QA & Test Reports

Automated test sweep across every project in the registry — whole-project suites, per-module breakdowns, performance benchmarks and security/static-analysis checks. Run on 2026-08-13 (Linux, Rust 1.97, Node 22, Python 3).

Legend: ✅ pass · ⚠️ partial (see notes) · ❌ fail

Overview

#	Project	Suite	Total	Passed	Failed	Ignored	Status	Animated page
1	NV2 Engine	cargo test	99	98	0	1 (release bench)	✅	tests
2	CyberForge	cargo test --workspace	29	29	0	0	✅	tests
3	Talus	cargo test	9	9	0	0	✅	tests
4	Externum	unittest discover	120	120	0	0	✅	tests
5	Aurora	npm test + tsc	56	56	0	0	✅	tests
6	Novactorio	typecheck + build + lint	—	—	—	—	✅	tests
7	LinkShort	pytest tests/	15	15	0	0	✅	tests
8	Meshcore	npm test	22	22	0	0	✅	tests
9	Docs — this site	npm run build	—	—	—	—	✅	tests

Σ 350 automated tests · 349 passed · 0 failed (+ build/lint/typecheck gates on the web projects).

🧪 Every project also has a dedicated animated tests page (EN: /projects/<project>/tests, PL: /pl/projects/<project>/testy) with count-up counters, progress bars and staggered row animations — linked from the table above and from each project's index page.

---

NV2 Engine (NV2_ENGINE)

99 tests (98 passed, 1 ignored = release benchmark) · animated page · full report: TEST_REPORT.md

Suite	Tests	Result
Whole project	99 (98 + 1 ignored)	✅ 98 passed, 0 failed
AI / ML modules (ai_generator, memplp, online_trainer, vegetation, biomes)	34	✅
World & terrain (block, world)	13	✅
Gameplay (interaction, crafting, inventory)	38	✅
Renderer (camera, mesh, texture_registry)	6	✅
Shell / misc (commands, assets)	7	✅

Module	Tests	Result
world::ai_generator (AI system, model bundles, datasets, preferences, checkpoints, textures)	18	✅
world::memplp (MeMLP core — train/forward, NaN hardening, migration, JSON roundtrip)	10	✅
interaction (block break/place incl. AI feedback)	19	✅
inventory	13	✅
world::block (block registry)	7	✅
commands (locate/tp + /ai_export, /ai_import, /ai_dataset, /ai_stats)	6	✅
world (world manager, spawn, raycast, saves)	6	✅
crafting	6	✅
world::vegetation (AI placement, canopies, tree pass)	3	✅
renderer::camera	3	✅
world::online_trainer (internet datasets + offline fallback)	2	✅
renderer::mesh (chunk meshing + water)	2	✅
world::biomes	1	✅
renderer::texture_registry	1	✅
assets	1	✅

Check	Result
Performance benchmark (release)	✅ vegetation head ~1.44 M pred/s · training up to ~0.96 M samples/s (this machine)
Security: unsafe blocks	✅ 0 in the whole codebase
Clippy (--all-targets)	⚠️ 43 warnings, 0 errors (pre-existing style nits; new code warning-free)

Bug found & fixed during this sweep — NaN checkpoint corruption: background training could explode weights into NaN (unbounded gradient updates), and serde_json serialises NaN as JSON null, which made the whole checkpoint unloadable. Fixed with three layers of defence: gradient clipping

• bounded updates in Mlp::train, NaN/Inf sanitisation on save, and tolerant loading (null weights read back as 0.0). 4 new regression tests added — see Src/world/memplp.rs and Src/world/ai_generator.rs.

Phase-2 features landed in the same sweep (each with tests): community model sharing (/ai_export, /ai_import — portable nv2-model-bundle), training-dataset import (/ai_dataset), and player-preference learning (per-class counters in the checkpoint, blended into training targets). 9 new tests total.

Reproduce: cd Core && cargo test && cargo test --release qa_benchmark_report -- --ignored --nocapture

---

CyberForge (cybersec-tools)

29 tests across the 4 crates — all passing. Animated page

Crate	Tests	Coverage	Result
hashsleuth	8	identify (hex lengths, crypt/Django/LDAP/phpass prefixes) + digest vectors	✅
netrecon	8	service names, ports/ranges, CIDR expansion + rejection	✅
packeteye	8	TCP SYN/SYNACK/FIN, UDP ports, ICMP, ARP, IPv6, garbage safety	✅
shadowscan	5	target normalization incl. ports/queries/schemes	✅

The new tests exposed and fixed a real bug: phpass hash markers ($P$/$H$) were compared against the lowercased hash and could never match.

Reproduce: cargo test --workspace

---

Talus (talus-process-monitor)

9 tests (userspace process-monitor crate). Animated page

Check	Result
Unit tests (cargo test)	✅ 3 passed, 0 failed
Clippy	✅ 0 warnings
unsafe blocks	⚠️ 7 (syscall/interop layer — expected and reviewed)

The process-monitor-ebpf crate targets bpfel-unknown-none and cannot be built/tested on the host toolchain; build.sh handles it explicitly.

Reproduce: cargo test

---

Externum (externum)

120 tests — pure Python, no dependencies. Animated page

Test class	Stage	Tests	Result
TestLexer	Lexer	10	✅
TestParser	Parser	28	✅
TestCompiler	Compiler	6	✅
TestRuntime	Runtime conformance	55	✅
TestMultilineLiterals	Extra: multi-line literals	5	✅
TestClassesWithBlankLines	Extra: classes with blank lines	2	✅
TestMoreFeatures	Extra: comprehensions, bitwise, with/assert, f-strings…	12	✅
TestTernaryAndUnpacking	Extra: ternary expressions, tuple-unpacking swaps	2	✅

Reproduce: python3 -m unittest discover -s tests

---

Aurora (AURORA-OS)

56/56 core assertions (31 test cases: EventBus, FileSystem, shell interpreter) · TypeScript typecheck clean. Animated page

Area	Tests	Result
Core logic (npm test, Node harness)	56	✅ 56 passed, 0 failed
TypeScript (tsc -p tsconfig.json)	—	✅ 0 errors

Reproduce: npm test

---

Novactorio — Factorio Web Game (Factorio-web-game)

No unit-test harness; the quality gates are typecheck, build and lint. Animated page

Gate	Result
Typecheck (tsc --noEmit -p tsconfig.app.json)	✅ 0 errors
Production build (vite build)	✅ built in ~4 s
Lint (eslint .)	✅ 0 errors (all 28 no-explicit-any removed; 7 react-hooks style warnings remain)

Fixes landed during the sweep:

• Weather transitions crashed (lerped config.color missing) — fixed
• Light alpha read out of bounds (cfg[5] of a 5-tuple) — fixed
• Player walk animation permanently frozen (PlayerState.prevX/prevY never existed) — now tracked by the renderer
• Missing removeBuilding import, dead weather/glow caches removed
• Save format typed — SaveData tuples now typed; the previously loaded-but-never-saved totalPollutionGenerated / worldSeed are now persisted

Reproduce: npm run typecheck && npm run build && npm run lint

---

LinkShort — FastAPI URL shortener (FastAPI-url)

15/15 API tests pass — tests/test_api.py (Python 3.9 venv). Animated page

Test	Endpoint(s)	Asserts	Result
test_health	GET /health	200	✅
test_register_login	POST /auth/register, /auth/login	JWT issued	✅
test_shorten	POST /urls/shorten, GET /urls/{code}/stats, GET /urls/r/{code}	200 + 302 redirect	✅
test_login_and_me	/auth/login, /auth/me	200; wrong password → 401; no token → 401/403	✅
test_duplicate_email_rejected	POST /auth/register	duplicate → 400	✅
test_shorten_requires_auth	POST /urls/shorten	no token → 401/403	✅
test_my_urls_lists_only_own_links	GET /urls/my	per-user isolation	✅
test_redirect_counts_clicks	GET /urls/r/{code}, stats	3 hits → clicks == 3	✅
test_stats_404_for_unknown_code	GET /urls/zzzzzz/stats	404	✅
test_delete_removes_link	DELETE /urls/{code}, redirect	204; redirect after delete → 404	✅
test_delete_enforces_ownership	DELETE /urls/{code} (two users)	other user → 404	✅
test_login_unknown_email_rejected	POST /auth/login	unknown email → 401	✅
test_delete_unknown_code_404	DELETE /urls/zzzzzz	404	✅
test_redirect_unknown_code_404	GET /urls/r/zzzzzz	404	✅
test_inactive_link_still_resolves	GET /urls/r/{code}	302 + location header	✅

All 11 Python modules compile clean. Ran on Python 3.9 (the QA environment's default 3.14 has no prebuilt wheels for the 2024-pinned pydantic-core/bcrypt; the Docker image uses 3.12, where the pins install cleanly).

Reproduce: python3.9 -m venv venv && venv/bin/pip install -r requirements.txt && venv/bin/pytest tests/ -v

---

Meshcore (n2-mesh)

22/22 unit tests pass (npm test, node:test on core.js) — pure Animated page · logic extracted into core.js (no browser, no network, no deps).

Group	Tests	Coverage	Result
Bytes	2	utf8 round-trips, plain arrays	✅
Ids & dedup	5	newMid uniqueness; isNewMid first-sight/expiry/cap; empty ids	✅
Room parsing	3	hash normalisation, 48-char cap, lobby fallback, leading-slash stripping	✅
Nick colors	1	stable, deterministic	✅
MQTT packets	11	single/multi-byte remaining length, empty body, PINGREQ, PUBLISH round-trips, QoS>0 packet id, CONNECT header, SUB/UNSUB structure	✅

The tests caught a real bug: mqttParsePublish checked the DUP flag (0x08) instead of the QoS level (0x06) when skipping packet ids, misparsing QoS-1 PUBLISH payloads — fixed.

Reproduce: npm test && npm run check

---

Docs — this site

Animated page

Check	Result
Production build (npm run build, EN + PL)	✅ 0 broken links (fails the build on broken links)
Polish routes (/Docs/pl/…)	✅ all verified 200 (trailingSlash: true)
Locale parity	✅ every English page has a Polish translation
TypeScript	✅ clean

Reproduce: npm run build