The Ratio That Predicts Failure: How Much Pilot Evidence Is Enough Before a First Commercial Plant
THE SHORT ANSWER Pilot evidence is enough when the first commercial plant sits within 30 times the largest integrated unit that has run successfully, the longest run outlasted the process's dominant failure mechanism, and mass and carbon balances close within ±5%. Of the three tests, the scale-up ratio predicts failure best. |
SECTION 01
Which Pilot Would You Fund?
Two technologies reach the same investment committee in the same quarter. Both turn waste into a drop-in fuel, and both want capital for a first commercial plant of 500 tonnes a day.
Pilot A reports a mass balance that closes at 96%. Its best continuous run lasted 100 hours, on a 1 tonne-a-day unit. The commercial plant would be 500 times larger than anything it has operated.
Pilot B closes at 94%. It has logged a 2,000-hour continuous run. It reached its current 100 tonne-a-day demonstration unit through two earlier steps of 5x, and the commercial plant is one more 5x step.
PILOT A AGAINST PILOT B
Measure | Pilot A | Pilot B |
|---|---|---|
Mass balance closure | 96% | 94% |
Longest continuous run | 100 hours | 2,000 hours |
Largest unit operated | 1 tonne a day | 100 tonnes a day |
Path to 500 tonnes a day | One 500x jump | A final 5x step, after two earlier 5x steps |
Scale-up ratio | 500x | 5x |
How I read it | Unproven | Generally acceptable |
Most committees I have sat with lean toward A. The yield looks better and the deck looks cleaner. B is the safer investment by a wide margin. A has never shown how its process behaves anywhere near the scale it is asking you to fund, and nothing in its pilot data can tell you.
That gap, between what a pilot proves and what a commercial plant demands, is where first-of-a-kind plants fail, whatever the chemistry. This piece sets out the three tests I apply across thermochemical, catalytic, biochemical and electrochemical technologies, ranks them by how well they predict failure, and gives the evidence bar for each.
SECTION 02
Why Pilot Hours Mislead Investment Committees
First-of-a-kind failures are expensive and public, and they are not confined to one technology. Each case below comes from a different corner of the energy transition. Read through the tests, each shows a gap that could be seen before the money went in.
THE CASE FILE AT A GLANCE
Case | Technology | What broke | Test that flags it |
|---|---|---|---|
KiOR, Columbus | Biomass-to-fuel | Yield near 22 gallons per bone-dry ton against 72 cited | Balance closure |
Range Fuels, Soperton | Wood gasification | Feed system could supply one of two gasifiers | Scale-up ratio, solids handling |
Fulcrum Sierra, Reno | Waste gasification and Fischer-Tropsch | Nitric acid corrosion; deposits over 10 feet thick | Run length on the worst feed |
Gevo, Luverne | Isobutanol fermentation | Bacterial contamination | Consecutive clean batches |
Northvolt, Skellefteå | Battery cell manufacturing | Rejected cells; output far below plan | Throughput ratio |
Shell, Bintulu to Pearl | Gas-to-liquids | Nothing: a step under 10x after 18 years at the smaller plant | Passes |
Thermochemical Conversion: KiOR's Yield Claim
COLUMBUS, MISSISSIPPI · BIOMASS-TO-FUEL · $230 MILLION PLANT · IDLED MARCH 2014
KiOR built a $230 million biomass-to-fuel plant in Columbus, Mississippi, backed by a $75 million loan from the state. The plant was designed to take 500 tons of wood a day. KiOR cited a yield of 72 gallons per bone-dry ton. Internal calculations reported by Biofuels Digest put the plant near 22. The plant was idled in March 2014, and KiOR filed for Chapter 11 that November.
WHAT THE TEST WOULD HAVE FLAGGED Closure. A yield figure is only as good as the mass balance behind it, repeated across runs, at a scale close to the one being financed. |
Solids Handling: Range Fuels' Two Gasifiers
SOPERTON, GEORGIA · WOOD GASIFICATION · SHUT 2011
Range Fuels received a $76 million Department of Energy grant, an $80 million USDA loan guarantee and more than $150 million in venture capital for a wood-to-fuel plant in Soperton, Georgia. According to MIT Technology Review, the feed system it bought could supply only one of the plant's two gasifiers. The plant made some methanol in 2010, at a loss, and was shut in 2011.
WHAT THE TEST WOULD HAVE FLAGGED The scale ratio, applied to solids handling. The first diligence question is the largest unit that feed system had fed continuously, on that wood, at that rate. |
Waste Gasification: Fulcrum Sierra's First Integrated Test
RENO, NEVADA · WASTE GASIFICATION AND FISCHER-TROPSCH · CHAPTER 11, SEPTEMBER 2024
Fulcrum BioEnergy raised more than $1 billion and spent over $200 million on its Sierra plant near Reno, Nevada, designed to turn about 175,000 tons of prepared landfill waste a year into some 11 million gallons of synthetic crude by gasification and Fischer-Tropsch synthesis. Production began in 2022. As reported by Waste Today, nitric acid formed in operation and corroded equipment, and after a restart in late 2023 a cement-like buildup, in places more than 10 feet thick, formed in the gasification system and had to be cleared with explosives. Fulcrum laid off most of its staff in May 2024 and filed for Chapter 11 that September.
WHAT THE TEST WOULD HAVE FLAGGED Run length on the worst feed. Corrosion chemistry and ash deposition show themselves over thousands of hours on real waste, and short campaigns on prepared feed will not reveal them. |
Industrial Fermentation: Gevo's Contamination at Luverne
LUVERNE, MINNESOTA · ISOBUTANOL FERMENTATION · STOPPED 2012, RESUMED JUNE 2013
Gevo converted its ethanol plant in Luverne, Minnesota, to produce isobutanol and began production in May 2012. It stopped about five months later. In the words of its chief executive, reported by Minnesota Public Radio, "contamination from unwanted bacteria was hurting efficiency." Production resumed in June 2013.
WHAT THE TEST WOULD HAVE FLAGGED The fermentation yardstick. Biological processes are judged on consecutive clean batches at scale, and contamination control in commercial-volume fermenters is a scale variable a laboratory or pilot campaign rarely exercises. |
Electrochemical Manufacturing: Northvolt's Yield at Rate
SKELLEFTEÅ, SWEDEN · BATTERY CELL MANUFACTURING · PRODUCTION ENDED JUNE 2025
Northvolt began cell production at its Skellefteå plant in Sweden at the end of 2022. By mid-2024, according to electrive, rejected cells were running far too high and deliverable output was far below plan. BMW cancelled a cell order worth about €2 billion in June 2024. Northvolt filed for Chapter 11 in November 2024 and for insolvency in Sweden in March 2025, and production at Skellefteå ended by 30 June 2025.
WHAT THE TEST WOULD HAVE FLAGGED The ratio, measured as manufacturing throughput. When scale means more production lines, the question is how much output has run at target yield and rate, and for how long. |
What Disciplined Scale-Up Looks Like: Shell, Bintulu to Pearl
BINTULU 1993 TO PEARL GTL 2011 · GAS-TO-LIQUIDS · A STEP UNDER 10X
Fischer-Tropsch technology can scale. Shell's first commercial gas-to-liquids plant at Bintulu, Malaysia, started in 1993 and produces about 14,700 barrels a day. Pearl GTL in Qatar, at 140,000 barrels a day, took first gas in 2011, and a Bintulu team helped design, commission and start it up from the Bintulu blueprint, as reported by the Borneo Post. The step from Bintulu to Pearl was under 10x, taken after nearly two decades of commercial operation at the smaller plant.
The Base Rate
RAND · 44 PIONEER PROCESS PLANTS · 34 COMPANIES
RAND studied 44 pioneer process plants built by 34 oil, chemical and mineral companies. More than half failed to reach their production goals in the second six months after start-up (Merrow, Phillips and Myers, 1981). The failures above sit inside that average, and Pearl sits outside it for a reason. The chemistry rarely fails. Scale does.
The chemistry rarely fails. Scale does.
SECTION 03
The Ratio, Ranked
The scale-up ratio is commercial capacity divided by the largest integrated unit that has run successfully. Integrated means feed preparation, reaction, separation and recycle running together, on the feed the commercial plant will buy. A reactor run on bottled syngas does not count toward a gasification-to-liquids ratio.
None of this is a case against new chemistry. The test measures evidence, and it is indifferent to novelty. A first-of-its-kind route that advances in disciplined steps passes. A mature chemistry pushed 500x in one jump fails.
I rank the three tests in this order.
Rank and test | Why it ranks here |
|---|---|
01 · Scale-up ratio | The physics that changes with size, such as heat removal, mixing, oxygen transfer, solids flow and residence time distribution, changes outside the range of the data. No amount of run time at small scale closes that gap. |
02 · Run length | Hours only matter against the failure mechanism. A run should last three to five times the process's dominant degradation cycle. If a unit fouls every 50 hours, a 100-hour run proves almost nothing, and a 500-hour run starts to mean something. |
03 · Balance closure | This is the most abused number in pilot reports, because yields get reported while unaccounted material is left out. Closure within ±5% is the minimum and ±2 to 3% is strong, reported as an average with its standard deviation and the number of balances. A 98% closure repeated 100 times is worth far more than one result of 99.5%. For carbon processes, carbon closure above 95% matters more than total mass, because unaccounted carbon is unaccounted revenue. |
The ranking also tracks money. In the same RAND study, the earliest cost estimates averaged 49% of what the plants finally cost. Cost growth tracked how far the technology departed from plants already built: the number of process steps that were new at commercial scale, and the share of the heat and mass balance that was not based on data from a prior plant. A step that has never run at scale raises the risk that the plant underperforms and the risk that it costs far more than the model says.
Test | Minimum credible | Strong |
|---|---|---|
Commercial capacity ÷ largest integrated unit run successfully | Under 30x | Under 10x |
Largest single scale-up step | Under 10x | Under 5x |
Continuous run | 300 to 500 hours | Over 1,000 hours |
Mass balance closure | ±5% | ±2 to 3% |
Carbon balance closure (carbon processes) | Over 95% | Over 97% |
HOW I READ THE RATIO
Ratio | Assessment |
|---|---|
Under 3x | Low concern |
3 to 10x | Generally acceptable |
10 to 30x | Needs strong engineering justification |
30 to 100x | Significant technology risk |
Over 100x | Unproven |
Pilot A sits at 500x. Pilot B sits at 5x. Shell's step from Bintulu to Pearl sat under 10x. That is the whole answer to the opening question.
The US Department of Energy builds the same idea into its technology readiness scale. Technology Readiness Level 6 marks the step from laboratory to engineering scale and the determination of the scaling factors that let a commercial plant be designed. A pilot that has not produced those scaling factors has not reached TRL 6, whatever its run hours.
SECTION 04
When Scale Means More Units, Not Bigger Ones
Not every technology scales by getting bigger. Battery cells, electrolyzer stacks, membrane modules and many modular reactors scale by numbering up: more identical units, more lines, more stacks. The physics inside each unit stays the same, so the reactor-size ratio looks reassuring. The risk moves elsewhere.
It moves to manufacturing yield at rate, to unit-to-unit variability, and to the balance of plant that ties hundreds of units together: gas and liquid distribution, power electronics, thermal management and controls. Northvolt shows the cost when yield at rate is assumed. For these technologies I apply the ratio to throughput: commercial output divided by the largest integrated output that has run at target yield and rate, across a full production line, for months.
Modularity reduces one scale-up risk and adds another. A diligence process should price both.
SECTION 05
Three Things That Broke Between Pilot and Plant
I evaluated a biomass gasification and Fischer-Tropsch technology for investors. Its pilot results were real. Three problems sat outside anything the pilot could show, and each maps to one of the tests.
Feeding Solids at Scale
TEST AT WORK · SCALE-UP RATIO
At pilot scale, the biomass was sorted by hand, dried and fed at a uniform size through small hoppers. A commercial plant has to move hundreds of tonnes a day of inconsistent, low-density agricultural waste into a pressurized gasifier. That material bridges and plugs commercial lock-hoppers, and holding a gas-tight pressure seal while feeding it continuously is an engineering problem of its own. None of it appears in a pilot fed by hand. Range Fuels' feed system is the public version of the same lesson, and RAND found that plants processing solids consistently performed worse than plants handling liquids and gases, with no improvement in newer plants. This is the scale ratio at work.
Catalyst Poisoning Over Thousands of Hours
TEST AT WORK · RUN LENGTH
The pilot ran on clean, uniform wood chips over short campaigns, so trace contaminants built up too slowly to notice. Agricultural biomass carries volatile alkalis, chlorine, sulfur and phosphorus. At commercial scale, breakthroughs at parts-per-billion levels accumulate over thousands of operating hours and deactivate a Fischer-Tropsch catalyst bed worth millions of dollars well before its design life. This is run length at work. Pilot hours should cover at least 5 to 10% of the catalyst life the financial model assumes. A 100-hour run behind a 20,000-hour life claim leaves the deactivation curve to guesswork.
Heat Removal in the Fischer-Tropsch Reactor
TEST AT WORK · STEP SIZE
Fischer-Tropsch synthesis is strongly exothermic. Small pilot reactors have a high ratio of surface area to volume and shed heat easily. As reactor diameter grows to carry commercial gas flows, the core struggles to reject heat. Local hotspots shift selectivity toward methane at the expense of liquid fuel, and at worst sinter the catalyst. This is step size at work. Heat removal does not scale in a straight line with volume, so heat-transfer-limited reactors need intermediate steps of 3 to 10x.
Surfacing all three before the investment decision moved the discussion from whether the chemistry worked to what the next demonstration unit had to prove.
SECTION 06
Design to the Worst Feed: The Brown Grease Lesson
Brown grease pretreatment, like most chemical processes, is designed to a feedstock specification. Brown grease has no organized market. Much of it comes out of municipal wastewater treatment plants as it is, carrying sulfur, metals, gums, water and free fatty acids that change from load to load.
A pretreatment unit designed to an average specification produced inconsistent output, and neither homogeneous catalysis (esterification and transesterification) nor heterogeneous catalysis such as HEFA hydroprocessing could accept it reliably. We helped the operator find that gap and design a pretreatment system that tolerated the full contaminant range and delivered steady output to the units downstream.
The rule carries to every variable feedstock: used cooking oil, agricultural residue, municipal solid waste, steel-mill off-gas, waste plastics and grid power with a shifting renewable share. Run the pilot on the worst feed in the supply plan, and size pretreatment to the range of the feed, not its average. The plants I have run handled brown grease at up to 100% FFA, and the feed was never the average.
SECTION 07
Evidence Bars by Technology
There is no universal 10x rule. The right bar depends on the physics that dominates as the plant grows.
Technology | What changes with scale | Minimum credible | Strong |
|---|---|---|---|
Simple unit operations: distillation, storage, blending | Hydraulics, well understood | Steps of 50 to 100x | Commercial reference units |
Conventional catalytic reactors: hydrogenation, hydroprocessing, methanol | Catalyst aging, where the reactor type is already commercial | 500 hours; steps of 10 to 20x | 2,000+ hours |
Heat-transfer-limited reactors: Fischer-Tropsch, exothermic hydrogenation, slurry bubble columns | Heat removal, non-linear with size | 1,000 hours; steps of 3 to 10x | Campaigns totaling over 5,000 hours |
Novel catalysts | Deactivation, poisoning, regeneration | 250 to 500 hours, covering 5 to 10% of claimed life | 1,000 to 5,000 hours with an aging curve |
Gasification, pyrolysis and other solids conversion | Feeding, tar, ash, fouling | 500 to 1,000 hours on representative feed; steps of 3 to 10x | Continuous feeding at scale on the worst feed |
Fermentation and biochemical routes | Oxygen transfer (kLa), mixing time, shear, contamination, strain stability | 10 to 20 consecutive successful batches; carbon recovery over 95%; electron balance over 90%; steps under 20x | 30 to 100 batches across a full campaign |
Electrochemical cells and stacks: electrolyzers, CO2 electrolysis, electrosynthesis | Stack degradation | 1,000 hours | 5,000 to 20,000 hours; 10,000+ hour equivalent for project finance |
Numbered-up manufacturing: battery cells, stacks, membrane modules | Yield at rate, unit-to-unit variability, balance of plant | A full line at target yield and rate | Months at target yield across the line, with integrated balance of plant |
Fermentation needs a different yardstick. One long run proves less than many batches, because the risks are reproducibility, contamination and strain stability. Thirty clean five-day batches show more than a single 1,000-hour campaign. Oxygen transfer and mixing cannot both be held constant as vessels grow, which is why bioprocess scale-up moves through intermediate fermenters. A direct jump from a 100-liter fermenter to a 100,000-liter commercial train is a 1,000x step and should stop a diligence process until it is explained.
SECTION 08
Where Simulation Is Enough, and Where It Is Not
A 2025 paper in Processes by Lütge, Schmidt, Köster and Strube of Clausthal University of Technology argues that demonstration plants are often an expense investors impose without needing them. Its authors report that investors without chemical engineering training want a scale factor of about 25 below the commercial plant, and propose that process modeling, a mini-plant and a quality-by-design risk study can replace the demonstrator, saving about 30% of investment and three to eight years. Its case study is a simulated moving bed separation of cresol isomers.
For that class of process, I agree. Liquid separations with well-characterized thermodynamics and hydrodynamics model well, and a validated model is a fair substitute for a demonstrator. Models are also how the scaling factors behind TRL 6 get used.
The argument does not carry to the processes where most first-of-a-kind energy transition plants sit. Solids flow, ash and tar behavior, trace-contaminant poisoning, microbial behavior at scale, heat removal in large exothermic reactors and manufacturing yield at rate are the variables models predict worst, and they are the ones behind the failures above. The paper's own investors also land on a scale factor of about 25, close to the 30x bar here. The real disagreement is over what counts as demonstrated scale.
SECTION 09
The Pilot Evidence Card
In diligence on a supercritical fluid process and a hydrothermal liquefaction process, I mapped every run and trial each company had completed by scale, duration and feedstock against these tests. Investors and the companies' boards could then see how far each technology had actually progressed, which claims rested on demonstrated data and which rested on models, and what the next unit had to prove before a commercial plant.
The card below is the one-page version, built to sit at the back of an investment committee memo. A single fail on the scale-up ratio outweighs passes on every other line.
Measure | Minimum credible | Strong | Ask management for |
|---|---|---|---|
Scale-up ratio | Under 30x | Under 10x | Capacity of the largest integrated unit run on representative feed, with dates |
Largest single step | Under 10x | Under 5x | The full scale path from lab to commercial, step by step |
Continuous run | 300 to 500 hours, and three to five times the degradation cycle | Over 1,000 hours | Run logs with every shutdown and its cause, not the best campaign summary |
Catalyst life coverage | 5 to 10% of claimed life | A measured aging curve | Deactivation data and regeneration history |
Mass balance closure | ±5% | ±2 to 3% | Average, standard deviation and number of balances |
Carbon closure | Over 95% | Over 97% | Carbon in and out by stream, including the unaccounted share |
Feedstock | Worst feed in the supply plan tested | Full feed range tested | Analyses of the feed tested against the supply contract specification |
Fermentation | 10 to 20 consecutive batches | 30 or more | The full batch record, including failed and contaminated batches |
Numbered-up systems | One full line at target yield and rate | Months at target yield | Yield, scrap and rate by line and by month |
SECTION 10
Frequently Asked Questions
What is a scale-up ratio in chemical engineering?
The scale-up ratio is commercial capacity divided by the largest integrated unit that has run successfully on representative feed. A ratio under 10x is strong evidence, under 30x is the minimum credible bar, and over 100x is unproven. It predicts first-of-a-kind failure better than run hours or balance closure.
How long should a pilot plant run before commercial scale-up?
Long enough to outlast the process's dominant failure mechanism three to five times over. In practice that means at least 300 to 500 continuous hours, over 1,000 hours for lender-grade evidence, and enough hours to cover 5 to 10% of any claimed catalyst life. Fermentation is judged on consecutive successful batches.
What scale-up factor is safe for a first commercial plant?
Commercial capacity within 30 times the largest integrated unit that has run successfully, and within 10 times for strong evidence. Single steps should stay under 10x, and between 3 and 10x for gasification, solids processes and heat-transfer-limited reactors. Simple unit operations such as distillation can scale 50 to 100x.
What mass balance closure should investors accept?
Within ±5% as a minimum and ±2 to 3% for strong evidence, reported as an average with its standard deviation across repeated balances. For carbon processes such as gasification, Fischer-Tropsch and SAF, carbon closure above 95% matters more than total mass, because unaccounted carbon is unaccounted revenue.
Why do first-of-a-kind chemical plants fail?
Most fail on scale, not chemistry. RAND found that more than half of 44 pioneer process plants missed production goals in their second six months. The usual causes are solids handling, contaminants that build up over thousands of hours, heat removal in larger reactors, contamination in fermenters and manufacturing yield at rate.
Can process simulation replace a pilot or demonstration plant?
For liquid separations with well-known thermodynamics, a validated model plus a mini-plant can replace a demonstrator. For solids handling, biological processes and heat-removal-limited reactors it cannot, because those are the variables models predict worst and the ones behind most first-of-a-kind plant failures.
SECTION 11
Before the Next Tranche
BEFORE CAPITAL MOVES If a pilot record is about to carry a term sheet, a first-plant tranche or a final investment decision, our scale-up gate review applies this card to the company's own data. Counsel in a dispute over scale-up representations will find our chemical engineering expert witness practice set up for exactly that question. |
ABOUT THE AUTHOR Gaurav Shah is Managing Partner of Trident Renewables. A chemical engineer (M.S. Chemical Engineering, University of Southern California) with 6 US patents and 1 PCT patent in waste-to-biofuel technologies, he has built and run 20+ renewable energy plants and sits on investment committees at VC firms in the US and India. |
SOURCES
RAND, A Quantitative Assessment of R&D Requirements for Solids Processing Technology, R-3216
National Energy Technology Laboratory, Technology Readiness Level definitions
Biofuels Digest, KiOR: The Inside True Story of a Company Gone Wrong, Part 5, 2016
Bloomberg, New-Fuel Hopeful Kior Opens First Major Wood-to-Oil Plant, 8 November 2012
Jackson Free Press, KiOR Files for Bankruptcy, but Not Miss. Unit, 10 November 2014
Associated Press, KiOR puts Lowndes Co. plant in idle, 18 March 2014
MIT Technology Review, The Death of Range Fuels Shouldn't Doom All Biofuels, 15 December 2011
Waste Today, Clean fuels company verges on collapse after financial troubles, layoffs, 2024
Minnesota Public Radio, Gevo resumes isobutanol production, 18 June 2013
electrive, Northvolt discontinues production at its main plant in Sweden, 22 May 2025
Electrek, BMW cancels $2 billion battery cell order from Northvolt, 20 June 2024
Borneo Post, Shell Malaysia marks 25 years of success in Bintulu, 29 August 2018
Comments