Handball Manager

A women's handball management simulation for Android, live on Google Play. Twenty-three countries' leagues and three European competitions run inside one world model — every club in every country plays its own season, whether or not the player ever looks at it — and the player's own matches are played live by an agent-based 2D match engine.

Built solo, end to end: the world simulation, the 2D match engine, persistence, the front ends, the test suite, the data collection, and the store release.

Google Play · Releases (APK / Windows) · Engineering notes · The 2D engine's source · Code samples

The game is closed source ahead of a commercial release. This repository holds the builds, the write-ups, the full source of the 2D match engine to read in engine2d/, and two MIT-licensed extracts in samples/. The complete source is available to reviewers on request — get in touch through the email on the Google Play listing.

_{A European League night on the 2D court, at normal speed: every player is an agent deciding ten times a second.}

A European week: the whole app in the competition's colours	League table with title, European and relegation places	The EHF European Cup, two-legged from Round 2
EHF coefficient rankings deciding each country's places	Player profile: attributes, contract, buyout clause, transfer list	The manager career: licences, a wage and your own balance
23 countries' leagues to start a career in	Every club with its hall, standing and honours	Club page: reputation, form and home arena

---

Stack

Language / runtime	C# 13, .NET 10
Mobile	.NET MAUI (Android, API 24+)
Desktop	WPF (Windows) — no longer updated; the model layer is shared
Persistence	EF Core 9 over SQLite, SQLitePCLRaw bundled for Android
UI pattern	MVVM, CommunityToolkit.Mvvm source generators
Tests	xUnit

Shape of the codebase

Project	Lines	Files	What it is
HandballManager.Core	65,500	199	Simulation, the 2D match engine, persistence, and every view model. References no UI framework.
HandballManager	14,300	91	WPF desktop head — views only, no longer updated
HandballManager.Mobile	16,900	149	.NET MAUI head — views, the 2D court renderer, session plumbing, Android packaging
HandballManager.Tests	17,500	70	xUnit; about 1,500 test cases, many of them full-season integration runs

51 services and 30 tables in Core. Putting the view models there rather than in either front end is the decision the whole project rests on: both heads bind to the same HomeViewModel, so a feature is wired once instead of twice, and the desktop app became a mobile port without the model layer being touched. The trade-offs are in Architecture.

The simulation

The world advances one day at a time. On any date the engine plays every fixture in every competition, ages and trains ~4,800 players, runs the AI transfer market, pays wages, sells tickets, and writes the news and statistics behind all of it.

Three problems that took the most work:

• Competition shape is data, not code paths. Twenty-three countries do not play the same game administratively. Montenegro's eight clubs meet three times. Slovenia's thirteen meet once — one club idle every round — then split into two groups carrying points down by finishing position. Türkiye splits and carries each club's points in full, so a group table where a club that won nine of twelve finishes second is correct. Poland pays three points a win and settles every drawn match with a shoot-out that leaves the scoreline a draw; Portugal pays a point for a defeat, and splits its table carrying half of everyone's points, rounded up.
• The default failure mode is silence. An eleven-club league generates no fixtures rather than throwing. A competition with no prize-money tier pays nothing to anyone, forever. A country with no youth name pool quietly produces Romanian teenagers. Most of the work on each new country goes into turning that class of bug into a test that fails loudly instead.
• European qualification has to stay exact. Three competitions, each with its own three-season coefficient ranking, however those rankings move and however many countries exist. The Champions League is sixteen: nine guaranteed places, an incentive place for the European League's leading country, and upgrades walked back down the ranking. The European League takes banded places (four, three, two or one by rank) into a field of 28 to 52, and the field size alone decides who plays Round 2, who enters at Round 3 and who goes straight to the groups. The EHF European Cup gives each country what is left of four places after the other two, tops up to thirty-two down its own ranking, and fills a short round with lucky losers. Written as rules rather than lookup tables, which is why the totals hold when a country is added. The original two-competition allocator is in samples/ with its tests.
• A save is more than what is stored in it. Careers run for dozens of hours and have no backup, so every release has to open every older save untouched. New columns are the obvious risk; the subtle one is what is rebuilt on load. League rounds carry no date — the calendar is recomputed each session and moves rounds off European weekends onto Wednesdays the cup has not claimed — so when 3.5 replaced six countries' group-stage cups with their real knockout formats, changing the cup's dates would have silently moved unplayed league rounds in seasons already underway. Old draws finish on the old path and the old calendar, held there by a test against a snapshot of the previous release's output, taken before the change was written. 4.0 did the same again for Europe: 253 calendars snapshotted before the European League's qualifying rounds and the European Cup moved any date.

The 2D match engine

_{A seven-metre stops the match to ask who takes it, ranked by the attribute that decides it.}

Panels folded: the court takes the screen

Tactics from the bench, applied from the next attack

Since 4.0 the player's own matches are played on a 2D court by an agent-based physics engine. Every player is an agent that perceives, decides and moves ten times a second: movement from pace, acceleration, agility and stamina; shots as geometry, with the release point, the aim error from finishing and composure under pressure, blocks by arms actually in the path, and a keeper's reach against the ball's flight time; passes that can be intercepted or fumbled; and referees applying the rules to what happens rather than rolling for it. The carrier prices shooting, passing, driving and holding in goals; defences play 6-0, 5-1, 3-2-1 or man-marking. The other bench is an AI coach that reads the score and the clock.

The hard part was not making it move but making it agree with the rest of the game. Every other match in the world is still simulated by the statistical engine — a season is thousands of them — so a watched match and a simulated one have to produce the same handball. The physics engine is calibrated over hundreds of matches to the statistical one: about 62.6 goals a match against 62.9, seven-metres and suspensions per team within a few tenths, and the same home-points share at every gap in squad strength. It only matched that curve once each match had its own day form; without it, the stronger side won too reliably. The anchors are pinned by a test that fails when a tuning change drifts them.

The engine's full source — about 7,000 lines, the simulation, the court renderer and its tests — is in engine2d/, with a reading guide.

Performance

Simulating a day is the hot path, and it degraded over a career rather than being slow up front — the more interesting kind of problem. SimulationProfiler (in the test project, inert unless HM_PROFILE=1) times each phase of a day against a database with real mileage on it.

	before	after
400 simulated days	449.8 s	166.1 s
avg day, first 20	947 ms	322 ms
avg day, last 20	1,276 ms	355 ms
EF-tracked entities by June	161,372	11,805

The causes were an EF change tracker that never released finished match rows, so every save walked a graph that grew all season, and player training running daily across every player in the world instead of weekly. Later passes found DetectChanges costing 23.7 ms per call — against 1.37 ms for a squad load — and four callers saving unconditionally on days with nothing to save.

Two plausible hypotheses measured negative and were dropped: indexing the match table (noise), and storing less match detail for other countries' games (~7%, inside run-to-run variance — the cost is deciding what happened, not writing it down). Full write-up: Making a career fast.

Testing

About 1,500 xUnit test cases, deliberately weighted towards integration:

• Season-completion tests hand the player a random club, play a whole season through the real view-model commands, then assert that nothing anywhere in the world is left unplayed. It exists because the world-simulation loop once looked only for the player's own supercup, leaving two other countries' finals unplayed all season with no error anywhere.
• Seed-data tests enumerate the embedded club data and check squad shape, position-correct attribute blocks, present crests and arena photos, prize tiers for every competition, real head coaches, and that nothing seeded narrates a season the player is about to play.
• Fixture-clash tests prove, without simulating anything, that no competition can schedule a knockout tie onto a date its own league already uses.
• Balance-report tests produce the cross-league comparison described below.
• Compatibility tests open a deliberately old-schema save through the paths that read it before upgrading it, and hold a season still playing an old cup format, or the old European formats, to the calendar the previous release produced, day for day.
• Engine tests play the 2D engine headless: determinism for a seed, knockouts that always produce a winner, the calibration anchors, and the match screen's view model at 32× — the court never shows a result before it happens.

The regression tests added for the bugs players reported against 3.4 were each run with their fix switched off first, to see them fail, before being kept.

The full suite plays roughly twenty seasons of a multi-country world and runs over an hour, which is its own design constraint — the seed-data subset runs in about 20 seconds for the iteration loop.

Data

253 clubs and 4,826 players with real squads, birthdates, positions, shirt numbers, heights and head coaches, assembled from twenty-three countries' federations plus the EHF. Very little of it was offered as a download; most came out of undocumented JSON behind Angular and Vue front ends, JavaScript object literals that are not quite JSON, and base64 PDF match reports.

Fetching it is the easy half. Federation sites drift to the current season while the game is pinned to 2025/26; one public table was two seasons stale with nothing on the page saying so; one federation registers a goalkeeper born in 2015 with eleven appearances; one reigning champion enters its league with an under-18 side and fields its internationals only in the final. Where positions are unpublished, shot-zone distributions classify them at 100% — but only where zone data exists, since on goals and shots alone the best classifier managed 45% against a 35% baseline. Full write-up: Building a world from federation data.

Balance

Twelve leagues rated country by country over several months have no agreed scale between them, and a domestic table cannot measure one league against another. Cross-border European matches are the only instrument, so the European competitions doubled as the tooling for the balance problem: five seasons with transfers off, cross-border results only, targets set on the raw attribute scale. The countries added since were pitched the same way before they shipped: squad strength set between two existing leagues, then checked on simulated seasons of cross-border results.

The most useful result was negative — between two runs with no rating change at all, one league went from three different champions to the same club winning five from five. Five seasons cannot resolve a title race, and a clean sweep is not evidence of a broken league. Levelling twelve leagues.

Engineering notes

• One core, three heads — the project split, why the view models live in Core, EnsureCreated versus migrations and what that costs forever, and the re-entrancy rule a shared DbContext imposes.
• Making a career fast — profiling a simulated day, the fixes worth 2.7x, and two well-reasoned hypotheses that measured as noise.
• Building a world from federation data — sourcing the first 2,698 players, the four ways the data lies, and the validation suite that rejects it.
• Levelling twelve leagues — measuring leagues against each other, and how many seasons it takes before a result means anything.

Code samples

The 2D match engine is published in full, to read, in engine2d/.

Two pieces of the simulation, extracted with their tests and MIT licensed — samples/:

• RoundRobinScheduler — a season's fixture list for any number of clubs and any number of meetings, with byes where the division is odd
• EuropeanPlaceAllocator — splitting 16 and 32 European entries between countries by coefficient ranking (the two-competition version, before 4.0)

The game itself

Twenty-three countries, 60 competitions

	Country	Competitions
	Romania	Liga Florilor · Cupa României · Supercupa României
	Hungary	NB I · Magyar Kupa
	France	Ligue Butagaz Energie · Coupe de France
	Denmark	Kvindeligaen · Landspokalturnering · SuperCup
	Norway	REMA 1000-ligaen · Norgesmesterskapet
	Germany	1. Bundesliga Frauen · DHB-Pokal · DHB Supercup
	Slovenia	1. SRL ženske · Pokal Slovenije
	Croatia	1. HRL Žene · Hrvatski kup
	Montenegro	Prva liga · Kup Crne Gore
	Sweden	Handbollsligan · Svenska Cupen
	Türkiye	Kadınlar Süper Lig · Kadınlar Kupası · Süper Kupa
	Poland	ORLEN Superliga Kobiet · Puchar Polski · Superpuchar
	Czechia	MOL Liga, then a Czech title play-off · Český pohár žen
	Slovakia	MOL Liga, then a Slovak title play-off · Slovenský pohár žien
	Spain	Liga Guerreras Iberdrola · Copa de la Reina · Supercopa de España (from 2026/27)
	Switzerland	Swiss Premium League 1 · Mobiliar Handball Cup
	Serbia	ARKUS liga · Kup Srbije · Superkup Srbije
	Austria	WHA Meisterliga · ÖHB Cup · WHA Supercup
	North Macedonia	Super Ženska Liga · Kup na Makedonija
	Portugal	1ª Divisão Feminina · Taça de Portugal · Supertaça Feminina
	Greece	A1 Women · Greek Cup · Greek Super Cup
	Netherlands	Super Handball League Women · Landelijke Beker · Handbal SuperCup
	Iceland	Olís deild kvenna · Powerade bikarinn · Meistarakeppni HSÍ
	Europe	EHF Champions League · EHF European League · EHF European Cup

Squad selection and live matches on the 2D court, contract and transfer negotiation, a transfer list, scouting, youth intake and development, club facilities, finances down to gate receipts, dynamic attendance, and a full honours history for every club.

Since 3.1 the manager is a career in their own right: coaching licences studied for out of your own balance, a contract and a wage, a board that sets objectives and grades you monthly against what a club of that size should be doing, and the sack if the grade stays bad. A job market runs across every country whether or not you are in it, so being out of work is a state the game supports rather than an ending. European ties bring the competition's own colours and backgrounds to the whole app for as long as the next fixture is a European one, with an intro film once a season.

3.3 took the match itself deeper: injuries, fitness that carries from one match to the next and the squad rotation that follows from it, two-minute suspensions and seven-metre throws, a training calendar, tactics changed from the bench during a match, and a dressing room that remembers the playing time it was promised. Since 3.5 the six cups that had stood in with group stages play their real formats — straight knockouts with seeded byes, and France's pool phase for the clubs not in Europe. 3.6 added Czechia and Slovakia's joint MOL Liga, Spain and Switzerland.

4.0 is the largest release since launch. Seven more countries — Serbia, Austria, North Macedonia, Portugal, Greece, the Netherlands and Iceland — take the world to twenty-three, enough for the EHF European Cup to arrive as the third European competition and for the European League to play its real qualifying rounds. The player's matches moved to the 2D engine described above, with the whole bench to manage while they run. Version history is in CHANGELOG.md.

License & copyright

© 2026 Vieru-Potecu Cezar-Mihai. All rights reserved. No redistribution, no commercial use; all rights to the match engine, simulation logic and branding are reserved. The engine2d/ directory is published to read, not to reuse — see engine2d/LICENSE. The samples/ directory is MIT licensed — see samples/LICENSE.

Disclaimer

Handball Manager is a fan-made project and is not affiliated with, endorsed by or connected to the EHF, any national handball federation or league, or any club or player represented in it.