1. Probability-weighted process mix removed
15 Nov 2026 · v0.3
There's no point imagining a frontier plant which uses a worse process, and anyway process doesn't seem to be the major issue here.
Alex Bates's rebuild, as he shared it (file v0.9.7, 16 November 2026), hosted by The Unjournal for discussion. Only the typeface has been changed to match this site. The main Unjournal model remains the master copy (Alex suggested this; neither model is fully validated). Comment with Hypothes.is: the ‹ tab on the right edge.
Unjournal pivotal questions · cultured meat · rebuild v0.9.4
For usability, a very simple comparison between dry cell mass and commercially available chicken is assumed.
Changes to discuss with the team.
15 Nov 2026 · v0.3
There's no point imagining a frontier plant which uses a worse process, and anyway process doesn't seem to be the major issue here.
15 Nov 2026 · v0.3
There's no point imagining a frontier plant which contracts manufacturer - if there was, we should just model the costs of the contract manufacturer.
15 Nov 2026 · v0.4
Relevant to the scope.
15 Nov 2026 · v0.4
Not relevant to the scope.
15 Nov 2026 · no model change
Finding. The Learn page puts the seed train at 5–10% of production cost, and a few lines later at 5–15%, "mostly labour". It cites Humbird's $1/kg labour figure, but that covers the whole plant's staff (95 people for fed-batch, 132 for perfusion) against a total cost of $37/kg, so all plant labour is about 3% of cost. The page also says seed-train medium volumes are orders of magnitude smaller than in production. In the model the seed train is equipment only, about 0.5% of cost, and that capital ratio is consistent with Humbird (about $2M of seed-train reactors against $34M of production reactors).
15 Nov 2026 · no model change
Finding. The cost breakdown summary says capital is 15–35% of total cost and fixed running costs 10–25%. In the model they are about 7% and 3.5% across all worlds, but about 13% and 22% in the cheapest 10% of worlds. Published cost studies model plants near parity with conventional meat, so their shares match the model's cheap worlds, not its typical one.
16 Nov 2026 · v0.6
The original bought a blended share of reactor capacity (55% ± 15) at food-grade prices and the rest at pharma prices. This doesn't make sense at a frontier factory, and is more relevant as an industry average.
16 Nov 2026 · v0.8
There's no point imagining a frontier plant which uses a worse process.
16 Nov 2026 · corrected in v0.8
Repeated issue with the model where technology which is cheaper on average is used even when simulated result makes it more expensive in some particular world.
16 Nov 2026 · no model change
Finding. The share of output lost to failed runs grosses up every cost line, including harvest. But a failed run is thrown away, not harvested, so harvest shouldn't scale with failures. Recovery losses in the harvest step itself do belong there.
16 Nov 2026 · v0.8.6
Change. (1) Reactor prices are entered as two prices, pharma-grade and food-grade (default 0.35 × pharma-grade), drawn together so they move as one; this replaces the original's separate 0.15–0.55 ratio. (2) The whole-plant multiplier also has a food-grade column, since a food-grade plant may need far less cleanroom space, sterile pipework and validation (default: no difference). If food-grade reactors work, the plant uses whichever installed cost is lower. (3) Fixed running costs ($1–6/kg wet at a 20 kt reference plant, scaled by plant size and utilisation) and utilities and consumables ($0.30–3/kg wet) are merged into one plant operating cost per kg: central $23.50, 95% optimistic $9.00 per kg dry ($4.70 and $1.80 per kg wet), matching the median and 2.5th percentile of their sum. The fixed-cost scale exponent goes with them.
16 Nov 2026 · corrected in v0.8.7
The original model has no limit to the scaling factor of capital, which means the best plant is just one which grows massive. This is not plausible, and leads to spurious effectiveness.
16 Nov 2026 · v0.9.4
Errors in the original model, corrected here.