07Fidelity & honesty
How accurate are the numbers? When to trust them, when not to. The soul of the platform.
Audience: everyone, but essential for anyone doing serious work. This page is the soul of the platform.
Most tools want you to believe they're perfect. Delta V does the opposite: it tells you exactly how much to trust every number. A tool that's honest about its limits is more useful than one that hides them. This is written into the Constitution as the Honesty Ledger.
Why "fidelity" instead of "accuracy"?#
Fidelity means how faithfully the simulation matches reality. No simulation is perfect — they all approximate. A higher-fidelity model includes more physical effects and matches reality more closely, but costs more to compute. The skill of mission design is choosing the right fidelity for the question:
- Planning a rough Δv budget? Low fidelity is instant and plenty.
- Predicting exactly where a satellite will be in three weeks? You need high fidelity.
Delta V makes the fidelity visible and selectable — and always labeled.
The fidelity ladder (orbital physics)#
The browser runs a deterministic Rust engine that climbs a ladder of physics models. The current level is shown as a badge in the Flight Deck top bar, and you toggle the rungs in the Fidelity panel.
| Level | Badge | What it models | Trust it for | The honest caveat |
|---|---|---|---|---|
| L0 | two-body (validated) 🟢 | Earth as a perfect point mass — pure Kepler orbits | Learning, Δv budgets, transfers, instant preview | Real Earth isn't a point; orbits slowly evolve in ways L0 ignores |
| L1 | two-body + J2 🔵 | Adds J2 — Earth's equatorial bulge | Sun-synchronous design, nodal drift, most LEO planning | Still no drag, no Sun/Moon; a prototype beyond J2 |
| L2 | drag (prototype) 🟠 | Adds atmospheric drag (exponential model) | Seeing decay trends qualitatively | Prototype — good to ~km, not operational |
| L3 | 3rd-body (prototype) 🟠 | Adds Sun & Moon gravity (analytic ephemerides) | Seeing third-body effects qualitatively | Prototype — good to ~km, not operational |
What "validated" means at L0: the engine is checked against analytic textbook reference values — the real ISS speed and orbital period, the classic Hohmann transfer Δv. It is not yet cross-validated against the professional tools GMAT or STK, and we don't market it as such. That honesty is a constitutional requirement.
What "prototype" means at L2/L3: the drag and third-body models are accurate to roughly a kilometre — fine for seeing the effect and learning, not for operational predictions. For that, you climb to the server.
L2+ — the Precision Engine (Orekit, server-side)#
When you need real, operational fidelity, the "Compute L2+ (Orekit, server)" button (Fidelity panel) sends your orbit to a server running Orekit — a mature, open-source (Apache-2.0) astrodynamics library used across the industry. It computes with the full instrument:
- A full gravity field (not just J2),
- NRLMSISE-00 atmospheric drag (a real density model),
- Solar radiation pressure (sunlight literally pushes spacecraft),
- Sun, Moon, and planetary third-body gravity (JPL ephemerides).
The result overlays as an authoritative green trajectory against the instant browser line, with a Rust-computed divergence number so you see exactly how much the quick model drifted. This has been verified end-to-end: e.g. a 300 km orbit decays ~2 km over 12 hours under realistic drag.
The design philosophy: the browser gives you an instant, deterministic, replayable preview (L0–L1, plus L2/L3 prototypes); the server gives you the authoritative answer (L2+). You get both — speed when exploring, precision when committing — and you always know which one you're looking at.
Why determinism matters (and why it's rare)#
The Rust engine is deterministic: identical inputs → identical outputs, bit-for-bit, on every device, every time. This is enforced with "golden bit" tests that would fail if a single bit ever changed.
Why care?
- Reproducibility — a result you compute, anyone can re-compute and verify. Science demands this.
- Shareable truth — when you send a mission link, the recipient's browser reproduces your exact scene.
- Anti-cheat — for future leaderboards/competitions, the server can replay your run and confirm it's real.
Most simulation tools can't promise this. Delta V was architected around it from day one — it's why the physics lives in one Rust engine that runs identically in the browser and on the server.
The honesty ladder spans three domains#
Fidelity labeling isn't just for orbits. The same discipline runs through the whole product:
| Domain | The honest label you'll see | What it's protecting you from |
|---|---|---|
| Orbits | L0 validated / prototype / L2+ Orekit | Trusting a quick preview as if it were operational |
| Links & coverage | "first-order link model," "quick-look" | Trusting a simplified radio/coverage estimate as a final budget |
| Conjunction (future) | "screening, not operational CA" | Confusing a rough close-approach screen with an operator's collision alert |
| Exoplanets | "candidate, not confirmed" | Calling a transit dip a "discovery" before real follow-up |
In every case the rule is identical: state precisely how much to trust the result, every time, with no exceptions.
What Delta V does not do (today, honestly)#
So you're never surprised:
- It is not STK or GMAT. It covers the ~20% of analyses that ~80% of missions need — propagate, transfer, target, access, link, coverage — not STK's full catalog (radar, EOIR sensors, aircraft/missile modeling). Missile/weapons modeling is deliberately out of scope (civil space only, by ethics and export-control prudence).
- The L2/L3 in-browser models are prototypes (~km); operational fidelity lives on the Orekit server.
- The orbital engine is validated against analytic references, not yet against GMAT/STK.
- Orbit determination, CCSDS message interop, and conjunction screening are designed but not yet built (the "Precision Desk," phase S7).
- The AI assistant is currently rule-based hints, not the full computed-answers-with-citations copilot (CAPCOM, phase S6).
None of this is hidden. It's all labeled in-product and laid out in For executives. Honesty isn't a weakness here — it's the feature that makes every number you do see worth trusting.
Next: the glossary for any term · or the engineer track for the capability map vs STK/GMAT.