A $50M cultivation expansion. A department of PhD agronomists, design engineers, architects and the tradesmen who built it. Total late-cycle mortality, no pathogen, no spread, no answer. This is the fishbone that found it — and you are invited to work it yourself.
The expansion added fifty percent more flower canopy to an operating facility — engineered from scratch, built by top-tier trades, commissioned as a state-of-the-art environmental system. It landed in the same quarter the adult-use market opened. Demand was effectively unbounded; every gram of capacity was already sold.
The first cycle in the new rooms went textbook through vegetative growth and into early flower. Then, at roughly week six or seven — the point at which a plant has consumed nearly all of the labor, nutrient, power and time it will ever consume, and is finally laying down the bulk that becomes product — the crop began to die.
Not stall. Not underperform. Die, completely and without exception, in a condition that yielded nothing salvageable. Maximum sunk cost, zero return. The loss was booked against the most expensive plants in the building.
Every discipline that touched the project was put on it: cultivation staff with decades of grow experience, agronomists and horticulturists with doctorates, the mechanical engineers who designed the system, the architects who wrote the specification, the contractors who built it. Weeks in, there was no hypothesis that survived contact with the evidence.
One hundred percent mortality. No partial recoveries, no resistant genetics, no rooms spared.
Onset clustered in the final two weeks, exactly as growth rate and biomass demand peaked.
No radial spread from an index plant, no hot spots, no gradient. Pathology behaves like a map; this did not.
Identical genetics, inputs and SOPs in the legacy rooms performed normally. Only the new building killed plants.
This is what made the problem unsolvable by inspection. The building management system reported every commissioned setpoint in tolerance, continuously, for the entire cycle. Nothing alarmed. Nothing drifted. Every instrument agreed with every other instrument.
The sheet opposite is empty because nobody had pulled it yet. It fills in as you work the diagram below — every investigation you run adds its instrument record here, and every record comes back clean. A monitored parameter that reads normal draws attention; an unmonitored parameter draws none at all.
No instrument records pulled yet. Work the fishbone below — each investigation files its reading on this sheet.
Eighteen candidate causes, sorted 6M with the domain function each one belongs to. Investigate any of them. Each investigation consumes standing crop while the cycle runs on — pick badly and the product burns down in front of you. One cause is the answer.
Nothing eliminated yet. Tap a candidate cause on the diagram to run its test.
The facility was designed for controlled CO2 augmentation. The bulk tank that would supply it had been fabricated and was sitting in a yard waiting on a concrete pad to be poured, so the gas train was never connected — correctly judged as harmless, because plants pull carbon dioxide out of ordinary air. Augmentation is an accelerant, not a requirement.
What no one connected to that decision was the air handling. Designed around augmentation, the rooms ran fully recirculating with the outside-air dampers shut, so the atmosphere would hold the injected gas instead of exhausting it. With no injection and no fresh air, the crop drew the CO2 down and there was nothing to replace it.
Through veg and early flower the demand was low enough that the residual held. Late flower is when a canopy does its heaviest carbon fixation — and in a gymnasium-sized sealed room, that demand stripped the air below the compensation point. The plants asphyxiated themselves, at the exact moment they were worth the most.
It presented as fatal, universal, non-communicable and confined to the new building because it was not biology at all. It was an unbuilt utility interacting with an air handling sequence — an interface failure between two scopes of work, each of which was individually correct.
Reconstructed from cycle records. Ambient outside air sits near 420 ppm; the compensation point for a heavily loaded canopy under 900+ µmol of light is far above the 15 ppm the sealed rooms reached at end of cycle — effectively nothing left to fix.
Carbon starvation — photosynthesis could not meet the demand of peak biomass accumulation.
Room CO2 had been drawn down to 15 ppm — a fraction of the 420 ppm ambient air would hold.
The rooms were recirculating with outside-air dampers closed — no fresh-air makeup.
The sequence was written for a CO2-augmented room, where retaining the injected gas is the correct behaviour.
The bulk CO2 tank was waiting on a concrete pad, so the gas train was never commissioned — and no one owned the gap between "not yet installed" and "already sequenced as if installed."
Introduce outside air and let ambient atmosphere supply the carbon until the tank was commissioned. Engineering agreed on the spot; the sequence was changed in about five minutes. Plants stabilised within the day and the cycle was carried to harvest.
Commissioning was amended so no room could run sealed without a verified CO2 source, and atmospheric CO2 was added to the monitored, alarmed parameter set.
A $50M cultivation expansion added fifty percent more flower canopy to an operating facility. The first cycle in the new rooms ran textbook through vegetative growth and into early flower, then died completely at week six or seven — the point at which a plant has consumed nearly all the labor, nutrient, power and time it will ever consume. One hundred percent mortality, nothing salvageable, booked against the most expensive plants in the building.
The failure signature defeated the textbooks. It was fatal, late-cycle only, non-communicable — no radial spread, no hot spots, no gradient — and confined to the new building, while identical genetics and SOPs performed normally in the legacy rooms. The building management system reported every commissioned setpoint in tolerance, continuously, for the entire cycle. Nothing alarmed and nothing drifted.
The root cause was carbon dioxide. The facility was designed for controlled CO₂ augmentation, but the bulk tank sat in a yard waiting on a concrete pad, so the gas train was never connected — correctly judged as harmless, since plants pull CO₂ from ordinary air. What no one connected to that decision was the air handling: designed around augmentation, the rooms ran fully recirculating with the outside-air dampers shut. With no injection and no fresh air, the crop drew the CO₂ down until, at peak late-flower carbon fixation, it fell below the compensation point. The plants asphyxiated themselves at the moment they were worth the most.
It presented as biology and was not. It was an unbuilt utility interacting with an air-handling sequence — an interface failure between two scopes of work, each individually correct. $16.4M of product recovered, five minutes to implement the fix, no capital required.