57f2). After a re-sweep: chart script →doc/charts/→ copy todoc/img/→ commit → copy to the wiki'simages/→ commit and push. -
The four explainer figures on the method pages (
images/explainer_*.png) are generated byscripts/make_figures.pyin the wiki repository (python scripts/make_figures.py images); regenerate them there, not from the deck. -
Pinned commit. Permalinks point at
e25416b. When Maarten pushes216ac77and the FRI re-sweep, update first: the results page, the numbers on Metrics aggregation and ranking and Scenarios S1 S2 S3, and §2 and §6 here. The method pages should not need to change.
| To check… | Open |
|---|---|
which branch/commit you have; what main has |
git log -1 --format='%h %ad %s' --date=short; git log --oneline ServiceAccess..origin/main
|
| the landbody rule |
cfg/main/Templates.dms around lines 87–94 (Roads_isConnected) and 763–801 (Strong_connectivity); cfg/main/SourceData/RegionalUnits.dms around line 117 (Country_Split) |
| the pharmacy readers |
cfg/main/SourceData/Locations.dms around lines 358–451 |
| hard-coded machine paths |
settings.jl lines 3–5; export_s1s2_locations.jl line 18; the Set-Location line of every run_*.ps1
|
| how refine logs are merged into a frontier |
doc/build_deck_data.py around lines 166–171 |
| the committed results and their dates | git log -- doc/deck_data.json doc/agg_data.json doc/pharmacy_descriptives.csv |
Permalink to the documented commit: https://github.com/ObjectVision/NetworkModel_EU/blob/e25416b/cfg/main/SourceData/Locations.dms#L358.
-
Three states, one wiki. Repository and GitHub at
e25416b, Maarten's working copy at216ac77+. Any number in an issue comment dated 11 September or later is not in the files you can open. Blocking action item: push (#52, #53, #54, #44). -
Open without a file on this machine: whether the September PL8 log is full-set (§6), the NUTS vintage and every parquet's
ref_date(§3–4), the 12 991 ESPON-count coincidence (§3), the JuMP/Arrow versions (§5). -
Two stale wiki images: the
lambda_axis_*.pngcopies are older renders thandoc/img/(§7). -
SweepResults/baselineinAnalyses.dmsaround line 381 still reads theQUADRATICfolder (fe8256a, 2 June 2026, pergit blame -L381,381 cfg/main/Analyses.dms; untouched since) — a travel function nobody reports; the consequences and the one-line fix are on GeoDMS OD matrix and choice set. - The methodology draft and the REGIO reviewers' comments are not in the repository;
doc/deck_changes_from_review.mdis the only traceable record (Decision log).
- Home · The pipeline in plain words · Scope rules and data gaps · GeoDMS OD matrix and choice set
- Running a sweep end to end · Log lines and output files · Deck and charts pipeline · Installation of Julia
- The lambda sweep and the Pareto frontier · Scenarios S1 S2 S3 · Metrics aggregation and ranking
- Decision log · Pharmacy results September 2026 · Pharmacy service access · Glossary
NetworkModel EU documentation
Start here
Understanding the method
- The facility location model explained
- Travel cost functions
- The lambda sweep and the Pareto frontier
- From LP fractions to real pharmacies
- Scenarios S1 S2 S3
- Scope rules and data gaps
- Metrics aggregation and ranking
- Background theory
Working with the code
- GeoDMS OD matrix and choice set
- Code walkthrough settings
- Code walkthrough lp_run
- Code walkthrough lambda_sweep_simplex
- Running a sweep end to end
- Log lines and output files
- Deck and charts pipeline
- Installation of Julia
Reference
- Lambda sweep method
- Pharmacy service access
- Pharmacy results September 2026
- Decision log
- Alternatives not pursued
- FAQ
- Glossary
- Service allocation procedure
External