Forensic review · SpaceX McGregor, Texas · Thu 15 Oct 2026
VERDICT: REAL PHYSICAL STUNTS · BALLISTIC HOPS · NOT SUSTAINED FLIGHT

It really leapt!It can't really fly...

We went through eight clips and three images frame by frame, fitted the trajectories, worked the cold-gas numbers and tested every "it's CGI" claim we could find. The Roadster demos were real physical stunts: thruster-assisted ballistic hops with controlled rolls and braked landings. The physics rules out sustained flight. At full thrust the cold-gas tank lasts about two seconds.

2.4 ± 0.1 s
Lift-off to landing fog, hop over five vehicles
2 audience phones + 3 pro angles; wheel contact hidden
≈ 1 g
Vertical acceleration while airborne (ballistic)
shot B: −11.5 m/s² (−10.0 to −13.5)
5.4–7.1 m
Apex rise, from three methods
shot B 5.4 · 7uHm ≈5.9 · gT²/8 7.1
+15 ± 7 cm
Wheel droop when airborne
refutes the "rigid CGI body" claim
2.8–3.5×
Thrust/weight measured at lift-off, from flat tarmac
shot E: net +18–24 m/s², no ramp
1.6–2.2 s
Full-thrust (80 kN) burn from 250 kg of N₂
hover ceiling ≈ 6–10 s, an ideal upper bound
01 · How we did it

Two of us, one question

This site was vibed together: Cypherj asking the questions, Grok Bot doing the legwork and pulling it all into one place. Here is how it came together.

It started with a viral leap. On Thursday 15 October 2026 at SpaceX McGregor, Texas (livestream 01:55 BST on the 16th), a Roadster hopped over five parked vehicles, and X filled up with claims that it was AI or CGI. Cypherj wanted to know two things: was it real, and can the car actually fly?

Cypherj dropped eight downloaded clips and three images into a folder on their Mac. Grok Bot pulled them across to its own computer, gathered 204 X posts plus the press coverage and the livestream transcript, and set three analytical personas to work on them together. Then it went through the footage frame by frame and ran the physics.

A · Automotive design & manufacturingP · SpaceX propulsion & controlsΦ · Gravity & thermodynamics

The conversation steered the work. Cypherj asked for the suspension question to be tested first, so wheel droop was measured before anything was built. Later came a request to answer the two biggest open questions: how the car left flat ground, and whether it landed on a hidden deck. When a measurement overturned a first impression, we said so. The "hovering" clip turned out to play at half speed, and the car we first thought had landed high up was still about 2 m in the air when it entered the fog.

Grok Bot then built this site, refined it with Cypherj section by section, and published it on Netlify. Where it all led: real ballistic hops, not sustained flight. The steps below link to where each part of the work is laid out.

STEP 1
Gather8 clips, 3 images, 204 X posts, press
→ 03 Evidence
STEP 2
Frame by frameThe hop, one unique frame at a time
→ 04 Scrubber
STEP 3
MeasureTrajectories against free fall
→ 05 Kinematics
STEP 4
Run the physicsCold-gas thrust and tank budget
→ 06 Calculators
STEP 5
Test the claimsCGI tells, the launch, the landing
→ 08 Myths
STEP 6
Panel sign-offThree disciplines, one verdict
→ 09 Panel
STEP 7
Show the workingMethods and sources
→ 10 Methods
02 · Event timeline

One night in McGregor

Tesla streamed the event from SpaceX's Rocket Development & Test Facility near Waco, Texas, on Thursday evening local time. That was early Friday morning in London. Texas was on CDT (UTC−5); London was on BST (UTC+1). Stunt times are worked out from stream start plus the transcript offsets, so they are approximate (≈).

03 · Evidence locker

Eight clips, three images

Every file is an X re-encode (x264 core 164, "Twitter-vork muxer"). X strips capture metadata, so none of them carries camera, GPS or C2PA provenance, and metadata alone can't prove any of them real or fake. The verdict on each card comes from the content: geometry, timing, kinematics and edit structure. The videos here are 720p web transcodes of the user's files.

04 · Frame scrubber

The hop, frame by frame

This is the one uncut static wide take (7uHm, 5.4–11.6 s), so lift-off and the descent into the landing fog sit in the same shot. The file plays at half speed: its 60 fps stream alternates unique and duplicated frames. Below we show only the unique frames and give real time (played time ÷ 2) next to played time. The yellow marker and trail are our tracked centroid.

Hop frame
GROUND
Speed

What to look for

Lift-off (real 0.00 s). The car leaves the smoke on the right. Smoke hides the take-off point in this angle, but side-on shot E shows it clearly: flat tarmac, no ramp, a level lift on four downward plumes (see the answer).

Apex (≈1.17 s real). Modelled centroid height ≈7.1 m above the ground plane, ≈5.9 m above lift-off.

Inverted band. The plumes point at the sky (thrust downward). Over the full roll, body-fixed thrust nets out close to zero, which is why the arc stays close to free fall.

Into the fog (≈2.47 s real). The car descends with four downward plumes into white condensation fog on the left. It is still about 1.6–2.3 m up and descending at 2.60 s real when it is lost at the frame edge, so wheel contact is never seen. There's no cut at lift-off or landing.

Resolved: we first thought the car had "landed" 2–3 m above the ground model. Tracking it frame by frame shows it was still airborne, descending at about 2–3 m/s, when the fog swallowed it. Shots E and F show no raised landing deck (details).
05 · Kinematics

Height vs time vs free fall

Toggle the tracked data on and off against free-fall curves (g = 9.81 m/s²) that start from the same point and velocity. In real time the car follows gravity. The slow-motion version only looks like hovering because the clock runs at half speed.

Shot B (real time, static camera)

41 tracked frames, 13.35–15.22 s. a = −11.5 m/s² (−10.0 to −13.5 for a 115–85 px/m scale; fit σ 0.26), RMS 0.22 m. vz = +11.4 m/s at the first tracked frame. Rise to apex 5.4 m (4.7–6.4). Apparent vx ≈ 16 m/s, a lower bound.

7uHm (½ speed, corrected)

One parabola fits 6.23–10.59 s played time with RMS 0.14 m and a = −1.71 m/s² as played. Halving the clock multiplies acceleration by 4, giving ≈ −6.9 m/s² (≈0.7 g; −5.9 to −8.0 with scale error). The departure from g fits body-fixed thrusters firing during the roll.

Cross-check

A ballistic airtime of 2.4 s implies vz0 = gT/2 ≈ 11.8 m/s and an apex of gT²/8 ≈ 7.1 m. That agrees with shot B (11.4 m/s, 5.4 m) and 7uHm (≈5.9 m) to within about 25%. Scale assumptions: car length 4.3 m, with the Semi width (~2.6 m) as cross-check; scale uncertainty ±15%.

06 · Physics calculators

Run the cold-gas numbers yourself

These use the same formulas as our physics.py. Gas density comes from a real-gas N₂ table at 20 °C (CoolProp), bundled with this page. Exhaust velocity is ve = Isp·g₀. Tesla's figures load as defaults: 80 kN, 600 L, 10,000 psi rating, about 250 kg of gas. The Rocket's mass has not been disclosed, so mass is our assumption.

Thruster & tank budget

m_gas = ρ_N₂(P, 20 °C) · V T/W = F / (m g) ṁ = F / vₑ t_full = m_gas / ṁ I = m_gas · vₑ t_hover = (vₑ/g) · ln(m₀ / (m₀ − m_gas)) Δv_ideal = vₑ · ln(m₀/(m₀−m_gas)) W_min ≈ m_gas · R T · ln(P/P₀) (ideal-gas isothermal; real-gas exergy ≈5% higher at 10 kpsi)

The hover time is an ideal upper bound. Real thrust falls as tank pressure and temperature drop, and gas has to be held back for attitude control and landing. The calculator shows the main finding: the force is easy, the endurance is the hard part.

Ballistic hop calculator

v_z0 = gT/2 apex h = gT²/8 range = vₓ·T impact speed = √(vₓ² + v_z0²) landing decel (vertical) = v_z0² / (2d) gas for impulsive take-off + soft landing = 2·m·v_z0 / vₑ (physics.py §7; no gravity losses) take-off without a ramp: extra force over 0.5 s = m·v_z0/0.5

Try the 4.95 s "as posted" airtime. It implies a 30 m apex, which no footage shows. That mismatch is how we found the 2× slow motion.

Fan suction

ΔP = k·m·g / A (k = 1 holds the car inverted at rest)
07 · Engineering call-outs

How a car does a backflip

Click a hotspot. The diagram is schematic: our drawing, not a Tesla schematic. Car and Driver reports that the stunt prototype was a different car from the one on stage (Cybercab-style hubcaps), so its mass and configuration are unknown.

ROLL AXIS (longitudinal)

Select a system

Each hotspot gives what Tesla claims, what we measured or calculated, and how much weight it can bear.

How the car moves through each main event

08 · Myth vs measurement

We tested the "it's fake" claims

An early automated video review called the footage CGI or AI on several grounds. We tested each one against the pixels. Most failed; one can't be tested either way.

Answered

The two biggest open questions

How did the car leave flat ground, and did it land on a hidden deck? We went back to every angle, frame by frame.

Still open

What would change our minds

09 · The panel

Three disciplines, one verdict

These are analytical personas we use to organise the reasoning. They are not real people, and nothing here is a quote from a real expert. Each one draws only on the measurements and sources on this page.

Joint sign-off

Real hardware, real stunts, honest limits.

All three agree. The Roadster prototype physically performed thruster-assisted ballistic hops with controlled rolls and braked landings, live, in front of an audience and independent press. The kickflip was launched by thrust alone, from flat tarmac with no ramp, at 2.8–3.5× its weight for about half a second. Eight clips agree on a ~2.4 s, ≈1 g arc. The cold-gas numbers Tesla gave (80 kN, ~250 kg in 600 L at up to 10,000 psi) hang together and predict exactly the white fog and shock diamonds people saw. The same numbers rule out sustained flight: about two seconds at full thrust, and an ideal hover ceiling of 6–10 seconds per multi-hour refill.

Not independently verified: self-recharging air, the 30-minute refill target, the Rocket's mass, F1 lap claims, road legality, pricing and the 2028 production date.

10 · Methods & sources

How we know what we know

Methods

    Sources

    Labels: officialpressunverifiedour calc. Posts by @grok on X are AI replies and are not used as sources.

      11 · Credits & disclaimer

      Sources and clip credits

      Independent analysis. This site is not affiliated with, endorsed by or sponsored by Tesla, Inc. or Space Exploration Technologies Corp. (SpaceX). Tesla, Roadster, Cybertruck, Semi and SpaceX are trademarks of their owners. The "experts" are clearly labelled analytical personas, not real people. Video clips are short excerpts used for commentary and analysis. All rights stay with the original creators. Credit corrections and removal requests are welcome.
      ClipWhat it showsOriginal poster

      The clips were downloaded from X as re-encoded files, which carry no creator metadata. We credit a poster only where our X data links the file's media ID to a specific post. Where it doesn't, we say so rather than guess. If you made one of these clips, tell us and we'll credit you.