EchoForge is a strict-open, radar-first, GPU-native synthetic sensing foundry. It publishes public-proxy object signatures, uncertainty-scored radar artifacts, and reproducible validation evidence so downstream work can be inspected, rerun, and compared without claiming measured truth or proprietary-equivalent sensor behavior.
The claim boundary stays narrow: public-source priors only, explicit uncertainty, hard negatives treated as robustness work, and no classified, vendor-private, or exact field-performance claims.
EchoForge Studio is a same-origin Rust + Vite interface for live radar playback, Monte Carlo campaign setup, run archive/restore/duplicate workflows, artifact review, and headless API usage. The UI keeps source cards, validation tier, uncertainty language, seed, and scenario hash visible before export.
| Command Center | Live Radar | Monte Carlo Builder |
|---|---|---|
 |
 |
 |
README media is tracked under assets/readme/studio with a deterministic capture manifest at assets/readme/studio/manifest.json. Regenerate after a Studio UI change with the same lane used by the post-merge sync:
rtk bash ops/run-lane.sh studio-sync- Public-proxy object and scenario packs that are readable as source artifacts, not hidden datasets.
- Detector realism surfaces that separate detection, tracking, classification, and fusion.
- Hard-negative and clutter fixtures that exercise robustness instead of optimizing evasion.
- Validation reports and smoke gates that keep generated outputs reproducible and reviewable.
- Source-pack fixtures: public-proxy dossiers, object cards, physics dossiers, and pack manifests.
- Detector surfaces: detector family matrices, detector cards, detector graphs, and sensor cards.
- Scenario fixtures: public-proxy scenes, airspace-object libraries, and weather profiles.
- Validation surfaces: benchmark reports, phase metrics, and smoke outputs that stay out of Git.
- Public-proxy dossier: object-packs/public-proxy/source_dossier.yaml, object-packs/public-proxy/object_card.yaml, object-packs/public-proxy/physics_dossier.md
- Public-proxy pack manifest: object-packs/public-proxy/pack.manifest.json
- Radar platform pack: object-packs/radar-platforms-v1/pack.manifest.json, object-packs/radar-platforms-v1/saab_giraffe_1x.yaml
- Hard-negative pack: object-packs/hard-negatives/pack.manifest.json, object-packs/hard-negatives/birds.yaml, object-packs/hard-negatives/rain_cell.yaml, object-packs/hard-negatives/wind_turbine_large.yaml
- Airspace-object library: configs/monte-carlo/airspace-objects.json
- Scenario fixtures: scenarios/public-proxy-clutter/scenario.yaml, configs/scenarios/uae-coastal-surveillance.json, tests/schemas/scenario.sample.json, tests/schemas/weather_profile.sample.json
- Detector family matrix: detection/detector_osint_family_matrix.py, detection/reports/detector_osint_family_matrix.md
- Detector card: schemas/detector_card.schema.json, tests/schemas/detector_card.sample.json
- Detector graph: schemas/detector_graph.schema.json, tests/schemas/detector_graph.sample.json
- Radar platform card: schemas/radar_platform_card.schema.json, tests/schemas/radar_platform_card.sample.json
- EO/IR card: schemas/eo_ir_sensor_card.schema.json, tests/schemas/eo_ir_sensor_card.sample.json
- Acoustic card: schemas/acoustic_sensor_card.schema.json, tests/schemas/acoustic_sensor_card.sample.json
- Passive RF card: schemas/passive_rf_sensor_card.schema.json, tests/schemas/passive_rf_sensor_card.sample.json
- Supporting reports: detection/reports/phase_pd_pfa.md, detection/reports/auc_table.md, detection/reports/track_lifecycle_baseline.md, detection/reports/benchmark_regeneration_report.md, detection/reports/passive_rf_eoir_passive_radar_backlog.md
The runit main run is the benchmark of record here: a 10,000-scenario-group,
30,000-phase-record fixed-wing pusher-prop public-proxy example with all
detector branches plus the layered fusion branch. It writes generated streams
and reports under outputs/; see docs/main_run.md for
the full split, leakage, and claim-boundary protocol.
IEEE-style paper: paper/echoforge_ieee.pdf.
Paper figure strip:
| Simulator stack | Phase ladder | EI evolution |
|---|---|---|
 |
 |
 |
Paper lane:
rtk python3 -m detection.paper_evidence --strict-ei-trace --force
rtk python3 paper/generate_metric_macros.py --strict
rtk python3 paper/generate_figures_focused.py --strict
rtk python3 paper/validate_visuals.py
rtk bash paper/build.sh --copy-tracked
rtk python3 paper/validate_paper.py --tex paper/echoforge_ieee.tex --bib paper/references.bib --pdf paper/echoforge_ieee.pdf --figures-dir paper/figures --paper-evidence-root outputs/paper-evidence/tier1-finalPaper artifact map:
| artifact | purpose |
|---|---|
paper/echoforge_ieee.tex |
IEEE-style source with strict public-proxy claim boundaries |
paper/echoforge_ieee.pdf |
tracked compiled paper for direct review |
paper/figures/architecture_stack.png |
readable simulator/evidence stack and model-card summary |
paper/figures/phase_method_ladder.png |
phase-wise human detector ladder through accepted fusion and EI |
paper/figures/ei_evolution_money_plot.png |
train/CV EI evolution trace with selected-lock marker |
paper/figures/false_alarm_breakdown.png |
selected-threshold false-positive burden by method and family |
paper/figures/radar_positive_vs_false_positive.png |
qualitative synthetic range-Doppler positives and false positives |
paper/validate_paper.py |
page, citation, bibliography, required-figure, and no-fallback-metadata checks |
rtk python3 detection/generate_main_run.py --profile fixed-wing-pusher-proxy-v2 --out-root outputs/training-data/runit-fixed-wing-pusher-proxy-v2-main-run --scenario-groups 10000 --seed 202605210136 --force
rtk python3 detection/run_main_run_detectors.py --data-root outputs/training-data/runit-fixed-wing-pusher-proxy-v2-main-run --out-root outputs/detection/runit-fixed-wing-pusher-proxy-v2-main-run --folds 5 --seed 202605210136 --forceFull-run holdout performance, all phases combined. Thresholds are selected on train/CV only, then applied to the blind holdout.
| method | holdout ROC AUC | holdout AP | accuracy | precision | recall | FPR | F1 |
|---|---|---|---|---|---|---|---|
layered_fusion_c2 |
0.937723 | 0.128188 | 0.986000 | 0.169492 | 0.416667 | 0.010947 | 0.240964 |
tabular_ml_baseline |
0.927521 | 0.125996 | 0.980667 | 0.090909 | 0.291667 | 0.015639 | 0.138614 |
sequence_ml_proxy |
0.917421 | 0.116239 | 0.993333 | 0.125000 | 0.041667 | 0.001564 | 0.062500 |
tactical_s_band_aesa |
0.897779 | 0.047577 | 0.966222 | 0.055556 | 0.333333 | 0.030384 | 0.095238 |
gbad_3d4d_cueing |
0.926655 | 0.040623 | 0.955111 | 0.045918 | 0.375000 | 0.041778 | 0.081818 |
high_resolution_xku_cuas |
0.898012 | 0.112345 | 0.986667 | 0.153846 | 0.333333 | 0.009830 | 0.210526 |
distributed_acoustic_cue |
0.901838 | 0.093170 | 0.987556 | 0.100000 | 0.166667 | 0.008043 | 0.125000 |
The complete generated table is outputs/detection/runit-fixed-wing-pusher-proxy-v2-main-run/performance_metrics.csv;
the JSON summary is outputs/detection/runit-fixed-wing-pusher-proxy-v2-main-run/performance_summary.json.
Advanced full-stream evolution ran against the same
outputs/training-data/runit-fixed-wing-pusher-proxy-v2-main-run corpus and
selected a locked meta-fusion winner after train/CV-only search.
| selected candidate | train/CV rank | train/CV AP | train/CV ROC AUC | holdout ROC AUC | holdout AP | accuracy | precision | recall | FPR | F1 | gate |
|---|---|---|---|---|---|---|---|---|---|---|---|
meta_fusion.top8.v2_aggressive.geodesic_odds |
1 | 0.957622026 | 0.997433776 | 0.996928025 | 0.942983898 | 0.993556 | 0.947368 | 0.789474 | 0.001140 | 0.861244 | pass |
The advanced-evolution outputs live under outputs/detection/runit-fixed-wing-pusher-proxy-v2-main-run-advanced-evolution/.
Synthetic public-proxy benchmark evidence only; these numbers are not measured truth or field-performance claims.
- Public-proxy clutter baseline: scenarios/public-proxy-clutter/scenario.yaml
- Scenario fixture: tests/schemas/scenario.sample.json
- Weather fixture: tests/schemas/weather_profile.sample.json
- Scenario config: configs/scenarios/uae-coastal-surveillance.json
rtk just fast
rtk just demo
rtk env HOST=127.0.0.1 PORT=8080 cargo run -p echoforge-studio --locked
rtk npm run web:build
rtk npm run web:smoke
rtk python3 detection/generate_ml_training.py --out-root outputs/training-data/fixed-wing-pusher-proxy-ml-training-smoke --scenario-groups 24 --seed 136 --scale-name smoke --max-time-s 150 --force
rtk bash detection/run_all.sh --smokeGenerated datasets, solver outputs, and benchmark artifacts stay under outputs/ and out of Git.
rtk npm install
rtk npm run web:build
rtk env HOST=127.0.0.1 PORT=8080 cargo run -p echoforge-studio --locked
rtk curl http://127.0.0.1:8080/api/runsCurrent same-origin Studio endpoints:
GET /api/runs,GET /api/runs/{id},GET /api/runs/{id}/artifactsPOST /api/runs,POST /api/runs/{id}/replay,POST /api/runs/{id}/archive,POST /api/runs/{id}/restore,POST /api/runs/{id}/duplicateGET /api/jobs,POST /api/jobs,POST /api/jobs/{id}/cancelGET /api/contracts,GET /api/catalog,GET /api/validation/latestWS /ws/radar: JSON lifecycle/validation/artifact/backpressure/session frames plus quantized binary scan frames
Example Monte Carlo request:
rtk curl -X POST http://127.0.0.1:8080/api/runs \
-H 'content-type: application/json' \
-d '{"scenario_id":"coastal-clutter","source_pack":"public-proxy-v1","object_pack":"airspace-objects-v1","hard_negatives":["bird_flock_dense","rain_cell"],"weather_profile":"uae_coastal_summer","detector_pipeline":"physics_cfar_track_fusion_v1","seed":2026052101,"run_count":12,"workers":4,"max_concurrent":2,"smoke":true,"validation_target":"V1 public-proxy"}'- Repo policy and agent instructions: AGENTS.md
- Architecture and stack decisions: docs/architecture.md
- Proof lanes, cost budgets, kill-switches: docs/testing.md
- Audit rubric and boundary guidance: docs/audit-rubric.md
- Validation tier definitions: docs/validation-tiers.md
- Bootstrap and smoke recipes: Justfile
- Ownership and generated-zone maps: agent/owner-map.json, agent/test-map.json, agent/generated-zones.toml
- Boundary slice definitions: agent/boundaries.toml
- Schema fixture guide: tests/schemas/README.md
- Validation surface: detection/reports/detector_osint_family_matrix.md, detection/reports/track_lifecycle_baseline.md
- Public-proxy pack entrypoint: object-packs/public-proxy/pack.manifest.json