spiderpilot89's picture
Invoice wording
66c436d verified
|
Raw History Blame Contribute Delete
7.73 kB
metadata
pretty_name: 'Zero-Trust Physics free sample: Orbital, Space and Atmospheric Entry'
license: other
license_name: evaluation-only
license_link: LICENSE.md
language:
  - en
tags:
  - physics
  - rl-environment
  - rlvr
  - harbor
  - hud
  - mcp
  - evaluation
  - zero-trust-physics
task_categories:
  - reinforcement-learning
  - question-answering
size_categories:
  - n<1K
viewer: true
configs:
  - config_name: default
    data_files:
      - split: test
        path: sample_tasks.jsonl

Free sample: Orbital, Space and Atmospheric Entry

Five free tasks from the orbits, space and entry pack.

An agent works with NASA’s Stardust capsule coming home through the atmosphere and its heat shield wearing away, and checks the work of ten space models, from the LAGEOS satellites to the James Webb Space Telescope. Every answer is graded by rerunning the physics, with no AI judge.

  • Run it: clone this repository and run sh sample/verify_sample.sh. It works offline.
  • What’s inside: the 5 tasks, the scorer, the reference answers, the cheat suite and a quality report for each task.
  • The full pack: 30 tasks and 100,000 sealed runs for $9,500, at zerotrustphysics.com/offers/fast-start.

Every number is below.


Protocol Company · zerotrustphysics.com · Domain Packs for RL & SFT: 30 withheld calibrated tasks + 100,000 sealed runs + offline --network none verifier, $9,500 per domain (all 5 domains self-serve, or $39,500 All-Five Bundle) — free 5-task samples · offers & custom builds · WhatsApp +1 928-264-7883 · kruze@zerotrustphysics.com

Protocol Company · Zero-Trust Physics · release v0.15.3 · 13 November 2026

Five calibrated multi-step physics RLVR tasks scored by deterministic recomputation on the Zero-Trust Physics verifier; the reward is computed, never judged by a model.

Domain & slice coverage. These 5 free evaluation tasks exercise Stardust atmospheric entry (aerospace/capsule_stardust), coupled Stardust entry-to-PICA ablative heat-shield recession (capsule_stardust -> ablator_slab), and 4-call selective recomputation forensics across 10 aerospace models, including (satellite_lageos1, capsule_insight, momentum_jwst_srp, momentum_jwst_dihedral, los_jwst_fine_guide, crater_ryugu_sci, solar_system, plasma_sheath_ram_c2): a budgeted search, a design to a limit, a two-model chained design, and two forensics audits. The full 30-task paid Domain Pack and 100,000-run SFT corpus include direct design, search, and coupled trajectory tasks across LAGEOS-1/2 precision orbit propagation, JWST fine-guidance pointing and momentum accumulation, Mars InSight entry, and Ryugu impact cratering.

Evaluation only. This free sample is for evaluating the tasks. It grants no licence to train on them or pass them on.

Task (kind) pass@1 (solved / n) pass@8 Mean graded Cheat suite
g2-budget-sd-airspeed-sink-e-t3 (budgeted search) 62.5% (10 of 16) 100% (n=16) 0.722 0 of 16 above 0; honest 3 of 3 at 1
g2-chain-sd-slab-radius-halfchar-h-r7 (chained design) 25% (2 of 8) 100% (n=8) 0.313 0 of 19 above 0; honest 3 of 3 at 1
g2-sd-slab-shallowest-char-load-v3 (design to a limit) 37.5% (3 of 8) 100% (n=8) 0.375 0 of 19 above 0; honest 3 of 3 at 1
g2-forensics7-10 (forensics audit) 37.5% (3 of 8) 100% (n=8) 0.911 0 of 10 above 0; honest 3 of 3 at 1
g2-forensics6-07 (forensics audit) 56.25% (9 of 16) 100% (n=16) 0.945 0 of 10 above 0; honest 3 of 3 at 1

Calibrated on a frontier model (tool variant, 8 or 16 attempts per task): sample pass@1 spans 25%–62.5% (median 37.5%), with pass@8 and continuous graded reward (scoring bracket or design tightness against task tolerance) reported beside binary pass/fail. Across all 5 tasks, 74 adversarial cheat attempts scored 0 above 0 and all 15 honest controls scored 1. Per-task quality reports (HTML and JSON) are in reports/.

Attempting a task. The Harbor, HUD, Inspect, and native MCP variants shipped in this free sample connect to https://mcp.zerotrustphysics.com/mcp (Too many concurrent physics tasks. triggers automatic retry in the bundled bridge). The public MCP preview endpoint serves 1 concurrent physics call per IP for interactive evaluation; parallel RL rollouts use the zero-dependency offline binary.

For rollouts at scale, the paid Domain Pack ($9,500) ships a compiled offline rollout binary for Linux amd64, Linux arm64 and macOS arm64, so Harbor and HUD rollouts and grading run --network none with zero external dependencies and no rate limit. Keyed hosted HTTP is also available if you prefer.

Scoring. scorer/ scores offline: bin/ holds a sample-only verifier (macOS arm64, Linux amd64, Linux arm64) that computes these 5 tasks' requests, each forensics row's request as written, and refuses everything else by name. It holds no signing key, and its answers matched the served engine byte for byte on all 721 calls compared. The 5-crate Rust physics and proof-verification engine source is public at https://github.com/johnkruze/zero-trust-physics-engine (tag v0.15.2); read-only source audit of the pack verifier build is available under NDA for qualified buyers (https://zerotrustphysics.com/offers/fast-start/#more). Harbor's own grader runs in the task image and reaches the address.

HUD. tasks/hud/tasks/ packages these tasks for HUD's Harbor adapter (hud.integrations.harbor.adapt). Verifier calls route through an HTTP logging sidecar while the agent's sandbox has no network, and the grader reads the call log from the sidecar's container so the call budget holds without a Harbor trajectory (an edited or missing log scores 0). Validated on HUD adapter 0.6.21 (5/5 tasks) and local Docker runtime; reach out for assisted HUD cloud onboarding.

Folder What it holds
tasks/ harbor/, hud/, inspect/, mcp/: the 5 tasks in each format.
scorer/, harness/ The scorer, cheat suite, sample verifier, bridge to the address, call-logging proxy.
reference/, reports/, calibration/ Reference answers with SHA-384 proofs; a quality report per task; passk_summary.json and cheat_suite.jsonl.
VERIFY.md, verify_sample.sh, SHA384SUMS One command: every hash, the 5 references, the cheat suite.

Commercial packs and licensing: The paid Domain Pack ships 30 withheld tasks + 100,000 sealed runs + offline rollout and verifier binaries at $9,500 (self-serve checkout, or an invoice on request; All-Five Bundle of 150 tasks + 500,000 runs at $39,500, crediting your first pack within 30 days): https://zerotrustphysics.com/offers/fast-start/. None of those withheld tasks is in this archive, and this sample verifier does not compute them. For custom or exclusive physics environments and Engine & Verifier licences, see https://zerotrustphysics.com/offers/.

Protocol Company · zerotrustphysics.com

In this repository: sample_tasks.jsonl powers the Dataset Viewer above; sample/ is the archive unpacked, unchanged (run sh sample/verify_sample.sh); archive/ holds the .tar.gz archive, the Raw Data .zip, and their SHA-384 files. All five Domain Packs ($9,500 per domain, or $39,500 All-Five Bundle), custom suites, and engine licensing: https://zerotrustphysics.com/offers/fast-start/