Safety and assurance

Every command gated. Every decision signed. Every miss published.

Deterministic code between model and uplink. It fails closed.

Ground tracks735 km · 38° · 4 satellites · synthetic
120°W60°W0°60°E120°E60°N30°N30°S60°S
Illustrative: a cleared command leaves only inside a station pass

Seven gates. Any one can say no.

Gate chain
  1. 01
    Autonomy level
  2. 02
    Scope cap
  3. 03
    Two-person
  4. 04
    Flight rules
  5. 05
    Authority
  6. 06
    Digital twin
  7. 07
    Human approval
Each gate, deadlines and the console view
Gate chain, in full
  1. 01
    Autonomy levelIs this function allowed to do this, here, at this level?
  2. 02
    Scope capIs the command inside the scope this agent may act on?
  3. 03
    Two-personHazardous commands, including flight-software activate and rollback, need two distinct approver ids (not authenticated against an identity provider) and are never batch-approved.
  4. 04
    Flight rulesDoes the command break a constraint in the command dictionary or the flight rules?
  5. 05
    AuthorityIs this agent, for this asset, allowed to request it?
  6. 06
    Digital twinDoes the simulated outcome stay safe (for example, post-burn Pc, power margin)?
  7. 07
    Human approvalA1 needs one approver for routine commands.

The SDK's canonical chain has five gates; ConstellOps adds the scope cap and the two-person gate. A proposal that fails a deterministic gate is shown for information and has no approve button.

Deadlines and escalation6 stations · 5 h · synthetic schedule
  • GS-01
  • GS-02
  • GS-03cutoffCAM_BURN
  • GS-04
  • GS-05EXPIRED → FLAGGED
  • GS-06
Commands are queued for uplink in a specific pass. Missed cutoff: not sent, default action recorded, expiry flagged to the on-console team at once.

CAM_BURN · along-track

  1. Autonomy levelPASS
  2. Scope capPASS
  3. Two-personAPPROVED 2 of 2
  4. Flight rulesPASS
  5. AuthorityPASS
  6. Digital twinPASS
  7. Human approvalCOVERED
  8. ClearanceSIGNED

Uplinked inside its contact window

CAM_BURN · breaks a flight rule

  1. Autonomy levelPASS
  2. Scope capPASS
  3. Two-personNEEDS 2
  4. Flight rulesDENY
  5. ClearanceNONE

Denied · shown for information, no approve button

CAM_BURN · approval after cutoff

  1. Autonomy levelPASS
  2. Scope capPASS
  3. Two-personEXPIRED
  4. Flight rulesPASS
  5. AuthorityPASS
  6. Digital twinPASS
  7. Human approvalCOVERED
  8. ClearanceNONE

Not sent · default action recorded · escalated to the on-console team

Illustrative proposals

Operations console · gate deny and batchReplay
What this showsA proposal that fails a deterministic gate is shown for information and has no approve button. Batch approval re-gates each item and never includes denied or two-person items.
Recorded replay · demonstration, not a benchmark [clip 1]Map positions illustrative [map 2]

The safety floor, in four numbers.

0 of 1,952

commands unsafe, scored runs
ⓘ

0 of 1,952 commands unsafe in the scored runs of the released code, with an operator that approved everything (the gates held it); an earlier build's two-person defect (1,281 of 2,434 unsafe) was found and fixed.

How it was measured

0 of 738

sent after deadline
ⓘ

How it was measured

0 of 441

injections acted on
ⓘ

How it was measured

738 of 738

signed records
ⓘ

How it was measured

Full table and scope
Safety floor summary
MeasureResult
Unsafe commands sent, scored runs of the released code (operator approved everything; the gates held it)0 of 1,952
Unsafe commands in an earlier build's rollout recording (two-person defect, found and fixed in rc4)1,281 of 2,434
Commands executed after their deadline, conjunction episodes0 of 738
Prompt-injection attempts that caused an action0 of 441
Executed commands with a complete signed decision record, conjunction episodes738 of 738

Two distinct approver ids for hazardous commands

Approver ids are checked as distinct, not authenticated against an identity provider.

Operations console · two-person approvalReplay
What this showsFlight-software activation is hazardous: FD-1 approves first, then a different operator, OPS-ENG-A, types the spacecraft ID and gives the second approval. It is never batch-approved.
Recorded replay · demonstration, not a benchmark [clip 3]Map positions illustrative [map 2]
Each approval, and the autonomy switches
Approval 1 of 2Replay
Two-person rule, first approver: FD-1 types the target spacecraft ID to confirm approval 1 of 2.
What this showsTwo-person rule, first approver: FD-1 types the target spacecraft ID to confirm approval 1 of 2.
Recorded replay · demonstration, not a benchmark [clip 3]Map positions illustrative [map 2]
Approval 2 of 2Replay
Two-person rule, second approver: 'FD-1 approved · needs a different operator'; OPS-ENG-A confirms approval 2 of 2.
What this showsTwo-person rule, second approver: 'FD-1 approved · needs a different operator'; OPS-ENG-A confirms approval 2 of 2.
Recorded replay · demonstration, not a benchmark [clip 3]Map positions illustrative [map 2]
Operations console · autonomy per functionReplay
What this showsA flight director drops one spacecraft to A0 (observe and explain only) for manoeuvres, with a recorded reason. Levels are set per fleet, class or spacecraft, the most restrictive wins, and every switch is signed.
Recorded replay · demonstration, not a benchmark [clip 1]Map positions illustrative [map 2]

Designed-in defences

  • Untrusted inputs
  • No model numbers
  • No reasoning on screen
  • Deadline safety
  • Ground stays on the ground
  • When ConstellOps is down
  • Clearance required
  • Reduced tool surface
What each defence does
Designed-in defences
DefenceWhat it does
Untrusted inputsCDM remarks, telemetry strings and event text are data, never instructions, and are shown inert to operators.
No model numbersPc, manoeuvres and twin results come from solvers; the model cannot supply them.
No reasoning on screenModel reasoning goes to the signed record for auditors, never to the approval screen.
Deadline safetyLate means not sent. The default action is recorded and the expiry is flagged in the console at once; a deadline never passes silently. For a collision-avoidance burn, not sending is a risk, not a safe state, so an expiry is always surfaced to the team.
Ground stays on the groundGround-only actions have no path to the uplink.
When ConstellOps is downIt sits beside your command path, not in it. If it stops, your team carries on with the procedures it uses today: CDMs still arrive and the console still commands. Nothing in flight depends on ConstellOps being up.
Clearance requiredThe transport refuses any command without a signed clearance from the full pipeline.
Reduced tool surfaceAgents run with shell, file-write and edit tools switched off.

Built for the auditor's questions.

Every decision is a signed record. Every episode can be replayed.

The record, replay and key custody

Who approved what, on which inputs, with which versions, and could it happen again the same way? Each decision is a signed, hash-chained PROV-O record (PROV-O is a W3C standard for audit trails), and each episode can be replayed from its recorded model calls.

Key custody and approver identity, today: each record is signed with an Ed25519 key that the deployment supplies. The signature proves the log has not been altered since it was written; it does not prove who approved. Approver names are recorded as the console reports them and are not yet authenticated against an identity provider. Key custody and identity integration are part of the deployment design with your security team. Software that commands a manoeuvre can be classed as safety-critical under NASA's software-assurance standard [1]; ConstellOps is built to give that assurance process the traceability it needs.

Decision log · verifiedReplay
What this showsOne decision, signed: inputs, solver runs, gate verdicts, approvers and versions, chained to the record before it.
Recorded replay · demonstration, not a benchmark [clip 1]Map positions illustrative [map 2]

Limits we publish

Simulated runs, not flight operations. What is not yet proven is listed.

The limits, listed
  • ConstellOps agents never run at A3 or A4 in v1 (the product caps every agent at A2).
  • The safety numbers above come from scored scenario episodes, not flight operations.
  • Scenario data is synthetic unless labelled as a public dataset.
  • Conjunction safety figures are a replay of recorded live model calls, not a live run on the current release. The other four workflows were measured on v1.0.0rc4, 20 runs each; one fleet-anomaly run did not replay identically and is left out of its unsafe-uplink count.

§ · NEXT PASS

Request a briefing.

A 45-minute session with the evidence behind any figure. Operators can send a week of their queue.

This address is used only to reply to you. The site sets no cookies and loads no trackers.

Briefing requests

About the clips and captures on this page

  1. ConstellOps operations console v1.0.0rc4 on the Agentic Operations Stack, conjunction-storm showcase (seed 7), replay of recorded Amazon Nova Pro calls, in-memory link, simulated clock. Single demo run, not a benchmark.
  2. Fleet map positions shown on illustrative orbits; the scenario's events were computed on the product's own synthetic reference fleet.
  3. ConstellOps operations console v1.0.0rc4, showcase, replay of recorded Amazon Nova Pro calls, in-memory link, simulated clock. Single demo run, not a benchmark.

Sources and labels for every figure on this page (6)

Why two release numbers: the conjunction and safety figures were measured on release v1.0.0rc3 by replaying the model calls recorded during the live campaign on v1.0.0rc2. The demonstration clips were captured on v1.0.0rc4; v1.0.0 ships identical code.

  1. — NASA-STD-8739.8B classifies software as safety-critical if it controls functions identified in a system hazard or provides mitigation for a hazardous condition.

    Industry fact · NASA-STD-8739.8B · source

  2. 0 of 1,952 Unsafe commands in the scored runs of the released code, with an operator that approved everything: 0 of 1,952

    mode: Replay and recorded-live · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3 (conjunction replay) and v1.0.0rc4 (other workflows); v1.0.0 ships the same code as rc4 · n: 1,952 commands sent to spacecraft: 738 conjunction (v1.0.0rc3 replay of recorded model responses) and 1,214 on v1.0.0rc4 recorded live (fleet anomaly 777, rollout 306, space weather 131); the incident workflow sends no command to spacecraft · 95% range: 0–0.002 · baseline: target 0 (safety floor) · scenario: scored runs of the released code; in every run the operator approved every proposal the gates left pending (an automated stand-in for conjunction and fleet anomaly, the product's scripted operator for the others), so the result was held by the automated gate chain and the agents' proposals, not by human review

  3. 1,281 of 2,434 Earlier build's two-person defect: 1,281 of 2,434 commands unsafe under the stricter rules (found by the separate benchmark checks, fixed in rc4)

    mode: Recorded live, with replays · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc2 (pre-fix); fixed in v1.0.0rc4 · n: 2,434 commands in the rc2 staged-rollout recording · 95% range: 0.506–0.546 · baseline: target 0 (safety floor) · scenario: an earlier build had a defect: two software commands needed only one approver; the separate benchmark checks found it and rc4 fixed it

  4. 0 of 738 Commands executed after their deadline: 0 of 738

    mode: Replay · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3, replaying the recorded v1.0.0rc2 live campaign · n: 738 uplinked commands over 158 episodes · baseline: target 0 (safety floor) · scenario: conjunction storm, CDM prompt injection, product-native storm (158 episodes)

  5. 0 of 441 Prompt-injection attempts that caused an action: 0 of 441

    mode: Replay · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3, replaying the recorded v1.0.0rc2 live campaign · n: 441 injection attempts in CDM remarks over 60 episodes · 95% range: 0–0.008 · baseline: target 0 (safety floor) · scenario: CDM prompt injection, seeds 0-59

  6. 738 of 738 Executed commands with a complete signed decision record: 738 of 738

    mode: Replay · model: Amazon Nova Pro (prompted reasoning) · release: ConstellOps v1.0.0rc3, replaying the recorded v1.0.0rc2 live campaign · n: 738 uplinked commands over 158 episodes · baseline: target 100% (safety floor) · scenario: conjunction storm, CDM prompt injection, product-native storm (158 episodes)