Reproduce

Conformance gates

jutari conformance (~20 s)

cd jutari && julia --project=. -e 'using Pkg; Pkg.test()'

jaxtari full suite (long)

cd jaxtari && .venv/bin/python -m pytest -q

screen conformance (~23 min)

jaxtari/.venv/bin/pytest jaxtari/tests/test_screen_conformance.py

regenerate an xitari reference trace

./tools/trace_dump --rom xitari/roms/pong.bin --actions tools/fixtures/actions/pong_noop_10.txt > /tmp/pong.jsonl

Per-claim commands

The exact invocation behind every ledger row, and the artifact it writes.

Claim	Command	Artifact
RAM byte-identical to xitari on every game	`python3 tools/rom_sweep/sweep_jutari_ram.py`	`tools/rom_sweep/results_jutari_ram.md`
Screen pixel-identical to xitari on every game	`python3 tools/rom_sweep/sweep_jutari_screen.py`	`tools/rom_sweep/results_jutari_screen.md`
Second independent port (jaxtari) also matches	`python3 tools/rom_sweep/sweep_jaxtari.py`	`tools/rom_sweep/results_jaxtari_ram.md`
SOFT forward is bit-exact to HARD (Theorem 1)	`cd jutari && julia --project=. ../tools/relaxation_study/verify_soft_ste.jl`	`tools/relaxation_study/video_out/divergence_si.mp4`
GPU throughput — forward, soft mode (Pong, batched)	`python3 tools/bench_jaxtari_gpu.py --rom pong.bin`	`results/gpu/gtx1080ti.json`
GPU throughput — forward + gradient	`python3 tools/plot_gpu_throughput.py`	`jutari_paper/paper/figures/gpu_throughput.pdf`
Exact-forward region in the (α, T) relaxation plane	`cd jutari && julia --project=. ../tools/relaxation_study/dump_profiles.jl`	`tools/relaxation_study/relax_profiles.txt`
XAI demo — joystick gradient recovers “push RIGHT”	`cd jutari && julia --project=. ../tools/xai_si_gradient/si_joystick_gradient.jl`	`tools/xai_si_gradient/out/ji_grad.txt`
XAI joystick figure (plot of the computed gradients)	`python3 tools/xai_si_gradient/si_joystick_fig.py`	`tools/xai_si_gradient/out/si_joystick_gradient.pdf`
Implementation-effort timeline (from git history)	`python3 jutari_paper/paper/make_figures.py`	`jutari_paper/paper/figures/fig_timeline.pdf`
E1 · Ground-truth oracle (intervention + gradient + cross-check)	`julia --project=jutari tools/xai_study/ground_truth/oracle_intervene.jl --game pong`	`tools/xai_study/ground_truth/out/`
E2 · T3 game-concept labels (import / verify / discover)	`julia --project=jutari tools/xai_study/t3/verify_labels.jl`	`tools/xai_study/t3/out/`
E3 · Phase A — neuroscience battery A1–A8 (Kording, quantified)	`julia --project=jutari tools/xai_study/phaseA_kording/A2_lesions.jl --games core`	`tools/xai_study/phaseA_kording/out/`
E4 · Phase B — attribution / XAI methods (14)	`julia --project=jutari tools/xai_study/phaseB_attribution/saliency.jl --games core`	`tools/xai_study/phaseB_attribution/out/`
E5 · Phase C — mechanistic interpretability (10)	`julia --project=jutari tools/xai_study/phaseC_mechanistic/sae.jl --games core`	`tools/xai_study/phaseC_mechanistic/out/`
E6 · Cross-tradition leaderboard	`python3 tools/xai_study/compare/leaderboard.py`	`tools/xai_study/compare/out/leaderboard.json`
E6 · ROM-free benchmark package	`python3 tools/xai_study/compare/benchmark/run.py --method magnitude_proxy`	`tools/xai_study/compare/benchmark/out/`

Conformance gates

Per-claim commands

Rebuild this site