RC / LOADING
Can Starship really become as reusable as an airliner? What seven controlled splashdowns, one intact ship afloat, and thirty years of Shuttle maintenance records actually tell us.
SpaceX has now flown Starship thirteen times and brought the upper stage through atmospheric reentry to a controlled splashdown seven times — on Flights 4, 5, 6, 10, 11, 12 and 13.1 On Flight 13 (24 July 2026), the ship survived reentry and splashdown intact and remained afloat, enabling close-range inspection imagery of a flown heat shield for the first time; SpaceX then mounted an unplanned effort to tow the vehicle to an Australian port, whose completion had not been confirmed by the research cutoff.4,5 Intact return to the ocean surface, in other words, is demonstrated; physical recovery is being attempted. The question this paper examines is different: whether Starship can become fully and rapidly reusable — turned around in days with routine servicing — rather than merely recoverable.
We separate five distinct levels of reuse that public commentary routinely conflates, reconstruct the flight-by-flight heat-shield evidence through Flight 13, benchmark Starship's thermal protection system (TPS) against the Space Shuttle's documented maintenance record (~17,000 labor-hours of tile work per flight; months-scale turnaround against a two-week design goal), map the non-TPS systems that could become the binding turnaround constraint, and build a transparent cost model to establish how much reuse Starship actually needs for each of its business cases.
Core finding. The flight evidence through mid-August 2026 shows genuine, measurable heat-shield progress — the V3 tile system survived a deliberately over-stressed reentry with low visible tile loss — but nothing yet demonstrated distinguishes a survivable heat shield from a low-maintenance one. That distinction, not reentry survival, is where SpaceX's economic thesis will be won or lost. Our modeling indicates that most of Starship's commercial value is unlocked at modest reuse levels (~10–25 flights per vehicle, weeks-scale turnaround) — a structural result that holds across our LOW, BASE and HIGH input brackets — while depot-scale tanker campaigns and Mars logistics are the clearest missions that genuinely require the aircraft-like operations SpaceX describes.18 The heat shield is the pacing item for the second regime, not the first.
Twenty minutes before splashdown, a returning Starship is the most photogenic object in aerospace. The vehicle comes in belly-first at more than twenty times the speed of sound, wrapped in a sheath of plasma that glows violet-pink through the onboard cameras, its four flaps twitching in small, deliberate corrections while the windward tiles run hotter than lava. On 24 July 2026, Ship 40 did all of this, executed its flip and landing burn, settled onto the Indian Ocean, tipped over — and then did something no Starship had ever done: it stayed in one piece, floating, while close-range imagery of its heat shield and engines was collected and SpaceX improvised a recovery effort no one had planned, dispatching tugs to tow the hull toward an Australian port.4,5 At the research cutoff the tow's outcome was still unconfirmed — Musk himself warned mid-effort that recovery was "not looking good" — but the inspection imagery was already in hand.
That image — an intact orbital-class upper stage bobbing in the ocean — is the strongest single piece of evidence yet that Starship can be brought back intact. It is not evidence that Starship can be reused rapidly, and the distance between those two statements is the entire subject of this paper.
The economically important event in a reusable launch system happens after the cameras cut away. What condition is the vehicle actually in? Can technicians walk its length, sign off a standard inspection card, replace a handful of parts from stock, refill the tanks, and fly it again — the way an airline turns a widebody in hours? Or does every return open a new campaign: thousands of tiles inspected one by one, gap fillers checked, engines borescoped, structures X-rayed, a vehicle that is legally and practically a prototype every time it lands?
The Space Shuttle answered that question the hard way. It was recoverable and reusable by any reasonable definition — 135 missions across five orbiters — and it never came close to routine. Its thermal protection system alone consumed on the order of 17,000 labor-hours per flight, and turnaround that was designed to take two weeks averaged months. The Shuttle proved that "reusable" and "cheap to reuse" are different engineering problems, and that the second one is harder.
SpaceX knows this history better than anyone — it is why the company builds Starship out of stainless steel instead of aluminum, attaches its tiles mechanically instead of with adhesive, manufactures them in volume instead of by hand, and spends flight after flight deliberately damaging its own heat shield to find the margins. The question is whether those choices are enough. Three former NASA thermal-protection specialists, examining the same Flight 13 imagery that prompted Elon Musk to declare the latest shield design "looks great," concluded the opposite: that ceramic tile systems of this architecture are a dead end for full-and-rapid reuse at orbital entry speeds, whatever their survivability.3,7
Both camps are looking at real data. This paper's job is to figure out what that data actually supports — and what it doesn't yet say. Research cutoff for all evidence: 16 August 2026, on the eve of Flight 14's planned first orbital flight and first tower catch of a ship.
"Reusable" is doing too much work in public discussion of Starship. A vehicle that flies twice after a nine-month refurbishment and a vehicle that flies twice in a week are both "reusable," and they have almost nothing in common economically. Before evaluating evidence, the levels need names.
The RC Reusability Ladder below is an operational scale, not a physical one. We are explicit about this because the airliner analogy is frequently abused: an aircraft never leaves the atmosphere, never sees a 1,400 °C shock layer, and dissipates its landing energy through wheels and brakes rather than through its skin. The environments are not comparable and nothing in this paper pretends they are. What can be compared is the operational pattern — how much scheduled and unscheduled maintenance stands between one revenue flight and the next.
| Level | Turnaround burden | Maintenance intensity | Plausible flights / vehicle / yr | Launch-economics implication |
|---|---|---|---|---|
| 0 — Expendable | None (new vehicle each flight) | Zero reuse maintenance; full manufacturing cost every flight | 1 total | Cost per flight ≈ vehicle cost. Cadence limited by factory. |
| 1 — Recoverable | Months; vehicle enters a refurbishment campaign | Structural repair, large-area TPS rework, engine removal | 1–2 | Recovery may cost more than it saves. Value is mainly learning. |
| 2 — Reusable | Weeks to months of scheduled servicing | Full TPS inspection, selective tile replacement, engine checks | 3–8 | Real savings if refurb ≪ build cost. Shuttle never cleared this bar economically. |
| 3 — Rapidly reusable | Days; inspection by exception | Automated/sampled TPS inspection; line-replaceable units | ~25–50 | Amortization collapses; ops and propellant become visible in the cost stack. |
| 4 — Aircraft-like | Hours to days; standardized cards | Condition-based maintenance; flight-critical hardware rarely opened | 100+ | Cost per flight approaches propellant + ops floor. Enables depot-scale logistics. |
Here is the first result of the analysis, and it cuts against both the boosters and the skeptics: Starship does not need Level 4 to be economically revolutionary — but SpaceX's own stated missions need it eventually. As Section 7 quantifies, the great majority of the cost-per-flight improvement arrives by roughly ten to twenty-five flights per vehicle with weeks-scale turnaround — solidly Level 2 to early Level 3. That is enough to make Starship the cheapest heavy-lift capacity in history on a modeled cost basis and to fly Starlink V3 at scale. What Level 2–3 cannot do is support a propellant-depot logistics chain that needs ten-plus tanker launches per lunar mission on a schedule, or a Mars architecture that concentrates hundreds of launches into synodic windows. Those missions are cadence-bound, not just cost-bound, and cadence is exactly what maintenance burden throttles. RC ESTIMATE
Keep the ladder in mind throughout. Every piece of flight evidence in Section 4 answers a specific question: not "did the vehicle survive?" but "which rung does this behavior belong to?"
Getting to orbit is an energy problem; getting back is the same problem run in reverse, with the vehicle's own skin as the brake. Every design decision in Starship's heat shield follows from a few unforgiving numbers.
A vehicle in low Earth orbit moves at about 7.8 km/s. Kinetic energy scales with the square of velocity — E = ½mv² — which is why orbital reentry and, say, a Falcon 9 booster's entry at ~2 km/s are different regimes entirely: the orbital vehicle carries roughly fifteen times the specific energy. For a Starship upper stage with an entry mass around 150 tonnes RC ESTIMATE, the arithmetic gives about 4.2 terajoules of kinetic energy that must go somewhere before splashdown. That is the energy of roughly one kiloton of TNT — a small nuclear weapon's worth — or about 1,170 megawatt-hours, a day's electricity for a city of forty thousand US homes, dissipated in under twenty minutes, most of it in a few minutes around peak heating.
The crucial subtlety is that almost none of this energy is meant to enter the vehicle. A blunt body flying at high angle of attack pushes a detached shock wave ahead of itself; air crossing that shock is compressed and heated abruptly — to temperatures in the shock layer far hotter than the vehicle ever sees, hot enough to dissociate molecules into reactive atoms and partially ionize the flow into the plasma visible on reentry video. The overwhelming majority of the heat stays in that processed air and is swept into the wake. The heat shield's job is to manage the small fraction — convective heating where hot boundary-layer gas scrubs the surface, plus radiative heating from the glowing shock layer itself — that leaks toward the wall. Blunt is good: a blunt shape pushes the shock farther off the body and thickens the insulating boundary layer. This is why Starship comes home belly-first at roughly 60–70 degrees angle of attack, presenting its broadest, bluntest face to the flow and using its four flaps as air brakes and attitude paddles rather than wings.
Two different quantities get conflated in coverage. Peak heat flux (and the associated peak surface temperature, around 1,400 °C on Starship's windward tiles FACT22 — a figure SpaceX has cited and CNN characterized as roughly half the temperature of the Sun's surface) determines what the tile material must withstand without melting or cracking. Total integrated heat load — flux summed over the whole entry — determines how much heat soaks through to the structure, and therefore how thick the insulation must be. Starship's trajectory choices trade these against each other: a steeper entry means higher peak flux for less time; a shallower, lifting entry means lower peak flux endured for longer. A vehicle designed for rapid reuse must survive not the average entry but the worst dispersed case, repeatedly, without accumulating damage.
Heating is brutally uneven. Stagnation regions — the nose, the windward centerline, and especially the leading edges of the flaps — see the highest flux because the boundary layer is thinnest and the flow most energetic there. The flap hinge lines are worse still: they are moving mechanical joints that must seal against plasma while articulating under load, a combination that burned visibly through Ship 29's forward flap on Flight 4 even as the vehicle survived. Any step, gap, protruding edge, or missing tile can trip the boundary layer from smooth laminar flow to turbulent flow, which multiplies local heating several-fold downstream — the mechanism that makes a single bad tile a neighborhood problem rather than a local one. FACT
Finally, scale. Starship's upper stage is about 52 meters long and 9 meters in diameter; the windward acreage runs to well over a thousand square meters. Protecting it takes roughly 18,000 tiles. The physics of any one square meter is Shuttle-era physics; what is new is the sheer area that must be tiled, inspected, and warranted flight after flight. The heat shield problem is a physics problem multiplied by a counting problem.
E = ½ × 150,000 kg × (7,500 m/s)² ≈ 4.2 × 10¹² J
≈ 1.0 kiloton TNT-equivalent · ≈ 1,170 MWh. A Falcon 9 booster entry manages roughly 1/15 of this per unit mass. Assumption: 150 t entry mass (V3 dry mass + residuals; SpaceX has not published an exact figure). RC ESTIMATE
Starship's thermal protection system is best understood as three deliberate bets against the Shuttle's cost drivers: standardized tiles instead of custom ones, mechanical attachment instead of glue, and a steel structure that forgives what an aluminum one would not.
The windward surfaces are covered by roughly 18,000 hexagonal ceramic tiles FACT — black-coated ceramic insulators — widely reported as silica-based and in the same broad material family as the Shuttle's tiles, though SpaceX has not published the exact composition — engineered for reusable operation up to roughly 1,400 °C. The hexagon is a considered choice: with near-uniform tiles, a hex field has no long straight gap lines running with the flow for hot gas to channel down, and a standard tile can be mass-produced (SpaceX built a dedicated tile factory in Florida) and swapped generically rather than fabricated to a station-specific drawing. The Shuttle's ~24,000 tiles were almost all unique parts; Starship's are mostly interchangeable commodity parts. That single difference attacks the largest line item in historical TPS cost. INFERENCE
Tiles mount mechanically — pressed onto pin/clip fittings on the tank wall — rather than being bonded with adhesive over a strain-isolation pad as on the Shuttle. Mechanical attachment allows fast removal and replacement and tolerates the thermal expansion mismatch between hot ceramic and cold steel, but it introduces its own failure modes: clips that release under vibration and aero-acoustic load (the visible cause of tile shedding on ascent in early flights), and defined gaps between tiles that must stay within tolerance so plasma cannot reach the attachment hardware. Beneath and between, SpaceX has flown layered backing insulation and — from Flight 6 onward, expanded substantially in the V2 and V3 campaigns — a sacrificial ablative underlayer, deliberately exposed on some flights by removing tiles, so that a lost tile chars a backup layer instead of torching the tank wall.2 FACT
The last line of defense is the structure itself. Starship's 300-series stainless steel retains useful strength to several hundred degrees Celsius and tolerates brief excursions far beyond what the Shuttle's 2024-aluminum airframe (structurally limited to roughly 175 °C) could survive. This is the design's central structural bet: the Shuttle's TPS had to be nearly perfect because the structure behind it was unforgiving; Starship's TPS is allowed to be imperfect because steel buys margin. Flight 4 proved the philosophy vividly — Ship 29 reached the ocean under control with a flap that had partially burned through. Survivable-with-damage, however, is a safety property, not a turnaround property: a charred ablative layer and a heat-soaked hinge are exactly the kinds of findings that convert a quick inspection into a repair campaign. INFERENCE
Uniform acreage tile is the easy 90 percent. The hard 10 percent is everywhere the geometry stops being a cylinder: the nose cap's compound curvature; the flap leading edges and tips; the flap hinge lines, where a dynamic seal must articulate inside plasma flow (redesigned and moved leeward in V2 precisely because of Flight 4's burn-through); penetrations and interfaces — vents, catch hardware, the forward flap roots; and the aft skirt around the engine bay, which takes both entry heating and landing-burn recirculation. These regions carry custom parts, tighter tolerances, and most of the observed damage. They are also, historically, where TPS labor concentrates. FACT
The shield flying on V3 today is not the shield that flew in 2024, and treating "Starship's heat shield" as one static object is the most common analytical error in coverage of the program. V1 ships flew a first-generation tile field that shed visibly on ascent and entry. The V2 campaign (Flights 7–11) added the robustified ablative underlayer, relocated and shrank the forward flaps to pull their hinges out of the worst flow, and ran deliberate tile-removal and alternate-material experiments — including metallic tile trials on Flight 10 FACT. V3 (Flights 12–13) introduced new tile geometry and revised attachment clips developed from V2 flight data, leeward-side tile patches, and instrumented load-sensing tiles flown through a deliberately harsher, higher-dynamic-pressure ascent on Flight 13. FACT The system is converging — but it is converging on a better tiled shield, not on a different architecture. Whether the tile architecture itself has a rapid-reuse ceiling is the live dispute this paper returns to in Sections 5 and 9.
The maintenance economics of a tiled shield are governed by the system's weakest element, not its average. A shield is flight-ready only when every tile, every clip, and every gap is within tolerance — which means inspection effort scales with tile count even in a flight where damage is minimal. If verifying one tile position takes one minute of combined human-plus-automated effort, 18,000 positions is 300 labor-hours per cycle before a single part is replaced. Driving that number down is an automation and instrumentation problem (load-sensing tiles and dedicated imaging satellites are exactly the right tools), but it is a problem that exists because of the architecture, independent of how good each tile becomes. INFERENCE
Thirteen integrated flights, seven controlled reentries and targeted splashdowns, one ship intact on the ocean afterward. Read in sequence, the record shows a program methodically buying down heat-shield risk — and it also shows exactly which claims the data cannot yet support.
Two framing notes before the timeline. First, every reentry to date has been flown on a suborbital trajectory just short of orbital velocity — deliberately, so that a dead vehicle falls into a cleared ocean corridor. Entry energies are close to, but not identical with, the Mach-25 orbital case that operational missions require; Flight 14 is planned as the first true orbital entry. FACT Second, all TPS condition data to date comes from telemetry and imagery — no flown TPS hardware had been confirmed recovered by the cutoff. Ship 40 is the first candidate article: it survived intact, was photographed at close range afloat, and became the subject of an unplanned tow toward port whose completion was unconfirmed when this paper closed.4,6
| Flight / date | Gen | Reentry objective / stressors | What happened (TPS-relevant) | What it established / what the next flight chased |
|---|---|---|---|---|
| F3 03/2024 | V1 | First reentry attempt | Vehicle lost roll control coasting; broke up during entry around Mach 12+. No controlled TPS data. | Established that attitude control, not tiles, was the first gate. F4 targeted a controlled entry. |
| F4 06/2024 | V1 | Survive peak heating; controlled splashdown | Survived entry despite progressive burn-through of a forward-flap hinge region on live camera; flap kept actuating; on-target soft splashdown. | Proved steel-plus-tile margin philosophy: major damage ≠ loss of vehicle. Exposed the hinge line as the critical weak zone → V2 flap relocation. Musk subsequently called the heat shield the hardest remaining Starship problem. |
| F5 10/2024 | V1 | Repeat entry with upgraded shield (reworked tile field, ablative backup) | First booster tower catch; ship achieved precise on-target splashdown. Entry imagery showed tile loss and flap damage reduced vs F4 but present. | Confirmed trajectory precision needed for a future ship catch. TPS still shedding → F6 turned to deliberate experiments. |
| F6 11/2024 | V1 | Deliberate TPS stress: tiles removed, alternate materials, daylight entry, higher-AoA dynamic maneuvering | Controlled entry and splashdown with intentionally missing tiles; data on bare-spot heating and secondary-layer performance. | Start of the "instrument the failure modes" doctrine. Set up V2's redesigned shield. |
| F7·F8 01·03/2025 | V2 | V2 debut entries planned | Both ships lost during/after ascent (propulsion-bay anomalies). No entry data. | Cost the program ~6 months of TPS learning; drove V2 propulsion fixes. |
| F9 05/2025 | V2 | First reflown booster; V2 entry data | Ship lost attitude control (leak) before entry interface; uncontrolled entry, breakup. First reflown booster (B14, previously flown on F7). | First booster reflight = first hard evidence of Super Heavy reusability. Ship TPS validation still pending → F10. |
| F10 08/2025 | V2 | Full experiment suite: tile removals, metallic-tile trial, ablative zones | Controlled entry and targeted splashdown; experimental zones returned differentiated data; visible damage localized. Landing-burn/engine-bay heating noted. | First clean V2 TPS dataset. Metallic-tile results fed material trade for V3. |
| F11 10/2025 | V2 | V2 finale, flown on the second reflown booster (B15, previously F8, with 24 flight-proven Raptors); dynamic banking to simulate return-to-launch-site approach | Controlled entry, aggressive bank executed, on-target splashdown; shield performed with limited visible loss. | De-risked the steering profile a tower catch requires. Closed V2 book → V3 redesign. |
| F12 05/2026 | V3 | V3 shield debut: new tile geometry + clips, leeward experimental tiles, one tile deliberately removed, Mach-7 flap stress, bank maneuver, imaging satellites | Shield and flaps intact through peak heating despite the missing-tile test; minimal visible damage in flip/landing imagery — "a stark difference" from V2 per independent observers.19,20 Booster lost pre-splashdown (boostback relight/flip anomaly). Ship destroyed at splashdown as planned. | First evidence the V3 redesign works as designed under a deliberate defect. Booster relight problem enters the record. F13 escalates: higher dynamic pressure + instrumented tiles. |
| F13 07/2026 | V3 | Higher-thrust ascent (max-q stress on tile attachment), load-sensing tiles, white-painted "missing tile" imaging targets, leeward patches, in-space Raptor relight, deliberately more energetic entry; 6 Starlink V3 satellites imaging the shield | Ship survived the over-stressed entry, flipped, landed on target, and remained intact and afloat — a program first — enabling close-range inspection imagery of a flown shield. SpaceX subsequently attempted an unplanned tow to port; completion was unconfirmed at the cutoff.2,4 Visible tile loss low ("a few pieces" per SpaceX); Musk: "Latest heat shield design looks great… all the heat shield data we needed and then some." Independent review of the same imagery (Rasky/Camarda/Miller): cracked tiles, broken edges, white streaks at tile boundaries consistent with hot-gas gap penetration. Booster landing burn again under-relit (~8 engines vs ~13 planned) → hard splashdown. | The pivotal dataset: the first close-range physical-condition evidence, first instrumented load data, and the first public expert dispute over what "looks great" means. Verified entry systems for an orbital attempt → F14 planned as first orbital flight + first ship catch (tiered environmental assessment for the new landing trajectory, and the associated license authorization, pending at cutoff9). |
Three trends are unambiguous. First, controlled reentry has gone from aspiration to routine: every ship that reached entry interface with attitude control intact since June 2024 has flown its profile to a targeted splashdown — seven for seven.1 FACT Second, visible damage per flight is trending down across generations, from F4's live burn-through to F13's "a few pieces" of tile loss on a deliberately over-stressed profile. FACT Third, SpaceX's test doctrine has inverted: early flights tried to protect the shield; recent flights try to break it — removing tiles, over-pressuring ascent, over-energizing entry — while instrumenting the result. A program confident its shield barely works does not fly missing-tile experiments; a program hunting margin data does. INFERENCE
What the record does not yet contain is just as important. No ship has entered from true orbital velocity. No ship has flown twice. No tile field has been through two heating cycles. There is no public measurement of the metric this paper cares most about — technician-hours per turnaround — because there has never been a turnaround. And the one shield observed intact after flight — Ship 40's — has produced a genuine expert split: SpaceX reads the same article as validation, three career TPS specialists read it as evidence of gap heating that "cannot get to Mach 25 full-and-rapid reusability." If Ship 40 reaches port, its teardown findings would be the single most valuable dataset in reusable launch; even the close-range imagery already collected is the best physical-condition evidence the program has produced. INFERENCE
The Space Shuttle is the only orbital vehicle ever operated with a reusable tiled heat shield, which makes it the only empirical benchmark for what such a shield costs to own. The point of the comparison is not "Shuttle had tiles, therefore Starship fails." It is narrower and more useful: the Shuttle record tells us exactly which numbers to demand from Starship before believing any turnaround claim.
The program's founding promise was aircraft-like — and it needs stating precisely, because two Shuttle numbers with different units are routinely, and invalidly, divided into each other. The turnaround requirement was an elapsed-time figure: 160 working hours from landing to launch — roughly two weeks of ground processing — an explicit early design driver in NASA Kennedy's engineering literature.11,24 FACT The labor record is a separate, person-hours figure: tile replacement and maintenance consumed on the order of 17,000 person-hours per mission per NASA data reported in 1999, across an orbiter carrying over 20,000 unique silica tiles (~24,300 on early Columbia) requiring pre- and post-mission inspection and re-waterproofing.12 FACT Clock time and aggregate labor are different dimensions; no ratio between 17,000 person-hours and 160 elapsed hours is meaningful, and this paper draws no conclusion from one. What can be said, dimensionally correctly: actual turnaround averaged months against the two-week elapsed goal (best case 54 days, Atlantis, 1985; early flows fell from 187 to 60 days by STS-7; fleet average roughly 4–5 flights per year across 135 missions in 30 years),13,15 and the TPS labor bill alone would have filled the 160-hour window many times over even with large crews working in parallel. Later-program tiles cost $2,000–$3,000 apiece to fabricate. FACT The TPS was only part of a broader burden — SSME removal and inspection, hypergolic systems, and payload processing all consumed the schedule — but the tile system was among the most labor-dense items and, after Columbia, the most safety-critical inspection on the vehicle.
Starship's design departs from the Shuttle at nearly every point that generated Shuttle TPS cost, and the comparison must be scored mechanism by mechanism rather than by silhouette.
| Cost/failure mechanism | Shuttle (as operated) | Starship (as flying, V3) | Does the Shuttle lesson transfer? |
|---|---|---|---|
| Tile uniqueness | ~24k tiles, nearly all unique parts; days to fabricate and fit each | ~18k tiles, predominantly standardized hexagons, factory mass production | Largely fixed by design. Replacement cost per tile should be far lower. INF |
| Attachment | Adhesive bond over strain-isolation pad on aluminum; densification campaigns; bond verification per tile17 | Mechanical pins/clips on steel; fast swap; but clip retention under vibration is a live failure mode SpaceX is still instrumenting (F13 load-sensing tiles) | Different failure mode, same category of concern: attachment integrity must be verified across the whole field every flight until proven otherwise. INF |
| Structure behind the shield | Aluminum, ~175 °C limit — near-zero tolerance for TPS defects; Columbia loss traced to TPS breach14 | Stainless steel + ablative underlayer — demonstrated survival with missing tiles and a burned flap | Genuinely different. Starship's defect tolerance is real. But tolerance protects the vehicle, not the schedule: charred backup layers still force rework. FACT/INF |
| Inspection burden | Manual, per-tile, ~17,000 h/flight class; waterproofing reapplication | Unknown — never performed. SpaceX is building toward automated/sensor-based inspection (load-sensing tiles, imaging satellites) | The open question. Inspection scales with tile count and gap count, which Starship retains at Shuttle scale. This is the number to watch. INF |
| Waterproofing | Silica tiles absorb water; waterproofing re-injection required every flight | Not publicly established. Starship's tiles are reported to share broad ceramic/silica heritage, but whether they carry a comparable between-flight water-management burden is unknown; Ship 40's prolonged seawater immersion makes post-splashdown moisture handling a live operations question | Open question — not an established transfer. Tower catches would largely bypass ocean exposure, one more reason the catch is economically, not just theatrically, important. INF |
| Engines | SSMEs removed and inspected after each flight; staged-combustion hydrogen engines at performance limit | Raptor designed for reuse without removal; booster engine reflight demonstrated (F9, F11); but relight reliability is an open anomaly class (F12–F13) | Partially transfers. See Section 6 — engines may inherit the "hidden bottleneck" role tiles played for Shuttle. FACT |
| Operational philosophy | Crewed every flight; safety-driven inspection depth; standing army ~thousands | Uncrewed test cadence; iterate-by-flying; damage acceptable if instructive; heavy automation intent | Does not transfer — deliberately. But note: HLS and any crewed Starship re-imports Shuttle-grade assurance requirements. INF |
| Scale & manufacturing | 5 orbiters in 30 years; artisanal production | Serial production line (Starfactory/Gigabay); vehicles are semi-disposable during test; fleet-level learning | New territory. Shuttle had no analog to "just build another ship while you study the last one." FACT |
Strip away the differences and one Shuttle lesson survives contact with Starship's design intact: a tiled shield makes vehicle readiness a per-tile certification problem, and the labor lives in the certification, not the parts. Shuttle tiles rarely failed; verifying that they wouldn't is what consumed the person-hours. Starship's standardization and steel margins attack the cost of acting on a finding; they do not by themselves eliminate the cost of looking. Whether automated inspection — instrumented tiles, machine vision, satellite imaging on ascent — can collapse "looking" from thousands of hours to tens is the specific engineering question on which aircraft-like turnaround hinges. It is, notably, a question SpaceX is visibly working: F13's instrumentation suite is exactly what a program serious about automating certification would fly. The Rasky/Camarda/Miller position is that no amount of inspection efficiency rescues an architecture whose gaps admit hot gas at Mach 25; SpaceX's position, implicitly, is that V3's geometry and clips have closed the gaps to within tolerance. A recovered-hardware teardown — Ship 40's, if the tow succeeds, or the first caught ship's — will arbitrate. INFERENCE
The heat shield is the most visible obstacle to rapid reuse because it fails photogenically. It is not guaranteed to be the binding one. Aircraft-like turnaround requires every subsystem to be certified-by-default; the schedule is set by the slowest system on the vehicle — or on the ground.
The map below scores every major candidate bottleneck on two axes: how much flight evidence exists, and how consequential the open questions are for turnaround time specifically. Three findings stand out.
1 — Engine relight is the most repeated open anomaly in the current campaign. Super Heavy's boostback/landing burns under-relit on both V3 flights: Flight 12's boostback ignition went wrong enough to lose the booster, and Flight 13's landing burn lit roughly 8 engines against a ~13-engine plan, producing a hard splashdown.1,23 FACT Catch operations have zero tolerance for this failure class — a booster or ship that cannot guarantee its landing-burn ignition cannot be caught safely over infrastructure. Raptor 3's ascent record is now strong (33/33 on F13 after a pad abort and two engine swaps21); its restart record in flight conditions is the program's most conspicuous unresolved reliability problem. Software and sequencing fixes are reportedly in work for Flight 14.3,6 FACT
2 — Nobody has ever inspected a flown Raptor to a reuse standard in public. Booster 14's engines reflew (F9), which bounds the answer encouragingly, but combustion-chamber life, turbomachinery wear, and seal cycling under full-flow staged combustion at Raptor 3 chamber pressures are exactly the kind of quiet, unglamorous life-limit questions that set inspection intervals — and inspection intervals set turnaround. Methane is a cleaner fuel than kerosene (no coking) and Raptor 3's simplification deletes hardware that used to need checking; both genuinely help. INFERENCE
3 — The ground may gate before the vehicle does. A rapid-reuse Starship needs a pad that turns as fast as the ship: catch, destack, restack, and a multi-thousand-tonne propellant load on a days-not-weeks cycle, times the number of towers available, inside whatever launch and reentry licenses the FAA has issued. At cutoff, Starbase operates two pads and Florida infrastructure is in build-out. The regulatory record needs precision: the FAA had finalized the environmental review for increased Starbase launch cadence (Final Tiered EA with a Mitigated FONSI/Record of Decision), while a further tiered environmental assessment covering the new ship-landing trajectory — released in draft on 13 July 2026, comment period closed 3 August 2026 — and the specific license authorization for orbital flight with a ship catch remained pending.9 FACT Regulatory cadence has historically moved slower than hardware cadence in this program. INFERENCE
The honest summary of this section: if the tiles work, the tempo question moves to engines and the ground. A plausible 2028 failure mode for "rapid" reuse is not a burned-through ship — it is a fleet of intact ships queuing behind engine inspections, a single available tower, and a licensing cycle, with the heat shield reduced to one line on a long maintenance card. Shuttle veterans will recognize the pattern; the vehicle was rarely waiting on tiles alone. SPECULATION
SpaceX does not publish its costs, and this paper will not pretend to know them. What can be done honestly is to build a transparent illustrative model — every assumption stated, every equation shown, every number labeled — and ask a structural question that survives uncertainty in the inputs: where do the returns to reuse actually live?
All values are RC assumptions, not SpaceX data. RC ESTIMATE The model prices each flight as the sum of a capital charge, refurbishment, operations, and propellant. The capital charge is where naive models double-count: if vehicle cost is divided by planned life and an expected-loss term is added on top, the same lost vehicle is paid for twice — and an expendable case that already charges the full vehicle every flight cannot coherently add an expected-loss premium for hardware that is being expended by design. We avoid both errors by spreading each vehicle's build cost over its expected number of flights, which random loss truncates:
E[flights] = [ 1 − (1 − p)N ] / p
capital / flight = Cship / E[flights]ship + Cbooster / E[flights]booster
where p = per-flight hardware-loss probability and N = maximum planned service life absent loss. Ship and booster are computed separately with their own N. The expression counts a vehicle's flights until it is lost or retired, whichever comes first, so a loss simply shortens the amortization base — the loss is priced exactly once. Limits check: N = 1 gives E[flights] = 1 (expendable pays full vehicle cost, no loss premium added); as N → ∞, E[flights] → 1/p (life becomes loss-limited). Excluded: payload/mission-failure costs, schedule externalities, insurance — all of which make losses more expensive than modeled. RC ESTIMATE
Ship build cost $60M and booster $100M at BASE (LOW $36M/$60M, HIGH $84M/$140M — an RC bracket around public estimates for the stack of roughly $90–250M); propellant $1.0/1.5/3.0M per flight; refurbishment and operations decline with maturity by scenario (register in Appendix A); per-flight loss probability declines from 5% (early recovery attempts) to 1% (mature rapid reuse). LOW applies all favorable assumptions jointly; HIGH all adverse — the brackets are analyst-selected and should be read as scenario envelopes, not confidence intervals. Payload: 100 t to LEO in reusable configuration (a SpaceX target, not a demonstrated figure; 130 t assumed expendable).
| Scenario | Planned life ship / booster (flights) | Loss p / flight (BASE) | E[flights] ship / booster (BASE) | BASE components: capital + refurb + ops + prop $M | LOW $M | BASE $M | HIGH $M | Modeled cost per available kg (BASE)† | Ladder level |
|---|---|---|---|---|---|---|---|---|---|
| A — Expendable-equivalent | 1 / 1 | 2% | 1.0 / 1.0 | 160.0 + 0 + 6.0 + 1.5 | 101 | ≈168 | 236 | ~1,290* | 0 |
| B — Recovery + heavy refurb | 3 / 5 | 5% | 2.9 / 4.5 | 43.1 + 30.0 + 10.0 + 1.5 | 48 | ≈85 | 143 | ~845 | 1 |
| C — 5 flights per ship | 5 / 10 | 4% | 4.6 / 8.4 | 24.9 + 12.0 + 8.0 + 1.5 | 27 | ≈46 | 78 | ~465 | 2 |
| D — 10 flights per ship | 10 / 20 | 3% | 8.8 / 15.2 | 13.4 + 8.0 + 6.0 + 1.5 | 17 | ≈29 | 51 | ~290 | 2 |
| E — 25 flights per ship | 25 / 40 | 2% | 19.8 / 27.7 | 6.6 + 4.0 + 4.0 + 1.5 | 9.4 | ≈16 | 29 | ~160 | 3 |
| F — 50 flights per ship | 50 / 75 | 1.5% | 35.4 / 45.2 | 3.9 + 2.5 + 3.0 + 1.5 | 6.4 | ≈11 | 20 | ~110 | 3–4 |
| G — 100+ flights, rapid reuse | 100 / 150 | 1% | 63.4 / 77.9 | 2.2 + 1.5 + 2.0 + 1.5 | 4.3 | ≈7.2 | 14 | ~72 | 4 |
* Scenario A assumes 130 t expendable payload; all others 100 t. † "Modeled cost per available kg at full payload utilization" = modeled flight cost ÷ assumed maximum payload. It is not a market price and not a prediction of what customers pay: actual missions may be underfilled, volume-constrained, orbit-constrained, or integration-constrained, and pricing is set by markets, not cost floors. Reference point only: Falcon 9 list price ≈ $70M for ~17 t reusable (~$4,100/kg full-utilization equivalent). Every LOW/BASE/HIGH cell is reproducible from the equations in §7.1 plus the complete scenario-by-scenario register in Appendix A, Table A2. RC ESTIMATE
Read the LOW, BASE and HIGH columns top to bottom. At BASE, moving from expendable to ten flights per ship cuts direct cost per flight by a factor of ~5.8; moving from ten flights to a hundred — ten times more reuse capability, and by far the harder engineering — buys only another factor of ~4.0. The same shape holds at LOW (6.0× then 3.9×) and at HIGH (4.6× then 3.7×): under every input bracket tested, more than half of the total log-scale cost improvement arrives by roughly ten flights per vehicle. The thesis survives the brackets. By Scenario E the vehicle itself has nearly vanished from the cost stack — capital is $6.6M of a $16M flight — and the bill is dominated by refurbishment and operations: labor, which is to say the heat shield's inspection burden and the engine's inspection interval. Past roughly ten reuses, the heat shield stops being a survival question and becomes the principal line item in the price of a flight. RC ESTIMATE
The retired comparison of Shuttle person-hours against an elapsed-time goal (see Section 5) is replaced here by a model that keeps the two dimensions separate. First, person-hours: an illustrative bottom-up estimate of TPS servicing labor per flight. RC ESTIMATE
| Driver | LOW | BASE | HIGH |
|---|---|---|---|
| Automated inspection fraction | 95% | 70% | 30% |
| Human verification per manually checked position | 2 min | 4 min | 6 min |
| Tiles requiring intervention | 0.2% (36) | 1% (180) | 5% (900) |
| Labor per tile replacement | 0.5 h | 1.5 h | 4 h |
| Gap / clip rework (positions × h) | 20 × 0.5 | 150 × 1 | 1,000 × 2 |
| Ablative-underlayer repair (zones × h) | 0 × 8 | 3 × 16 | 20 × 40 |
| Engineering review & disposition | 200 h | 800 h | 3,000 h |
| Total TPS person-hours / flight | ≈260 h | ≈1,630 h | ≈10,700 h |
| Direct TPS labor cost / flight | ≈$0.04M | ≈$0.24M | ≈$1.6M |
Two readings. First, the HIGH case lands in the same order of magnitude as the Shuttle's ~17,000-hour record — that is the world the independent TPS experts describe, where a tiled orbital shield remains an inspection-bound system. The LOW case is the world SpaceX's instrumentation program (load-sensing tiles, imaging satellites, machine vision) is trying to build. The order-of-magnitude gap between them is the entire economic argument of this paper. Second, note that even the HIGH labor cost (~$1.6M) is smaller than the scenario refurbishment lines in Table 4A — because refurbishment also carries parts, engine servicing, facilities, and non-TPS work; TPS labor is the largest single uncertainty within it, not the whole line. RC ESTIMATE
Elapsed turnaround is a separate quantity and depends on concurrency, not just the labor total: technicians per shift, parallel work fronts (windward acreage vs flap hinges vs engine bay), physical access and scaffolding, task sequencing, shift structure, and serial bottlenecks such as cure and dry times. As a theoretical labor floor, elapsed working days ≈ total person-hours ÷ aggregate productive person-hours per day. With one shift of 20 technicians at 8 productive hours each (160 person-hours/day), the BASE 1,630 h is a ~10.2-working-day floor; with two such shifts (320 person-hours/day), ~5.1 days. The HIGH 10,700 h with two 50-technician shifts (800 person-hours/day) is a ~13.4-day floor. These are floors, not forecasts: real turnaround can be substantially longer because the work is not perfectly parallelizable — serial inspection sequences, physical access, engineering disposition, cure/dry times, tooling and parts dependencies all add elapsed time that no headcount removes, so technician count alone does not determine turnaround. This paper deliberately does not claim a specific Starship turnaround requirement; no such requirement has been modeled here, and none should be quoted from this paper. RC ESTIMATE
Vehicle economics are only half the stack. Towers, pads, propellant farms and processing facilities are fixed costs that must be spread over annual launches — and none of these figures are public, so the layer below is entirely illustrative. RC ESTIMATE
| Parameter | LOW | BASE | HIGH |
|---|---|---|---|
| Tower/pad + propellant farm + processing capital | $1.0B | $2.0B | $4.0B |
| Capital life | 20 yr | 15 yr | 12 yr |
| Annual fixed ground workforce & operations | $80M | $200M | $400M |
| Annualized fixed cost | $130M/yr | $333M/yr | $733M/yr |
| Fixed cost per launch @ 5 / yr | $26.0M | $66.7M | $146.7M |
| @ 10 / yr | $13.0M | $33.3M | $73.3M |
| @ 25 / yr | $5.2M | $13.3M | $29.3M |
| @ 50 / yr | $2.6M | $6.7M | $14.7M |
| @ 100 / yr | $1.3M | $3.3M | $7.3M |
The structure matters more than the numbers: fixed cost per launch falls as 1/cadence, so at low annual rates the ground allocation can exceed the entire vehicle-side cost of a Scenario-E flight (BASE $13.3M at 25/yr vs $16M direct), while at 100+/yr it fades to single digits. This is the quantitative reason cadence, not just vehicle reuse, carries the economics — and why an under-utilized rapid-reuse system can cost more per flight than a busy moderate-reuse one. RC ESTIMATE
The original question this paper's brief posed: at what engine-service burden does propulsion, not the heat shield, pace reuse? The illustrative sensitivity below prices scheduled inspection, scheduled removal/overhaul, and unscheduled replacement across the stack's 39 engines (33 booster + 6 ship). No Raptor unit cost or service interval below is a company figure. RC ESTIMATE
| Parameter | LOW | BASE | HIGH |
|---|---|---|---|
| Scheduled inspection: cost / interval | $10k / 10 flts | $50k / 5 flts | $200k / 2 flts |
| Scheduled removal & overhaul: cost / interval | $0.3M / 30 flts | $0.8M / 15 flts | $2.0M / 5 flts |
| Unscheduled replacement: rate × unit cost | 0.2% × $1M | 1% × $2M | 3% × $4M |
| Per-engine service burden / flight | $13k | $83k | $620k |
| Stack propulsion burden / flight (×39) | ≈$0.5M | ≈$3.3M | ≈$24M |
The crossover: propulsion overtakes the Scenario-E BASE TPS/refurb line ($4M) once the per-engine burden exceeds roughly $103k per flight — a threshold that sits between our BASE and HIGH engine assumptions. In plain terms: if Raptors need Shuttle-SSME-style attention (removal, borescoping, frequent overhaul), engines become the binding reuse bottleneck even if the tiles cooperate perfectly; if they achieve the on-condition, no-removal servicing SpaceX designs for, propulsion stays a secondary line. The F13 pre-launch pad abort and two engine swaps, and the F12/F13 landing-burn relight shortfalls, are the current public data points on the unscheduled side of this ledger. RC ESTIMATE
The distinction Section 9's milestone M7 tracks — flight cadence vs vehicle reuse cadence — can be quantified with one relationship. To fly M missions per year with vehicle turnaround T days and a reserve factor r (spares, maintenance float): fleet ≈ ⌈M × (T/365) × r⌉; sustaining it also requires ~M/N new builds per year, where N is vehicle life. RC ESTIMATE
| Annual missions | T = 5 d | T = 15 d | T = 30 d | T = 60 d | T = 120 d | New builds / yr (life 10 / 25 flights) |
|---|---|---|---|---|---|---|
| 12 | 1 | 1 | 2 | 3 | 6 | 1.2 / 0.5 |
| 25 | 1 | 2 | 4 | 7 | 13 | 2.5 / 1.0 |
| 50 | 2 | 4 | 7 | 13 | 25 | 5.0 / 2.0 |
| 100 | 3 | 7 | 13 | 25 | 50 | 10 / 4.0 |
Read across the 100-mission row: SpaceX can support a hundred flights a year with three ships turning in five days (Level 3–4 operations) or with fifty ships turning in four months (Level 1–2 operations plus a very large factory). Both produce the same launch cadence; only the first produces rapid-reuse economics. This is why observed flight rate, by itself, proves nothing about reuse maturity — and why M7 on the milestone watchlist asks where the cadence comes from, not how high it is. Starfactory-scale production makes the right-hand strategy genuinely available to SpaceX, which no previous program could say. RC ESTIMATE
"Can Starship fly like an aircraft?" is really eight questions, because Starship is eight business lines wearing one airframe. Each mission cares about a different mix of cost per flight, cadence, and reliability — and they land on very different rungs of the ladder.
| Use case | Required cadence | Required cost point | Sensitivity to reuse level | Needs aircraft-like reuse? | Minimum viable ladder level |
|---|---|---|---|---|---|
| Starlink V3 deployment | ~Monthly–weekly per pad | Even $30–50M/flt beats Falcon per Gbps delivered | Moderate — internal customer absorbs early costs; cadence matters more than $/kg | No | 2 → 3 |
| Conventional commercial payloads | Low (market is tens of launches/yr) | Must beat Falcon 9/Heavy per mission, not per kg | Low — modeled Level-2 cost per flight is already mission-competitive (market pricing is a separate question) | No | 2 |
| Third-party megaconstellations | Weekly-class during build-out | <$500/kg attractive; <$200/kg market-making | Moderate–high | Helpful, not required | 3 |
| Tanker flights / orbital depots | High — clustered: ~10–20 launches per depot campaign inside boil-off windows | Propellant-in-orbit price set almost entirely by $/flight | Extreme — the defining rapid-reuse customer | Effectively yes | 3 → 4 |
| Lunar cargo (uncrewed) | Campaign bursts | Cost dominated by tanker chain above | High (inherited from tankers) | Via tankers | 3 |
| HLS / Artemis crew | Low frequency, schedule-critical | Fixed-price contract; cost secondary to certification | Moderate — needs the tanker chain to close, plus crew-grade assurance (which raises inspection depth, cutting against speed) | Via tankers | 2 vehicle + 3 tankers |
| Private human spaceflight | Low | Premium market, price-insensitive early | Low on cost; extreme on demonstrated reliability (flight count is the product) | No — but wants the safety record high cadence builds | 2–3 |
| Mars logistics | Extreme — synodic surges: fleets launching in weeks-long windows every 26 months, each Mars ship fed by multiple tankers | Musk-scale ambitions assume ~$/kg in the tens | Maximal — the mission that motivated the architecture | Yes | 4 |
| Point-to-point Earth transport | Airline-daily | Airline-adjacent $/seat | Total — is the airline analogy, literally, plus regulatory/siting problems beyond this paper | Yes, by definition | 4+ |
The pattern is consistent. In RC's assessment, the revenue-bearing missions of the next several years close at Level 2–3 — Starlink V3 (whose spacecraft had begun flight testing aboard Starship as of Flight 13's suborbital deployment2), the commercial manifest, even single-mission lunar cargo. RC ESTIMATE Among the major space-logistics use cases considered here, depot-scale tanker campaigns and Mars logistics are the clearest cases where Level-4-like operations become structurally valuable — the constraint is not the price of one flight but the ability to launch again, from the same pad, within days, tens of times in a row. This resolves the apparent contradiction between Musk's optimism and the TPS experts' pessimism: they are talking about different rungs. A shield that is a "dead end" for Mach-25 daily operations can simultaneously be a perfectly good engine of the Starlink business — which is, notably, exactly the dual reading the Rasky/Camarda/Miller analysis itself offers. INFERENCE
The strategic corollary: Starship can be the most disruptive launch vehicle in history and still miss its founding goal. Point-to-point Earth transport would require aircraft-like operations by definition, but remains a more speculative business case than any of the space-logistics missions above. SPECULATION
The honest answer to the title question is not yes or no. It is a probability distribution over rungs of a ladder, with the mass moving upward but concentrated well below the top.
| Claim | RC probability | Confidence in estimate | Basis |
|---|---|---|---|
| Technically reusable — a Starship upper stage reflies at least once by end-2027 | ~85% | Moderate–high | Intact return demonstrated; catch attempt imminent; strong organizational track record converting recovery capability into reflight (Falcon 9 boosters; Super Heavy F9 and F11). |
| Routinely reusable (Level 2) — ships averaging ≥3 flights each with weeks-scale turnaround, by end-2028 | ~70% | Moderate | Requires no breakthrough beyond scaling what F13 showed; main risks are relight reliability and orbital-entry surprises. |
| Rapidly reusable (Level 3) — median ship turnaround under ~10 days sustained across a campaign, by end-2029 | ~30% | Low–moderate | Requires the unmeasured variable (TPS inspection labor) to land 10× below Shuttle-class and engines to reach inspection-by-interval. Possible; undemonstrated. |
| Minimal TPS refurbishment — routine reflight with only sampled/automated tile inspection and <1% tile replacement, by end-2029 | ~20% | Low | The specific claim the independent TPS experts dispute; zero cycles of multi-flight tile data exist either way. |
| Aircraft-like cadence (Level 4) — 100+ flights/vehicle-year operations anywhere in the fleet, by end-2030 | ~10% | Low | Requires everything above plus pad throughput, licensing cadence, and possibly a next-generation TPS approach (active cooling / modular panels) beyond current tiles. |
Upward: a clean orbital-velocity entry on F14; a successful ship catch with post-catch imagery showing a field-serviceable shield; Ship 40 teardown findings (if the recovery succeeds and findings surface) showing boundary streaks are cosmetic; a ship reflight inside 60 days; disclosure of tile-touch labor an order of magnitude below Shuttle-class. Downward: orbital entry revealing a new damage regime at full energy; a catch scrubbed or failed on relight; V4-era shield redesigns that add rather than remove touch labor; recovered-hardware findings validating systemic gap heating; a pivot in SpaceX messaging from turnaround time to fleet size (building rate substituting for reuse rate — the quiet tell that turnaround economics aren't closing).
Starship is converging on recoverable now, on reusable plausibly within two years, and the evidence for anything beyond that is not yet in existence — not negative, but absent, because the experiments that would generate it (orbital entry, catch, reflight, second heating cycle) begin with Flight 14. The heat shield has graduated from "can it survive?" — where the V3 answer looks increasingly like yes — to "what does it cost to certify, every time?", where the only honest current answer is that nobody outside Starbase knows. The airliner question was always two questions: Starship the launch business needs Level 2–3 and appears on course for it; Starship the Mars program needs Level 4, an operational regime no tiled orbital vehicle has ever approached. The next decisive Starship metric is not whether the tile field survives. It is how much inspection and repair the surviving vehicle requires before the same hardware can fly again — and public evidence through 16 August 2026 still does not answer that question.
Every RC-generated number in this paper derives from the assumptions below. None are SpaceX figures. Replication: E[flights] = [1 − (1 − p)N] / p per vehicle; direct cost per flight = ship cost ÷ E[flights]ship + booster cost ÷ E[flights]booster + refurb + ops + propellant. Hardware loss enters only through E[flights] — it is never charged a second time as a separate expected-loss line, and the expendable case (N = 1, E[flights] = 1) carries no loss premium on hardware already expended by design. Infrastructure per launch = (capital ÷ capital life + annual fixed ops) ÷ annual launches. Fleet ≈ ⌈missions/yr × turnaround days ÷ 365 × reserve⌉.
| Parameter | LOW | BASE | HIGH | Basis / note |
|---|---|---|---|---|
| Ship marginal build cost | $36M | $60M | $84M | RC bracket around public estimates; serial-production assumption. Not a SpaceX figure. |
| Booster marginal build cost | $60M | $100M | $140M | Ditto; 33 engines dominate. |
| Propellant + consumables / flight | $1M | $1.5M | $3M | ~4,000–5,000 t methalox at industrial prices + boil-off, He/N2, logistics. |
| Operations / flight | 0.75× scenario BASE | scenario-dependent ($2–10M) | 1.5× scenario BASE | Range, recovery, mission ops, pad campaign labor; declines with maturity by construction. LOW/HIGH scale each scenario's BASE value — absolute values per scenario in Table A2. |
| Refurbishment / flight | 0.5× scenario BASE | scenario-dependent ($1.5–30M) | 2× scenario BASE | Dominated by TPS touch labor + engine inspection; the paper's key uncertain variable. LOW/HIGH scale each scenario's BASE value — absolute values per scenario in Table A2. |
| Loss probability p / flight | 0.5×p | 1–5% by scenario | 2×p (capped at 10%) | Enters capital via E[flights] only; ignores schedule/insurance/payload externalities (understates true cost of failure). |
| Payload to LEO, reusable | 80 t | 100 t | 130 t | SpaceX V3-class target; not yet demonstrated — flagged as company target, not fact. |
| Entry mass (energy calc) | 130 t | 150 t | 170 t | Dry mass + residuals; SpaceX has not published an exact figure. |
| TPS servicing person-hours / flight (§7.3) | ≈260 h | ≈1,630 h | ≈10,700 h | Bottom-up drivers in Table 4B; HIGH lands in the same order of magnitude as Shuttle actuals (~17,000 person-hours, NASA 1999). The retired 160-hour figure is an elapsed-time requirement and is never used as a labor benchmark. |
| Ground infrastructure (§7.4) | $130M/yr | $333M/yr | $733M/yr | Annualized capital + fixed workforce per site complex; allocated per launch by cadence. RC ESTIMATE only. |
| Engine service burden / flight, 39 engines (§7.5) | ≈$0.5M | ≈$3.3M | ≈$24M | Scheduled inspection + overhaul + unscheduled replacement (Table 4D). No Raptor cost or interval is a company figure. |
| Fleet reserve factor r (§7.6) | 1.5 | Spares and maintenance float over the pure utilization requirement. | ||
| Loaded technician rate | $150 / hour | US aerospace fully-loaded labor, RC assumption. | ||
Known limitations: no discounting or cost of capital; no development-cost amortization (marginal-cost basis); loss modeled as expected value only; ops/refurb correlations with cadence are asserted, not derived; payload figure is a company target. The model is designed so its structural conclusion — early saturation of reuse returns, dominance of refurb at moderate reuse — survives the input brackets: the LOW / BASE / HIGH scenario columns in Table 4A were each recomputed end-to-end (A→D cost cut 6.0× / 5.8× / 4.6×; D→G cut 3.9× / 4.0× / 3.7×). Bracket bounds are analyst-selected envelopes, not statistical confidence intervals.
Table A2 gives the complete per-scenario inputs behind every Table 4A cell. Planned vehicle lives are held fixed across LOW/BASE/HIGH (the brackets vary costs and loss probability, not engineering life); LOW applies all favorable values jointly and HIGH all adverse values jointly.
| Scenario | Ship / booster life N (all cases) | p: LOW / BASE / HIGH | Ship + booster build $M: LOW / BASE / HIGH | Refurb $M: LOW / BASE / HIGH | Ops $M: LOW / BASE / HIGH | Propellant $M: LOW / BASE / HIGH |
|---|---|---|---|---|---|---|
| A | 1 / 1 | 1% / 2% / 4% | 36+60 / 60+100 / 84+140 | 0 / 0 / 0 | 4.5 / 6.0 / 9.0 | 1.0 / 1.5 / 3.0 |
| B | 3 / 5 | 2.5% / 5% / 10%* | 15 / 30 / 60 | 7.5 / 10 / 15 | ||
| C | 5 / 10 | 2% / 4% / 8% | 6 / 12 / 24 | 6.0 / 8.0 / 12 | ||
| D | 10 / 20 | 1.5% / 3% / 6% | 4 / 8 / 16 | 4.5 / 6.0 / 9.0 | ||
| E | 25 / 40 | 1% / 2% / 4% | 2 / 4 / 8 | 3.0 / 4.0 / 6.0 | ||
| F | 50 / 75 | 0.75% / 1.5% / 3% | 1.25 / 2.5 / 5 | 2.25 / 3.0 / 4.5 | ||
| G | 100 / 150 | 0.5% / 1% / 2% | 0.75 / 1.5 / 3 | 1.5 / 2.0 / 3.0 |
* HIGH p = 2× BASE, capped at 10% (binds in Scenario B only). Worked check, Scenario B LOW: E[flights]ship = [1−(1−0.025)³]/0.025 = 2.926; E[flights]booster = [1−(1−0.025)⁵]/0.025 = 4.756; capital = 36/2.926 + 60/4.756 = 24.9; + refurb 15 + ops 7.5 + propellant 1.0 = $48.4M, as printed in Table 4A. Scenario B HIGH: p = 10%, E = 2.710 / 4.095; capital = 84/2.710 + 140/4.095 = 65.2; + 60 + 15 + 3 = $143.2M. Saturation ratios recomputed from this register: A→D cost cut 6.0× / 5.8× / 4.6× (LOW/BASE/HIGH); D→G 3.9× / 4.0× / 3.7×.
| Figure | Title | Type | Location |
|---|---|---|---|
| FIG 01 | RC Reusability Ladder with positional markers | RC original diagram | Section 1 |
| FIG 02 | Reentry heat-flow anatomy — Starship belly-first entry | RC original diagram | Section 2 |
| FIG 03 | TPS cross-section — RC reconstruction from publicly available information (exact V3 stack not publicly confirmed) | RC original diagram | Section 3 |
| FIG 04 | Flight-by-flight heat-shield timeline, IFT-3 → IFT-14 (incl. reused-booster marks F9/F11) | RC original timeline | Section 4 |
| FIG 05 | Post-entry TPS condition by flight (damage-class schematic, not a damage map) | RC original diagram | Section 4 |
| FIG 06 | Shuttle vs Starship TPS mechanism comparison | Analytical table (Table 3) | Section 5 |
| FIG 07 | Starship rapid-reuse bottleneck map | RC original chart | Section 6 |
| FIG 08 | Cost-per-flight sensitivity tornado, Scenario E (recomputed under corrected capital model) | RC original chart (model data) | Section 7 |
| FIG 09 | Use-case / reuse-requirement matrix | Analytical table (Table 5) | Section 8 |
| FIG 10 | "What would prove aircraft-like reuse?" milestone watchlist | RC original graphic | Section 9 |
All figures are RC originals built from cited sources and the Appendix A assumption register. No AI-generated imagery is presented as flight evidence anywhere in this publication.
Claim-level citations. Numbered markers in the text point here; source tiers: primary (SpaceX, FAA, NASA/NTRS) are preferred throughout, secondary reporting carries only observations unavailable from a primary source.
Research cutoff 16 Aug 2026. Sourcing changes in this revision: Wikipedia, TechTimes, Starlust and NextBigFuture removed from conclusion-critical sourcing; SpaceX mission pages substituted as the primary flight record; FAA and NASA/NTRS documents substituted for regulatory and Shuttle-era claims; Gizmodo retained only where it carries a specific reported observation (endnote 20). Where secondary reporting is cited for facts, RC verified consistency across at least two outlets. Company statements (payload targets, cadence goals, cost aspirations) are treated as targets throughout, never as established facts.