Additive Manufacturing Qualification

Established Qualification is an evidence problem, not a printing problem. A fracture-critical part — one whose failure loses the aircraft, the crew or the patient — is certified on a package of evidence: design allowables derived from statistically controlled material data, a frozen manufacturing process, nondestructive evaluation with known detection limits, and a paper trail from raw material to installation. For forged and machined metal that machinery took decades to build and now works routinely. The question this brief examines is whether the same machinery can be rebuilt for additive manufacturing (AM), where the printer makes the material and the geometry in the same operation.

Established Additive manufacturing breaks the classical assumption that material properties and part geometry can be certified separately. In laser powder bed fusion, the dominant metal AM process, a focused laser melts successive ~30–60 µm layers of powder; every location in the part receives its own thermal history, set by laser parameters, scan path, local geometry, position on the build plate, powder condition and machine state. The part is not cut from a billet whose pedigree was established elsewhere; the part is the material sample, and there is only one of it.

Frontier The field’s answer is to qualify the process rather than the part: lock every parameter, monitor every build with in-situ sensors, validate process models against benchmark measurements, track powder and data provenance end to end, and argue statistically that a part built inside the locked window inherits the properties of the coupons that qualified it. Each element of that programme exists in partial form. None is finished, and the joints between them — especially the statistical joint between sensor data and fracture-critical acceptance — are where the open problems live.

Frontier The decisive question is therefore narrow and testable: can fracture-critical printed parts be certified repeatably — by evidence rules that transfer across parts, machines and sites — rather than heroically, one part number at a time? The record below says: not yet, anywhere, and the distance remaining is measurable in coupons, dollars and probability-of-detection curves.

Established Start with what the classical system demands, because it prices everything that follows. The MMPDS handbook (Metallic Materials Properties Development and Standardization, successor to MIL-HDBK-5, maintained by Battelle under FAA sponsorship) is the reference for aerospace design allowables. An A-basis value is one that 99% of material meets with 95% confidence; B-basis is 90% with 95% confidence. Deriving them requires data from on the order of ten independent production lots and roughly a hundred specimens per property, per condition, per orientation — and for conventional alloys a “lot” is a heat of material, a well-understood object.

Established For AM the lot concept explodes combinatorially. Properties measurably depend on machine make and model, parameter set, powder lot and reuse state, build layout and location on the plate, build orientation, thermal post-processing and surface condition. A qualification basis must either fix all of these — producing allowables valid only for one alloy on one machine configuration at one site — or sample across them at a cost that multiplies accordingly. Frontier Programmes that have actually produced AM design allowables report coupon counts in the thousands and multi-year timelines; commonly cited full costs for one alloy-machine-process combination run from the low millions to above ten million US dollars. Those figures circulate in industry presentations rather than audited budgets, so this brief carries them as a range, not a number. Frontier Around 2022, MMPDS incorporated its first additively manufactured design allowables — reported as a laser powder bed fusion titanium alloy, developed through an America Makes-funded programme at Battelle. That a sixty-year-old handbook admitted its first AM entry only in this decade, while AM parts had already been flying for years, is the statistical basis problem stated as an institutional fact. The announcement is cited from the bibliographic record, and the alloy identification is flagged accordingly.

Established The flight-qualified record is real, growing and narrower than its reputation. The FAA cleared its first additively manufactured commercial engine part in 2015: GE’s T25 compressor-inlet temperature sensor housing, retrofitted to several hundred in-service GE90 engines (vendor’s own figures). The LEAP engine’s printed fuel nozzle tip — roughly twenty conventional parts consolidated into one, about 25% lighter, with a durability improvement the maker states as fivefold (vendor) — entered service in 2016; GE announced its 100,000th printed nozzle tip in 2021. The GE9X, certified in 2020, carries on the order of three hundred printed parts across a handful of component families, including titanium-aluminide low-pressure turbine blades made by electron beam melting. In rockets, SpaceX flew a printed main oxidizer valve body on a Falcon 9 in January 2014 (company’s own account) and the SuperDraco escape engine with its laser-sintered Inconel chamber has flown on every crewed Dragon since 2020. Relativity Space’s Terran 1 — roughly 85% printed by mass — survived maximum dynamic pressure in March 2023 before a stage-two failure, a structural demonstration even in a failed orbital attempt.

Frontier The RS-25 and RL10 stories show what qualification costs when the part matters most. Aerojet Rocketdyne’s RS-25 restart programme printed the pogo accumulator assembly — a large structural component of the engine’s oscillation-damping system — and began hot-firing it at Stennis in late 2017; the new-production engine, carrying dozens of printed parts against a roughly 30% unit-cost reduction target, completed its certification test series in 2024 (programme’s own announcements) but, to this brief’s knowledge, had not flown as of early 2026. On the RL10, printed components have been hot-fired extensively in the RL10C-X development line, including a full additively built copper-alloy thrust chamber, and the manufacturer has stated printed-chamber variants will enter service on Vulcan-era upper stages; this brief could not verify a flight date for a printed-chamber RL10 and says so rather than asserting one. Frontier Military aviation’s firsts are similarly specific: NAVAIR flew a flight-critical printed titanium link-and-fitting assembly on an MV-22 nacelle in 2016; a printed titanium replacement part was installed on an F-22 in 2019; and the F-35’s printed inventory, though numbering many part numbers, sits overwhelmingly in non-structural and sustainment roles. No publicly documented fracture-critical printed structural part was flying on F-35 as of this brief’s knowledge cutoff — a thinner record than the programme’s reputation suggests.

Established The quiet success is medical. An electron-beam-melted titanium acetabular cup received its CE mark in 2007; the FDA cleared a printed polymer cranial implant in 2013; by the time the FDA issued its guidance Technical Considerations for Additive Manufactured Medical Devices in December 2017, the agency itself counted more than a hundred AM devices on the US market. Printed porous-titanium spinal cages and hip cups are now mainstream, with cumulative implant counts in the hundreds of thousands at minimum and industry claims running to the millions (vendor figures; no independent audit located). Established The reason medical worked is structural, not accidental: implants are certified device by device, mostly through predicate-comparison pathways with device-level bench testing, in compression-dominated load cases where AM’s porous lattices are the clinical advantage. Medicine never needed MMPDS-style allowables, so it never hit the statistical wall aerospace is still climbing.

Established The standards state. NASA published NASA-STD-6030, Additive Manufacturing Requirements for Spaceflight Systems, in 2021, generalising the Marshall standards MSFC-STD-3716/3717 of 2017: consequence-based part classification, a Qualified Material Process locked to a specific machine configuration, statistical process control against a Process Control Reference Distribution built from witness specimens, per-build witness testing, and first-article inspection including computed tomography, with equipment and facility control split into companion standard NASA-STD-6033. It is the most complete public statement of “qualify the process” anywhere. Established Neither the FAA nor EASA has an AM-specific rule; both certify printed parts through the existing airworthiness framework, project by project, steered by policy documents such as EASA’s Certification Memorandum CM-S-008 and by joint FAA-EASA AM workshops. Consensus standards fill in below: ASTM F42 and ISO/TC 261 (the ISO/ASTM 52900 series, including 52904 for metal powder bed fusion in critical applications and 52920 on qualification of AM production sites), and SAE’s AMS7000-series process and powder specifications. Frontier The result is a workable but bespoke system: every certification so far is a point solution — one part, one process window, one machine configuration, one negotiation with one regulator.

Frontier In-situ monitoring’s measured record is far more modest than its marketing. The review literature (Grasso and Colosimo 2017; Sanaei and Fatemi 2021, among others) catalogues melt-pool photodiodes and cameras, layerwise imaging, thermography and acoustic emission, with published defect-detection accuracies typically between 70% and 95% — measured on laboratory datasets with seeded or deliberately induced defects, usually on one machine, with detection thresholds tuned after the fact. Frontier Studies that compare commercial melt-pool monitoring flags against computed tomography ground truth report weak correlation at the individual-pore scale: gross lack-of-fusion is detectable, but the sub-100 µm porosity that initiates high-cycle fatigue failure largely is not, and false-positive rates that look small per voxel condemn whole builds at part scale. Established Accordingly, no regulator anywhere grants certification credit for in-situ monitoring in place of post-build nondestructive evaluation on fracture-critical parts; NASA-STD-6030 treats it as supplementary information. That gap — between sensor coverage of every voxel and statistical trust in none of them — is the exact frontier of this subject.

Frontier Question one: can in-situ monitoring earn a probability-of-detection curve? Conventional NDE methods are admitted to fracture-critical service through POD studies — blind trials establishing, say, 90% detection of a critical flaw size at 95% confidence. No published study has put an in-situ monitoring system through the equivalent blind, multi-machine protocol. Until one does, monitoring data is engineering insight, not certification evidence, and the entire “self-certifying build” vision rests on an unmeasured quantity.

Frontier Question two: do allowables transfer? Whether coupon data from one machine bounds the properties of a nominally identical machine next to it — let alone a different model, site or powder vendor — is genuinely contested. Round-robin exercises have repeatedly found measurable inter-machine variation under nominally fixed parameters. An accepted statistical equivalence rule for transferring a qualification basis would collapse the cost of the whole enterprise; none exists.

Frontier Question three: can models substitute for coupons? Model-assisted qualification — process simulation plus Bayesian updating to shrink coupon counts — is the stated ambition of NIST’s qualification programme and much of the ICME community. The AM Bench benchmark series (2018 and 2022, with a further round in preparation) exists precisely to test whether simulations predict melt-pool geometry, residual stress and microstructure blind. Models have scored well on some challenge problems and poorly on others; nobody has yet carried a model-assisted argument through a fracture-critical certification.

Frontier Question four: is there a fatigue path for as-built surfaces? Machining every surface surrenders much of AM’s geometric advantage, but as-built roughness can cut high-cycle fatigue strength by half or more relative to machined, hot-isostatically-pressed material, with life scatter spanning an order of magnitude. Whether surface-state-specific allowables, chemical polishing or design rules can make an as-built fracture-critical surface certifiable is open. Frontier Question five: where is the powder reuse endpoint? Oxygen pickup, spatter contamination and size-distribution drift degrade reused titanium powder toward specification limits, yet economics demand reuse; a defensible, sensor-verifiable end-of-life rule for powder is still argued vendor by vendor. Speculative Behind all five sits the data question: whether the terabyte-scale build archives now accumulating can be pooled across competitors into shared statistical bases, or whether every manufacturer re-derives the same curves in private, forever.

Established The first bottleneck is arithmetic: coupons cost more than the parts. A statistically defensible basis for one alloy on one locked process consumes thousands of specimens, months of machine time and millions of dollars, and it buys exactly one point in a configuration space with a dozen axes. Any subsequent change — a firmware update, a new powder vendor, a moved machine, a resized part — reopens some or all of the argument. Frontier Change control is therefore the operational bottleneck: qualification freezes processes at the very moment machine vendors iterate hardware and software annually, and the industry has no agreed taxonomy of which changes are minor, which require bridging data and which restart everything.

Established Second: nondestructive evaluation strains against AM’s geometry. Computed tomography resolution degrades with part size and wall thickness — tens of micrometres of voxel at best for real parts — while the fatigue-initiating defect population extends below 50 µm; rough as-built surfaces mask surface-breaking indications from both CT and penetrant methods; and the internal channels that justify AM in the first place are exactly the regions hardest to inspect. Frontier Third: the in-situ data pipeline has no accepted decision rule. A monitored build can generate from tens of gigabytes to terabytes of imagery and photodiode traces, but there is no standardised mapping from signal anomaly to defect probability to accept/reject decision — every vendor’s flag means something different, and none of them means anything to a regulator yet.

Established Fourth: provenance is fragmented. The digital thread that qualification presumes — powder heat and reuse genealogy, build file hash, machine log, sensor archive, post-processing records, inspection results, all linked to a serial number — exists today as paper certificates and incompatible vendor formats. Published sabotage demonstrations (a 2017 academic team printed a covertly weakened drone propeller that failed in flight) established that an unsecured build chain is an attack surface, not just an audit inconvenience. Frontier Fifth: people. The certifying workforce — designated engineering representatives, NDE Level III examiners, materials review board engineers who understand both fracture mechanics and scan strategies — is scarce enough that several programmes name reviewer availability, not data generation, as their schedule driver. Speculative If AM adoption accelerates while that workforce stays flat, the queue at the certification office becomes the binding constraint regardless of what the sensors learn to see.

Established This subject imports its hardest problems from four older disciplines. From fracture mechanics and fatigue: defect-population-based life prediction — treating the pore ensemble measured by CT as the initial crack distribution and propagating it to a life estimate — is the intellectual core of any argument that a monitored build is safe, and the review literature (Sanaei and Fatemi 2021) rates it promising but unvalidated for blind prediction. Frontier From process physics and ICME: melt-pool, residual-stress and microstructure models must predict, not postdict; the AM Bench series is the field’s standing referee for that claim, and its challenge-problem results are the honest scorecard — mixed.

Established From NDE physics: probability-of-detection methodology, developed over decades for cracks under aircraft skins, is the template any in-situ monitoring credit must satisfy; the transfer of POD statistics to voxel-scale sensor streams is an open statistical problem, not a software feature. Frontier From statistics proper: small-data allowables, tolerance bounds under multi-level variance (machine, build, location, day), and equivalence testing for process transfer are the mathematical machinery under every shortcut the field wants; the methods exist in the literature, but their regulatory acceptance for AM is piecemeal. Established Beneath everything sits metrology — calibrated melt-pool radiometry, traceable powder characterisation, reference defect artefacts — which is precisely the layer NIST’s AM qualification programme and its Additive Manufacturing Metrology Testbed exist to supply. A field that wants to certify from sensors needs the sensors themselves certified first.

Frontier The experiment this brief ranks first has never been run in full. The decisive test is a blind, multi-site round robin: nominally identical fracture-critical builds on at least five machines, with in-situ monitoring verdicts locked before inspection, followed by computed tomography and fatigue testing to failure on every part, published as probability-of-detection curves against defect size. Seed a known fraction of builds with process excursions; let each site’s monitoring stack commit its accept/reject calls and defect maps in escrow; then compare against CT and against where fatigue cracks actually started. One such campaign would convert in-situ monitoring from marketing to measurement, in either direction — and would give regulators the first number they can act on. Frontier No agency or consortium has funded the full blind version; NIST’s AM Bench series runs the adjacent model-calibration half — blind prediction of melt pools and microstructure — but not the probability-of-detection half on fracture-critical hardware.

Frontier Second: the transfer experiment. One locked parameter set, one powder specification, ten nominally identical machines across at least three sites, full mechanical characterisation at each — published with machine identities anonymised but distinguishable. This is the direct measurement of the quantity every qualification economics argument assumes: the between-machine variance component. Fragments exist in round-robin literature; the full designed version, sized to support an equivalence rule, does not.

Frontier Third: the defect-to-life atlas. Build specimens spanning the realistic defect population, CT every one, fatigue-test every one to failure, and require models to predict life distributions from CT data alone, blind, before the tests. This is the validation the “treat pores as initial cracks” framework needs before a monitored defect map can ever substitute for a witness coupon. Frontier Fourth: powder end-of-life. Carry a single titanium powder lot through reuse cycles to failure of specification, with interleaved mechanical testing and full chemistry at each cycle, to replace vendor-by-vendor reuse folklore with a sensor-verifiable retirement rule. Established Every one of these experiments is buildable with existing hardware; what they lack is a funder whose incentive is public curves rather than private advantage, which is why this brief’s institutional section is not decoration.

Established A production cell that could pass tomorrow’s audit is specifiable today, and NASA-STD-6030 is close to its bill of materials. Configuration-controlled machines under an equipment standard (NASA-STD-6033’s role): calibrated laser power, spot size and gas flow, with drift monitored against schedule. A qualified material process locked in writing, with witness specimens on every build plate feeding a statistical process control chart — the Process Control Reference Distribution pattern — so that each build proves membership in the qualified population rather than merely resembling it. First-article inspection with computed tomography; serialised parts; archived build files, machine logs and sensor records tied to each serial number.

Established Post-processing is not optional finishing; it is half the property set. Hot isostatic pressing to close internal porosity, stress relief and heat treatment to set microstructure, machining or surface conditioning wherever fatigue lives — each a controlled process with its own qualification burden, and each a reminder that “printed” parts are really printed-plus-four-operations parts. Frontier The engineering gap is integration, not invention: sensor archives that an auditor can query years later, powder genealogy that survives vendor changes, monitoring flags wired into dispositioning workflows with recorded thresholds. Firms demonstrably operate cells like this for their own locked part numbers; what does not exist is the interoperable version — formats, thresholds and records portable between companies and reviewable by a regulator without a bespoke translation project. Frontier First-part-correct — certifying serial number one from process evidence alone, with no destructive sibling — is the stack’s stated destination, and every element above must hold simultaneously for it: that is why it remains a destination rather than a practice.

Established Every neighbouring manufacturing frontier runs into this brief’s problem wearing different clothes. In orbit, NASA’s In-Space Manufacturing project and ISS printer operations exist to make spare parts and reduce logistics for spaceflight (Prater and colleagues’ 2017 and 2019 programme updates) — manufacturing where the witness-coupon, teardown-and-test apparatus of terrestrial qualification simply cannot follow, which is why Space-Based Manufacturing keeps that business carefully separate from its own. Frontier A NASA-convened research campaign on the sciences of space manufacturing (2021, with industry co-authors) sketches the research agenda; the qualification chapter of that agenda is this brief’s subject transplanted to a place with no test lab. Established The structural version of the same story ran through Archinaut: a 37-metre-plus printed beam on the ground in 2017, a $73.7 million NASA contract in 2019, a 7.0-metre flight-like beam in 2020, and project conclusion in 2023 without an orbital print — while DARPA’s NOM4D programme (agency’s own release) manifested its Phase 3 in-space structure demonstrations for 2026 flights toward an ambition of 100-metre-class apertures. Orbital Shipyards owns that assembly story; the certification of what gets assembled will be owned by whoever solves the problem described here.

Established On the ground, Automated Construction Systems documents the civil-engineering twin: printed concrete where a print stoppage is a structural event rather than a scheduling inconvenience (Kazemian and colleagues), reinforcement named first among the field’s open challenges (Shahzad and colleagues), and a decade of demonstrations with no code-grade characteristic values — qualify-the-process failing for want of a round robin, in a different material system. Frontier At the far end, Moon-Based Manufacturing carries the comparative evaluation of regolith solidification (Bao and colleagues, 2024), where laser powder bed fusion of regolith spans 4.2–31.4 MPa on simulant and the top-ranked construction technique is bagging the dust rather than printing it — a reminder that where qualification is impossible, engineering retreats to what needs least of it. The nearer adjacencies are quieter: powder metallurgy and welding metallurgy supply the defect taxonomy; hot isostatic pressing comes from the castings world; computed tomography from medical imaging; and probability-of-detection statistics from decades of aircraft NDE — additive qualification is less a new field than a forced merger of five old ones.

Established The central institutional fact: allowables are a public good being produced privately. Coupon data is expensive, and the firms that generate it treat it as competitive advantage; the result is that the same curves are re-derived, in private, dozens of times, while the public handbook gained its first AM entry only around 2022. The MMPDS model — government-sponsored, industry-fed, publicly usable — solved exactly this problem for wrought metal in the twentieth century; America Makes and the ASTM AM Center of Excellence are attempting the AM re-run with consortium funding, against member incentives that pull the other way. Frontier Whether pooled, precompetitive AM property databases reach critical mass this decade is a live institutional bet, and the honest current answer is: partially, alloy by alloy, slowly.

Established The regulatory landscape is a spectrum of the same idea. NASA wrote the strictest public version (NASA-STD-6030’s process-lock-plus-witness regime); the FAA and EASA run case-by-case airworthiness findings steered by memoranda and workshops rather than AM-specific rules; the FDA published device-level guidance in 2017 and lets predicate comparison carry most of the load; naval and defence authorities qualify part families for specific platforms. Frontier Standards production, meanwhile, is fragmented across ASTM F42/ISO TC 261, SAE AMS-AM and national bodies — dozens of documents, overlapping scopes, and no single document a design office can follow from powder to certificate. Established NIST’s qualification programme occupies the layer none of these bodies can: reference measurements, benchmark data and the metrological plumbing that lets one lab’s melt-pool number mean something in another’s. Speculative The institution nobody has built is the clearinghouse the decisive experiment needs — a body with the standing to run blind trials on vendors’ monitoring stacks and publish the resulting detection curves under rules the vendors cannot veto; existing consortia are member-funded and structurally unable to embarrass their members.

Established Qualification is where safety is actually decided, which makes its statistics an ethical subject. The people inside an aircraft or under an implant bear the tail risk of every shortcut in the allowables argument; a 90/95 basis is a promise about the worst material, not the average coupon, and pressure to cheapen qualification is pressure applied directly to that tail. Frontier The distributed-manufacturing future the technology invites — print the spare at the point of need — redistributes accountability along with production: when a file, a powder lot and a machine in three jurisdictions jointly produce a failed part, existing liability doctrine has no settled answer for who certified what. Established The sabotage question is demonstrated, not hypothetical: the 2017 printed-propeller attack study showed a compromised build file producing a part that passes casual inspection and fails in service, which makes build-chain security a safety obligation, not an IT preference. Frontier There is also a bench-worker dimension — fine metal powders are inhalation and combustion hazards handled daily by technicians — and a fairness dimension: if qualification stays artisanal and expensive, only incumbents can afford to certify, and a technology sold as democratising manufacturing consolidates it instead. Speculative Against these sits the ethical upside case: parts consolidated from twenty pieces to one remove twenty inspection seams and their failure modes, and a genuinely sensor-verified process could someday make manufacturing more honest than the paper-certificate world it replaces. That case is coherent; it is simply not yet earned.

Frontier Qualification is the governor on how fast manufacturing can decentralise. The printer is portable; the evidence package is not. Whether a machine in a forward depot, a shipyard or a Mars habitat can make a part anyone is allowed to trust depends entirely on whether certification can travel as data — process records, sensor archives, provenance chains — rather than as an institution with a test lab. Every supply-chain-resilience argument for AM runs through that gate. Established The precedent is instructive: interchangeable parts, standardised threads and material handbooks were civilisational infrastructure precisely because they let strangers trust each other’s components; MMPDS-class allowables for AM are the same project for a process where every machine is slightly its own foundry. Speculative If first-part-correct certification is achieved — serial number one trusted from process evidence alone — manufacturing’s economics shift from tooling-amortised production runs toward certified batches of one, with consequences for inventory, repair culture and small-polity industry. Speculative Off Earth the stakes sharpen: settlements that cannot ship witness coupons home must either carry the whole qualification apparatus with them or accept lower assurance, and which of those happens will be decided by the statistics developed — or not developed — in this decade’s terrestrial programmes. Handwave The strongest civilizational claims — fully autonomous factories certifying their own output with no human review — assume the closure of every gap this brief documents, and are asserted in roadmaps rather than argued from any demonstrated detection or transfer statistic.

These horizons track evidence and institutions — detection curves, transfer rules, pooled databases — not printer hardware, which already outruns them all.

• 10 yr: Frontier The first blind probability-of-detection results for in-situ monitoring are published; regulators grant the first partial inspection-reduction credits on locked processes; MMPDS-class AM entries grow from one to a handful of alloy-process pairs; medical and non-fracture-critical aerospace AM continue compounding quietly.
• 25 yr: Speculative Model-assisted qualification with pooled databases cuts coupon counts by an order of magnitude for mature alloy-process combinations; a statistical equivalence rule for machine-to-machine transfer is accepted by at least one major regulator; fracture-critical printed parts become routine within qualified facilities, though not portable between them.
• 50 yr: Speculative Certification travels with the part as a data object — process record, sensor archive, provenance chain — accepted across sites for most part classes, with first-part-correct practice established for all but the highest-consequence hardware.
• 100 / 250+ yr: Handwave Fully autonomous qualification — machines certifying their own output beyond human audit, on and off Earth — is asserted in long-range roadmaps; nothing in the current statistical record grounds a date, and the honest entry here is that no one knows.

• Depends on nothing on this map: the printers, sensors and computers exist, and the blockers are statistical bases and institutional machinery that no other brief on this map is producing. Its intellectual inputs — fracture mechanics, NDE statistics, process metrology — are mature terrestrial disciplines rather than frontier results.
• Requires (not on this map) blind probability-of-detection curves for in-situ monitoring, which no one has funded; fatigue-life models validated by blind prediction from measured defect populations; a regulator-accepted statistical rule for transferring allowables between machines and sites; a pooled precompetitive coupon database whose data rights let competitors actually use it; powder genealogy that travels machine-readably across the vendor chain; and fleet-scale repeatability data measuring the between-machine variance every economic argument assumes.
• Enables trusted distributed and point-of-need production generally: certified spares for remote and military logistics, printed primary structure in launch vehicles and engines at production rather than demonstration cadence, and any off-Earth industry whose parts must be trusted without a terrestrial test lab behind them.
• Adjacent Space-Based Manufacturing and Orbital Shipyards inherit this problem wherever their products must be certified; Automated Construction Systems is the same institutional failure in concrete, blocked at a standards committee rather than a laboratory; Moon-Based Manufacturing shows engineering retreating to low-qualification techniques where the apparatus cannot follow.

Established “GE prints engine parts, so AM qualification is solved.” Each certified part — the T25 housing, the LEAP nozzle tip, the GE9X family — is a point solution: one geometry, one locked process, one machine configuration, one negotiated evidence package. Nothing about the nozzle’s certificate transfers to the next part number; that is precisely the unsolved problem, as the vendors themselves say in standards committees.

Frontier “In-situ monitoring watches every voxel, so printed parts will certify themselves.” Watching is not detecting. Published detection rates of 70–95% come from seeded-defect laboratory studies with post-hoc thresholds, correlation with CT at the individual-pore scale is weak, and no blind probability-of-detection study of any monitoring system exists. No regulator grants monitoring any inspection credit on fracture-critical parts today, and until the blind numbers exist that is the correct posture.

Established “Printed metal is as good as forged.” Sometimes better in static strength, routinely worse in the properties that kill: high-cycle fatigue driven by porosity and as-built surfaces, anisotropy with build direction, and property scatter that widens every design margin. Post-processing — HIP, heat treatment, machining — closes much of the gap at the cost of being four more processes to qualify. The accurate sentence is: printed metal is as good as its process control, which is the whole subject.

Handwave “Blockchain / the digital twin will solve provenance and certification.” A ledger secures records; it cannot make the records mean anything. The missing quantities are detection probabilities, variance components and validated models — measurement problems, not bookkeeping problems — and no digital-thread architecture supplies them by construction. Frontier The defensible core is real but narrower: tamper-evident build chains are worth having, because the sabotage demonstrations are published and the paper-certificate status quo is worse.

Established “Regulators are blocking additive manufacturing.” The record shows the opposite: FAA-certified engine parts since 2015, an FDA guidance and hundreds of cleared devices, a NASA standard that tells you exactly what evidence to bring. The gate is open; what it demands is evidence that costs millions per alloy-process pair, and the queue at the gate is an economics problem the regulators did not create.

Frontier “Qualification travels with the file — print a certified part anywhere.” Under every current framework, qualification attaches to the file plus a specific machine configuration, site, powder chain and post-processing line; ISO/ASTM’s site-qualification standard exists precisely because the site is part of the material. The certified-anywhere future is this field’s destination, not its present, and the transfer statistics that would license it are unmeasured. Frontier “Medicine proves aerospace is just slow.” Medicine certified devices, not materials: predicate pathways, device-level bench tests, compression-dominated implants where porosity is the feature. Aerospace certifies tension-and-fatigue-critical structure against statistical allowables. The quiet medical success is genuine — and genuinely non-transferable.