Docs / Testing `_AugmentedIntelligence`

Testing `_AugmentedIntelligence`

Layered test strategy for the Typed Commands tree. Prefer fast isolated suites before full-binary and live world tests.

Quick start

# Pack A — PR gate (MinGW g++ on PATH)
./tests/run_math_tests.ps1

# Orchestrated packs (math + bridges + fuzzer + command batch)
./tests/run_all_tests.ps1

# Optional MSVC SelectedFiles compile gate
./tests/run_all_tests.ps1 -CompileGate

# Live command batch when exe exists
$env:AI_EXE = 'x64\Release\_AugmentedIntelligence.exe'
./tests/run_command_batch.ps1

Test packs

Pack	Script	Purpose
A — Isolated cores	`tests/run_math_tests.ps1`	MinGW behavioural suites (math, DE catalog, algebra/calc extensions, Core.hpp modules, bridge fixtures, expr fuzzer)
B — Command batch	`tests/run_command_batch.ps1`	Structural + planned smokes for `de`/`calc`/`algebra`/`nn`
C — Compile gate	`tests/run_compile_gate.ps1`	MSVC `SelectedFiles` compile of high-risk units
D — Fuzzer	`tests/run_command_fuzzer.ps1`	Calc parser crash-only robustness
All	`tests/run_all_tests.ps1`	Orchestrates A–D

Isolated suites (`tests/suite_*.cpp`)

Built via shim tests/test_shim.hpp → AugmentedIntelligence.hpp (no OpenSSL/CUDA/MySQL).

Suite	Coverage
`suite_numeric_nn_ml`	ODE, calculus, autodiff, optimize, MathNN, MLApp
`suite_differential_eqs`	Control, dynamics, world ODE, RSI
`suite_college_algebra`	Classic college algebra kernel
`suite_ode_catalog`	Catalog models, twins, SDE, DAE, sensitivity
`suite_math_extensions`	expm, LSQ, Lyapunov, Sturm, Markov, Fourier, Padé, Green
`suite_bridge_fixtures`	MC/SE JSON protocol fixtures
`suite_expr_fuzzer`	Random calc expressions → no crash
`suite_*` Core.hpp	Reasoning, plugin, mesh, shooting, CodeRL, …

Fixtures

tests/fixtures/minecraft/*.json      # TCP JSON-lines shapes (:8765)
tests/fixtures/space_engineers/*.json # AiSe shapes (:8766)
tests/command_batch.json             # Full-binary smoke command list
tests/eval_tasks_seed.json           # Offline EvalHarness task seed
tests/eval_tasks_seed.sql            # MySQL import for cognitive_loop

EvalHarness

_EvalHarness loads enabled rows from cognitive_loop.eval_tasks and records eval_runs under a run_label. Checkers: file_exists, file_contains, output_contains.

mysql ... cognitive_loop < tests/eval_tasks_seed.sql

Then from the app (when wired):

eval run baseline

Use pass rate over time as an ungameable RSI metric.

Safety rules

Synthetic RL smokes before live Minecraft / SE / desktop bots
Dry-run flags on bridge actions
High-stakes domains: human approval before actuation
SDE/RL tests use seeds or statistical bounds, not bit-exact equality

CI recommendations

Job	Command	Notes
PR	`run_math_tests.ps1`	Requires MinGW g++
Nightly	`run_all_tests.ps1 -CompileGate`	MSVC + MinGW
Release	Full CPU-AVX2 build + EvalHarness + manual live dry-run	When link is healthy

Extending

Pure logic → *Core.hpp + suite_mymodule.cpp
Register in run_math_tests.ps1
Add smoke lines to command_batch.json
Optional EvalHarness SQL row

See Developer-Extensibility.md tested-core pattern.

Generated from the project markdown docs on 2026-07-24. This is a static, self-contained site.

Testing _AugmentedIntelligence