FForge Outcome Delivery
Starship Build Engineering System

Limitations

← Back to outcome

LIMITATIONS — Starship Build Engineering System

Everything simulated, assumed, or out of scope, stated plainly.

Disclosed seams

  • DISCLOSED_SEAM — Notional vehicle data. src/vehicle.mjs (operations,

times, precedence, weld samples, tolerance chains, part counts) is self-authored notional engineering data. It is not SpaceX proprietary data, an as-built process sheet, a real station layout, or a measured production figure. It is structured like the real artifact so the tooling is realistic, but no claim is made about any real vehicle or factory.

  • DISCLOSED_SEAM — No hardware. Nothing physical was built, welded, cleaned,

stacked, or measured. All outputs are computed from the model.

  • SIMULATED — Synthetic benchmark. The 31.25% station reduction / +45.5%

efficiency result quantifies the line-balancing algorithm on the notional model. It is not a real production-rate or factory-throughput claim.

Assumptions

  • Process capability normality. Cp/Cpk-derived PPM and sigma level assume

the measured feature is normally distributed and the process is in statistical control. Non-normal or out-of-control processes need a different model.

  • RPW heuristic. Line balancing uses Ranked Positional Weight, which is

near-optimal but not guaranteed globally optimal; the verification asserts provable invariants (precedence, station ≤ takt, theoretical-min bound) and the baseline comparison rather than optimality.

Scope boundaries

  • 1-D tolerance only. The stack-up engine is linear; it does not perform 3-D

GD&T, datum-shift, or Monte-Carlo tolerance analysis.

  • Finish-to-start precedence only. The scheduler models FS dependencies; it

does not model start-to-start/finish-to-finish leads/lags or resource leveling.

  • Deterministic, no variability simulation. Cycle times are point estimates;

there is no discrete-event queueing or stochastic line simulation.

What remains unverified

  • Real-world accuracy of any number against a real Starship build.
  • Behaviour on real, messy production data (missing samples, non-normal welds,

resource contention).