Where the published numbers disagree

In one paragraph. Several of the figures circulating about Vietnamese agriculture and about alternative protein economics do not survive being checked against the source they are attributed to. Rather than average the disagreements away or quietly pick one number, every case is recorded here with the position this report takes. Anyone quoting this work, or quoting the same underlying sources, should know which figures are safe and which are not. Three entries marked new in this version record disagreements between this report and an external review of it. Codes carry a topic prefix: R regulation, F fermentation, CF crops and feedstocks, NP plant protein, CS cultivated meat and cross-cutting.

Regulation#

R-C1. "Cash support of up to 10 percent of new fixed asset investment" against the decree text. A widely cited Asia-Pacific site-selection study presents the grant under Decree 182/2024/ND-CP as up to 10 percent of new fixed asset investment. The decree applies 1 to 10 percent of the cost of new fixed assets capped at 0.5 percent of total investment capital, and only to enterprises meeting high capital or revenue thresholds. The position taken here: the decree text wins. The headline rate is accurate and the description of what a first-of-kind alternative protein plant could claim is not. A USD 20 to 50 million fermentation plant clears none of the thresholds. The policy ask is therefore the eligibility floor, not the rate. Amended in this version. The first version of this report stated those thresholds incorrectly while criticising the advocacy source for misdescribing the same decree. The corrected figures are in section 7.6, and they should themselves be confirmed against the decree text. The position taken is unchanged.

R-C2. Whether the feed pathway for a novel raw material is a thirty-working-day registration or a multi-stage evaluation. New in this version. This report states a 5-plus-25-working-day procedure under Circular 04/2020, and simultaneously records, at R-G2, that no source explains how a new single raw material joins the permitted circulation list. An external review asserts that Article 11 of Decree 13/2020 applies instead, requiring compositional testing, heavy metal assays, mycotoxin screens and controlled feeding trials, at 12 to 18 months. The position taken here: this report's own claim is withdrawn, because it was never supported and contradicted its own open-question register. The review's replacement is not adopted either, because its only citation is the source this study found silent on the question and its timeline carries no citation at all. Until action 3 is answered, the honest statement is that the feed regime has a defined authority and dossier and an undocumented timeline for this case. Anyone planning around a fast feed approval is planning around a figure nobody has published.

Fermentation#

F-C1. "Limited feedstock supply at scale" against a large exportable cassava surplus. The same site-selection study scores Vietnamese feedstock 26 out of 100 using sugar, and reads low export volumes as evidence of no surplus. Vietnam produced about 10.5 million tonnes of cassava roots from 517,800 hectares and exported 2.3 million tonnes of cassava and cassava products worth USD 711.5 million in the first half of 2025, ranking third globally. The position taken here: both are right about different things. For precision fermentation, which runs on refined sugar or glucose, the pessimistic score holds unless cassava glucose is substituted. For biomass fermentation, which runs on starch and woody side streams, sugar is the wrong denominator. Evidence strength: the cassava figures come from a Vietnamese government trade portal and were consistent across two retrievals; the sugar figures come from a single industry-commissioned study.

F-C2. A widely reproduced table assigns biomass values to the wrong organisms. A version of the results from Karimi et al. (2019) in circulation reports Neurospora intermedia at 118.5 g of dry biomass per litre and Aspergillus oryzae at 34.3 g per litre. The paper reports the opposite: A. oryzae 118.5 ± 3.9, N. intermedia 85.1 ± 3.9, R. oryzae 34.3 ± 2.4 g per litre, in 5 percent vinasse, 250 mL flasks with 100 mL working volume, over 72 hours. The position taken here: the table in section 10.1 was read from the paper and is the correct version. Anyone building a case on these numbers should open the paper rather than a secondary source, and should note that the headline value is unusually high and its methods have not been examined by anyone in this project.

F-C3. Two downstream cost figures could not be substantiated. Two claims are attributed to Karamerou et al. 2020: that fermentation, harvesting and drying together account for about 87 percent of total process cost, and that removing drying cuts cost from USD 1.81 to USD 1.55 per kilogram. Neither appears in the accessible abstract, which instead gives wet extraction at USD 1.16 per kilogram. The position taken here: treat the 87 percent split and the USD 1.55 figure as unverified and publish neither. The scenario costs quoted in section 10.6 come from the abstract and are the safer set.

F-C4. The capital cost benchmarks are not comparable and must not be averaged. EUR 5,000 per cubic metre is a feed-grade estimate attached to an anaerobic digestion plant, resting on a 2003 cost correlation. USD 13,800 per cubic metre is a 2024 vendor quotation for a bioreactor in an industrial chemical plant making isopropyl alcohol. Neither is a food-grade sterile protein plant, and the two differ by a factor of about three before any adjustment for grade. The same caution applies to the per-kilogram figures: the lowest numbers in section 10.6, at USD 1.20 and USD 0.81, come from a microbial oil model rather than a protein one. The position taken here: quote every figure with its basis attached, or quote none of them. The honest statement is that no food-grade sterile fermentation capital cost per cubic metre exists in the accessible literature.

Crops and feedstocks#

CF-C1. Vietnam's rice straw resource has four national estimates and the extremes differ by more than a factor of two. About 43 million tonnes follows from the one-to-one straw-to-paddy basis applied to 43.12 million tonnes of paddy. A 2025 mushroom industry report gives 42 to 47 million tonnes. One Vietnamese paper states 70.7 million dry tonnes. An energy-potential study calculates approximately 97 million tonnes, alongside 9 million tonnes of rice husk, while itself conceding a significant lack of data on availability and quantities. The position taken here: quote roughly 43 to 47 million tonnes, attribute it explicitly to the straw-to-paddy ratio basis, and note that the two independent-looking low estimates may rest on the same ratio rather than corroborating each other. The two higher figures exist and are not corroborated. This is publishable as a finding in its own right, and it is a reason to treat any straw-based business case with suspicion until the resource is measured rather than inferred.

CF-C2. Competing-use percentages attributed to a source could not be confirmed in it. A set of Vietnamese shares is attributed to the Agronomy energy-potential study: rice straw burned in the open at up to 95 percent, rice husk 70 percent incinerated in rural Vietnam, and bagasse about 70 percent to cogeneration with over 30 percent excess. A targeted read of that paper surfaced its tonnage figures and its statement that availability data are lacking, but not those percentage shares. The position taken here: this is not proof the shares are absent from the paper, since one read can miss table content, but it is enough that none of them should be quoted until someone finds them in the PDF. The regional shares from Thailand, the Philippines and Punjab in section 8.4 are on firmer ground because each was given with its own tonnage.

CF-C3. A bagasse crude protein figure of 12.50 percent is implausible. Sugarcane bagasse is a woody residue and its crude protein is normally on the order of 1 to 2 percent. The position taken here: not used anywhere in this report. It is most likely a row-assignment error of the same kind as F-C2, and a reason to treat single-row extractions from secondary compilations as unverified.

CF-C4. A dried cassava chip storage loss of 50 percent per month is not credible. No commercial dried-chip inventory could lose half its mass monthly; the cassava chip trade would not exist if it did. The position taken here: almost certainly a misreading of 0.5 percent, or a model parameter meaning something other than mass loss. Not used. The storage conclusion that does survive is the two to three day limit on fresh roots, which is consistent across sources and is the operationally important number.

CF-C5. Sugar industry figures differ across sources and years, which is not a contradiction but reads like one. The 2020/21 season is reported at 7.498 million tonnes of cane, 25 mills and 0.763 million tonnes of sugar. One study gives 1.35 million tonnes of sugar for 2024. A trade portal gives over 1.2 million tonnes by the end of May in the 2024/25 season, with 25 to 30 operating mills, down from more than 40. The position taken here: these are consistent with a sector that contracted in mill count and then recovered in output after trade measures against imported sugar. Quote them with their years attached and do not build a trend line from three sources using different methods.

CF-C6. Aquafeed tonnage against addressable fishmeal. New in this version. Earlier versions of this report quoted a 3.4 to 4.0 million tonne aquafeed market as the market for a fishmeal-replacing ingredient. An external review, working from the same underlying census, gives species-level fishmeal inclusion that puts most of that tonnage in pangasius at 2 to 5 percent inclusion, and totals the displaceable fishmeal pool at 350,000 to 450,000 tonnes. The two accounts are not in conflict as facts; this report used the wrong one of them as its headline. The position taken here: quote the displaceable pool, not the feed tonnage, and treat pangasius as a separate functional-ingredient proposition rather than a fishmeal-replacement market. The two accounts also differ on sub-totals, giving shrimp at 1.0 to 1.07 Mt against 920,000 t and an overall 3.78 to 4.47 Mt against 3.4 to 4.0 Mt, which should be reconciled in one reading of the census or, better, by asking the mills.

Plant protein#

NP-C1. A single tidy DIAAS comparison does not survive a targeted follow-up. A first pass gave DIAAS as soy 91, mung bean 86, pea 70. Asked for values backed by named sources, the same search returned mung bean DIAAS spanning 57.7 to 93, with cooked mung bean at 68, 76, 86 and 93 in different studies and age groups, PDCAAS 0.58 to 0.76, and digestibility from 52 percent in the laboratory to 89.4 percent measured in animals. The position taken here: the range is the finding and the point estimate is an artefact of picking one study. Publish ranges with their basis, never a single DIAAS number. It also means that any competitor or advocacy document quoting a single figure should be read as having made a choice.

NP-C2. Rice bran protein's limiting amino acid is reported three different ways. One source calls rice bran protein "lysine-rich" and then states that the limiting amino acid is the sulfur amino acids. Another lists rice bran among the proteins limited by lysine. The quote usually offered as decisive compares cystine and methionine content between rice bran protein concentrate, isolate and whole protein, which is a comparison between fractions and not a score against a reference pattern. An independent review adjudicates it: rice bran has 24 to 43 percent albumin against 4 to 8 percent in the endosperm, "so it displays a higher content of lysine", while "in rice, the lysine acts as the limiting amino acid". The position taken here: both statements are partly right. Rice bran protein is rich in lysine relative to milled rice protein, lysine remains limiting for rice protein generally, and some rice bran fraction studies find the sulfur amino acids lowest. Do not quote a single limiting amino acid for rice bran protein without naming the fraction and the reference pattern.

NP-C3. Bench-scale results are not commercial results, and the gap runs the wrong way for optimism. Bench-scale mung bean isolates show nitrogen solubility above 85 percent while commercial pea and soy isolates show below 60 percent, and laboratory-extracted soy emulsifies far better than commercial soy. The position taken here: this is not evidence that mung bean beats commercial soy. It is evidence that commercial processing destroys performance in every legume protein tested, and that a commercial mung bean isolate would probably lose much of its bench advantage. Any product claim built on bench data is a claim about a product that does not exist yet.

Cultivated meat and cross-cutting#

CS-C1. Approval lists differ on which jurisdictions have granted approval for cultivated meat. One compilation reports approvals granted in Singapore, the United States, Israel, the United Kingdom for pet food, and Australia and New Zealand for quail. A February 2026 trade publication lists Singapore (Good Meat chicken, Vow quail, Parima chicken), the United States (Upside Foods, Good Meat, Wildtype salmon, Mission Barns pork fat, Believer Meats poultry) and the United Kingdom for pet food as approved, while placing Israel, Australia and New Zealand in a category of frameworks in development or safety review completed. The position taken here: the disagreement is probably about what counts as approval, since safety clearance, market authorisation and a licence to manufacture are separate steps in several of these systems. Do not quote a count of approving jurisdictions. Quote the pattern, which both sources agree on: approvals are few, narrow, and concentrated in Singapore and the United States. Amended in this version. The Wildtype coho salmon clearance is dated 28 May 2025 and took the form of an FDA "no questions" letter. That a May 2025 clearance was still described as non-existent in a report compiled in September 2026 is a currency failure in this report, not only a disagreement between sources of different vintage. The date and form should themselves be confirmed against the FDA letter rather than trade coverage.

CS-C2. A documented blend DIAAS above 100 is weaker than the field's rhetoric. Complementation between a cereal protein and a legume protein is real and directionally well supported, but a measured cereal-legume blend DIAAS at or above 100 does not appear in the evidence set. Blends reaching 100 were shown by calculation or simulation. The position taken here: the field routinely presents calculated blend scores as though they were measured. This report does not, and says so, which is the cheapest credibility available in this area.

CS-C3. Availability of aquatic cell lines. New in this version. Earlier versions of this report stated that no publicly available immortalised cell lines exist for catfish, tilapia or salmon and no publicly available crustacean cell lines at all. An external review names continuous thymus and gill lines from Pangasianodon hypophthalmus deposited by ICAR-NBFGR, and a hybrid Penaeus monodon lymphoid line fused with Sf9. The position taken here: this report's claim is narrowed to food-grade commercial production lines, which both accounts support. The accessions should be confirmed. The correction makes the recommendation in 12.2 more actionable, not less, but it also retires this report's claim that a Vietnamese cell line would be a unique global contribution: the contribution available is adaptation to food-grade serum-free suspension culture, which is a narrower and more accurate claim.