You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Access to the DeepSpace code and its full record is granted case by case by Prometheus7 Research Institute. Tell us who you are and what you would use it for.

Log in or Sign Up to review the conditions and access this model content.

Oracle7 DeepSpace 2.4

Far enough from Earth, every question takes hours to answer, so a spacecraft has to look after itself. Oracle7 DeepSpace reads its own telemetry, finds the fault, fixes it, and writes down why. Every result below is a simulation: written down before it ran, run once, with every miss published in its verdict.

Access is by request. This repository is gated; each request is reviewed by Prometheus7 Research Institute.

What is here

Folder Version Flight code SHA-256 (recipe eval/eval_final.py:code_sha256)
v2.4/ DeepSpace 2.4 611e4d36e46d0b44ba60387d37380455d76aadf55c50604ffec5298a7ffd0ab2
v2.0/ DeepSpace 2.0 (2.0b) 4fabe84a88beb0b1e18922dd9d52e19da658093cda62aba65c6c0f8e3481dead

Each folder holds the code, its pre-registrations and the sealed results. Recompute the hash with cd eval && python -c "import eval_final as E; print(E.code_sha256())". Python standard library only; no learned weights, no GPU.

Measured (sealed runs, simulation)

  • Version 2, fresh faults: 1,683 of 1,683 faults that version 1 also knew were handled, with 0 harmful actions. On two kinds version 1 could not handle, a stuck sensor and a power reading that lies, it handled 400 of 400 each.
  • A fleet of 8 spacecraft at their real positions: in a sealed week, what one craft learned protected another 21 times. 35 deliberately wrong rules were offered, all 35 were refused, and no craft got worse.
  • Built from NASA's public record of Psyche, with no Psyche telemetry: 599 of 600 propulsion faults were handled with no harm. Where the thrusters' swirl loads the reaction wheels, opposite-twist thrusters cut wheel unloads from 46.8 to 1.4 a week, if the real torques differ in sign (an assumption, swept).
  • Version 2.4 keeps these results. Study 1 and all eight earlier spacecraft were checked identical on it.

Verdicts: v2.4/results/fleet_verdict.md, v2.4/results/psyche_verdict.md, v2.4/results/psyche_att_verdict.md.

Limits

These are simulations of spacecraft built from public records, not flight results. Assumed values are labelled and swept in each pre-registration. A rule is used on a craft only after that craft's own checks admit it.

Prometheus7 Research Institute ยท prometheus7.com/DeepSpace7

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support