Appearance
The engineering record
Internals. Nothing on this page is needed to install or play aowlspt — start at Installation for that.
This page is the material that used to sit on the aowlspt front page: what the project is made of, the gate that runs over it, and what the automated tests do and do not establish.
Why it is on this site
Every other page in these docs documents a piece of the toolchain. aowlspt is something the toolchain built — and it is the largest such thing, so it is also the most direct answer available to "can you write a real program in nimony." The numbers, measured 2026-08-19:
| nimony source in the repository | 139,587 lines across 202 modules |
| of which: mods | 68,209 lines, 119 modules |
of which: tooling (aowl and the gate binaries) | 34,194 lines, 24 modules |
| of which: hosts + backend + the mod library | 23,090 lines |
| header-only C, hand-written, under all of it | 12,855 lines across 15 headers |
| documentation in the repository | 13,491 lines of Markdown |
aowl test, end to end on a quiet tree | 143 checks, 0 errors |
Every host, the backend, the installer, the build tool, the release tool, every gate binary and every mod is nimony. One C# program is left — tools/SptReflect, which reads SPT's assembly metadata and is managed because the thing it reads is.
The case for reading it as a case study is on its own page: what the language made easy, what it made hard, what broke, and the long list of things this project says it has not proven.
Why the design is what it is
Post-1.0 Tarkov changed the client's scripting backend. Pre-1.0 clients are Mono: Assembly-CSharp.dll is an ordinary .NET assembly, BepInEx loads into the Mono runtime, and Harmony patches methods by name. Post-1.0 clients are IL2CPP — no managed assemblies at all, only GameAssembly.dll and global-metadata.dat. Nothing built for one loads into the other, and the whole existing modding ecosystem is built for the one that is gone.
The usual answer is to roll the client backwards to a pre-1.0 build. aowlspt's installer refuses to do that, and --force does not enable it. So it does the other thing: it hosts mods in the post-1.0 client directly, as a native DLL driving IL2CPP's own exported C API, with no BepInEx, no Il2CppInterop, no generated proxy assemblies and no second runtime in the process.
That decision is what shapes everything else. There is no managed layer to hide in, so the reach into the game is native x64 inline detours and a bound method-pointer fast path — and the price of every one of those is measured rather than asserted. See Reaching into IL2CPP.
The gate
aowl test143 checks, 0 errors, measured end to end on a quiet tree on 2026-08-19 — with aowl.exe rebuilt first, which matters, because the gate does not rebuild itself and an earlier run that day silently used a binary predating every phase it was meant to exercise.
Each of those checks is a named external step. Underneath them sit the suites, each reporting its own count, each with a loopback port of its own so that none of them can pass against a server somebody already had running:
| suite | checks | what it drives |
|---|---|---|
soak, 24 closed play cycles | 915 | the game server over a long session, checked by invariant |
emutest | 577 | the game server through the client's boot sequence, against a fixture database |
livectl | 213 | a mod disabled and re-enabled while the backend is serving |
realtest | 168 (1 skipped) | the server against a 39.40 MiB database imported from a real SPT install |
fuzzwire | 152 | what the server does when the client is hostile |
sain selftest | 124 | the rewritten bot decision layer, with no Tarkov under it |
framelen | 102 | hostile Content-Length framing |
modclient | 90 | the manager reading the client host's ledger |
firstrun | 64 | a synthetic vanilla client → install → importdb → verify → boot → restart |
pttguard | 62 | Path To Tarkov's graph against the real database |
wstest | 62 | the notifier websocket: does a push reach the session |
hostharness | 28 verdicts | the client host against a stand-in IL2CPP runtime |
ovsync | 24 | the client host's poll actually reaching the backend |
tickfault | 22 | a mod whose on_update fails |
ws_stall | 15 | a websocket peer that stops reading |
aowl-coverage --check | 12 | the coverage document against the code |
detour_race | 11 | a patch installed under live callers |
storeguard | 9 | a reader holding a key across a commit |
Two of those rows are worth more than their size. hostharness is counted as 28 verdicts and not by grep -c '^ok', which answers 31 — three of those lines are the harness reporting its own staging before it has read a line of the host's log. And storeguard's 9 is annotated with the fact that 2 of them fail against the store as it was, because a handle that never sees a change is trivially never torn. That annotation is the house style, and it is the subject of the case study.
What the gate does not settle
aowl testruns the client host againsttests/mockil2cpp, a stand-in that exports the same C API asGameAssembly.dll. Resolve, call-with-arguments, live objects, fields and patching are all exercised for real against it. What that cannot establish is that BSG's implementation behaves identically. Every performance figure on these pages is measured against that stand-in — the real boxed/bound ratio is expected to be larger than the published table, not smaller.
That gap is closed by launching the game rather than by adding tests, and it has been: the system now installs onto a real post-1.0 client, boots it, and plays offline raids. What each test suite is still worth is exactly what it says on its row — an invariant checked cheaply, on every build, before anyone launches anything.
Historical note, because it appears throughout older material in the repository: until 2026-08-19 the most repeated sentence in the project was "Nothing in this repository has ever run against BSG's client. Not once." It appeared at the top of seven documents, all ten mod READMEs, the overlay README and the module header of every client-side mod. That sentence is no longer true, and wherever it still survives it is stale text, not a current claim.