Project 33

C7 — Deployment reliability evidence

Field	Value
Status	UNRESOLVED — An open question, conflict, or missing measurement blocks a stronger statement.
Strongest accepted evidence	MODEL ONLY
Evidence classes cited	IMPLEMENTATION, TEST PROCEDURE, SIMULATED
Gate reached	SOFTWARE REPRODUCTION (next: SYNTHETIC TEST)
Last review	CURRENT — snapshot recorded 2026-09-28 by tooling snapshot (no human review)

Question. Does simulated deployment succeed in at least 99.9% of trials under manufacturing tolerances, and is that simulation informative?

Current statement. A seeded 10,000-trial deployment simulation records 10,000 successes, and the denominator is now explicit. The model cannot fail its velocity criterion for a locked trial and uses a different fin and speed than C6, so the 100% figure does not discriminate between designs. No physical deployment has been observed.

Requirements

ID	Requirement	Check	Bound	Observed	Result
R-C7-RELIABILITY	Simulated deployment succeeds in at least 99.9% of trials.	machine	≥ 0.999 fraction	1	SATISFIED (simulated)
R-C7-PHYSICAL	Repeated inert deployments of a physical mechanism lock at the deployed position.	physical measurement	none stated	NOT MEASURED	NOT MEASURED

Predictions and measurements

ID	Quantity (condition)	Predicted	Model uncertainty	Measured	Abs. error	Rel. error	Criterion	Result
PRED-C7-LOCK-FRACTION	fraction of deployments that reach and hold the lock position (simulated tolerance ranges of the C7 run record)	1 fraction [simulated]	not quantified	NOT MEASURED	—	—	not pre-registered	NOT MEASURED

Known contradictions

• None on record.

Open questions

• The success criterion cannot fail for a locked trial, so the result cannot discriminate designs (D-013).

Discrepancies

• D-010 (OPEN) No shared design point across the mass, spring, reliability, and structural models
• D-013 (OPEN) The reliability simulation cannot produce the failures it counts
• D-014 (RESOLVED) The historical 10,000-trial summary could not be reconstructed from its CSV

Assumptions

• A reliability statement needs a trial count, seed, model revision, and outcome definition.

Limitations

• Simulation outcomes do not measure physical reliability.
• No empirical confidence bound is asserted.
• The simulated fin, speed, and spring treatment differ from C6 (D-010, D-013).

Measurement needed to close the next gate

• A reviewed model whose failure modes can occur, run at the shared design point.
• Repeated inert deployment trials of a physical mechanism once C1 is resolved.

Pre-registered experiments

• None.

What this claim does not establish

• Any inert physical measurement: none has been accepted for this claim.
• Physical performance of any hardware.
• Flight readiness or safety for live propulsion.

Evidence

Artifact	Class	SHA-256 (first 12)	What it is
Simulation/reliability_sweep.py	IMPLEMENTATION	08c76f04ad04	Seeded Monte Carlo deployment simulation.
docs/EVIDENCE/C7_reliability.csv	SIMULATED	3eb6d15631e4	Failure-only export. Zero rows is only meaningful together with the run record’s denominator.
docs/EVIDENCE/C7_run_record.json	SIMULATED	1fb2b424083f	Trials, successes, failures, seed, sampled ranges, success criterion, and model-structure notes.
tests/test_reliability.py	TEST PROCEDURE	59ebcda5985b	Denominator and model-structure checks; no reliability threshold is asserted by CI.

Earlier versions (failure archive)

• evidence/archive/notes-v1/C7_reliability.md — Historical C7 note.

A change to any of these files makes the last review insufficient

• Simulation/reliability_sweep.py
• docs/EVIDENCE/C7_reliability.csv
• docs/EVIDENCE/C7_run_record.json
• tests/test_reliability.py

---

All claims · Traceability · Discrepancies · Evidence Observatory

This site is open source. Improve this page.