29. What frontier actors asked, and what the answers changed

In one paragraph. Part III was written for the people who could move alternative protein in Vietnam, so we asked whether it answers their questions. We named 80 real actors in four groups, read what each says it funds, buys, builds or regulates, turned that into 118 decision questions and checked each against Part III. It answered one in full and 71 in part; for most, the evidence behind the answer was weak or missing. The gaps were of a kind: Part III sized channels and prices well but did not map who decides, which rules already exist, who would pay, who already competes, or what a deal would look like. We then ran sixteen research lines in three waves, each asking a different kind of question, and stopped when the next candidates would add citations without changing a decision. The core conclusion survives: demand pulls new protein as an ingredient inside familiar foods, sold to incumbents. What changed is how to act on it: schools already reserve plant-protein days that tofu fills; a domestic textured-protein line is a debt-financed import-substitution project with no tariff shelter; soy is the price a new extender must beat; most of the label rules Part III proposed are already law; a sugar tax may reach sweetened plant milks from 2027; and only four of the original nineteen demand moves have a funder whose rules fit and that accepts Vietnam.

The method, the yield of each line and the reasons for stopping are in M4. Actor check and waves. The corrections are folded into chapters 11 to 18 and 25, the briefs and the executive summaries.

#29.1 Who we asked about

We identified 20 actors in each of four groups from their own documents: requests for proposals, strategy pages, annual reports and meeting documents from 2024 to 2026, laws and programme documents ^1,2,3,4 VN-directMedium stated. For each actor we wrote the demand questions it would need answered before acting in Vietnam, and checked Part III for an answer.

Source: actor_register.csv, actor_questions.csv; our assessment VN-directMedium inferred.

How well Part III answered the actors

118 demand questions from 80 named actors, by group and how far v0.3 answered them

Across all 118 questions the evidence was moderate for 27, weak for 49 and absent for 42; none was strong. Of the 72 answered in full or in part, 26 had moderate evidence, 41 weak and 5 none (our assessment) VN-directMedium inferred.

#29.2 What Part III missed

The misses share a pattern. Each is a different kind of question from the ones Part III asked, not a harder version of them.

• Decisions, not attitudes. Market-shaping funders count "an identifiable decision-maker, a reason they might move, and a window" ¹ generalHigh stated. Part III sized canteens and schools but did not say who writes the menus, which rules govern them or when those rules change.
• Rules already on the books. Several moves Part III proposed (school guidance without an animal share, percentage labels on blends, chay definitions) turned out to exist already, and a new tax it did not see (the sugary-drink excise) bears on its one proven modern category.
• Transactions, not pools. Investors now ask for "strong offtake agreements", "contracted revenue" and "real unit economics" ^2,5 generalMedium stated. Part III named likely first buyers from labels and trade data but had no price, cost, margin or order.
• Who pays. Part III told funders to fund measurement, without checking whose rules allow it.
• Who competes. Part III framed competition as Chinese commodity protein; regional suppliers and cheap domestic dry soy pieces already sit inside the channels it recommends.
• Units. Welfare and climate funders count animals and CO2e; Part III counted tonnes of protein.

#29.3 What the answers changed

Sixteen lines followed (listed in expansion_lines.csv). These are the findings that change what an actor should do.

Schools already have plant-protein days, and tofu fills them. Part III said school meals follow an old target to raise the animal share of children's protein. That target (Decision 1340, 2016) ended in 2020. The Ministry of Health's school-meal guidance of December 2025 sets no animal share and asks for legumes or tofu at least twice a week ^6,7 VN-directHigh revealed. In 48 published school-weeks from Ho Chi Minh City, Hanoi and Hai Phong, plant protein appeared on 15.1% of days, 9 weeks met the rule, tofu was served beside meat, and no dish used textured soy or mock meat ^8,9 VN-directHigh revealed. The menu software schools use still prints an "animal protein at least 48%" standard ¹⁰ VN-directMedium revealed. So a plant ingredient in the plant slot competes with tofu and replaces tofu, not meat; the only school route that removes meat is a labelled blend in minced-meat and fish-cake dishes, and the rule window is the School Health Programme 2026 to 2035 and the menu tools that follow it ^11,12 VN-directMedium revealed. 15.1% of days (Channels)

The Food Safety Law window runs into 2027. The National Assembly takes first comments in October 2026 and considers passage in May 2027 ¹³ VN-directMedium. Every supply-side and demand-side recommendation tied to "the October to November 2026 session" now has until early 2027 (Rules, Policy options).

A domestic textured-protein line is an import-substitution project, not a venture thesis. Chinese, ASEAN and Korean textured protein, isolates and gluten enter duty-free under ACFTA, ATIGA, AKFTA and RCEP ^14,15 VN-directMedium. Chinese dry-textured-soy lines are listed at USD 11,000 to 100,000; installed in Vietnam a line costs about USD 0.3 to 1.4 million, and working capital is the larger need ¹⁶ VN-directLow. Whether it pays turns on the delivered price of food-grade soy flour, which we could not verify ^17,18 VN-directLow inferred. The benchmark path supports about one line by 2035. The natural owner is a chay maker, co-packer or distributor with buyers, financed by bank debt and trade finance. 0% (Demand to frontier)

Soy, not meat, is the price a new extender must beat. Replacing a fifth of the lean in chả lụa (Vietnamese pork roll) with hydrated textured soy saved money at every hog price since 2019, so processors already own the cheapest extender ^19,20,17 VN-directLow inferred. A fungal or novel extender must replace more meat unnoticed than soy, or carry a label buyers prefer. No national standard sets a minimum meat content, and a named meat must already be quantified on the label ^{21,22,23,24,25} VN-directHigh. What buyers punish is hidden cost-cutting, called độn (filler), not a declared ingredient ^26,27 VN-directMedium revealed (Frontier demand).

Most of the label moves are already law. A high-protein claim needs 10 g per 100 g under a voluntary Codex-based standard ²⁸ VN-directHigh; the national vegetarian standard allows egg, milk and honey ²⁹ VN-directHigh; no rule reserves the words for meat or milk; and the GMO label applies only above 5% of ingredients, so lactoferrin-type uses are not caught ³⁰ VN-directMedium. The remaining asks are enforcement, one definition (unqualified chay should mean vegan) and one agency ruling on recombinant proteins (Demand to frontier).

A sugar tax may reach plant milks. From 2027 drinks above 5 g sugar per 100 ml pay an 8% excise (10% from 2028) and dairy is exempt ^31,32 VN-directHigh. Whether soy and nut milks are in scope is unresolved. New plant milks and the isolates that go into them should be designed at or below 5 g (Consumers).

Retailers' own labels are a route to shelf; corporate pledges are not a volume lever. Saigon Co.op's Co.op Select chay line is made by a named co-packer, so a protein upgrade can reach shelf without a consumer brand ^33,34 VN-directMedium revealed. Only one published corporate commitment binds Vietnamese units (Accor's 50% plant-based dishes by 2030, which counts dishes, not protein), and no Vietnamese protein buyer publishes a target ^35,36,37 VN-directMedium revealed (Channels, Business buyers).

Feed demand is revealed but priced low. Vietnam imported about 9,200 t of inactive yeast (food and feed together) in 2025, 95% of it at about USD 1.1 per kg, while the main aquaculture feed standards neither require nor reward microbial protein ^38,39 VN-directMedium revealed. A Vietnamese microbial feed ingredient must win head to head against the yeast mills already buy (Export demand).

The export base is wrapped foods. Vietnam exported USD 128 million of dumplings and other stuffed pasta in 2023, against almost no textured protein ^40,41 VN-directHigh revealed. European retailers' protein targets are flat and met with European products; the buyers are importers and brand owners who require IFS or BRCGS certification. Rules of origin bind on imported flour in the wrapper, not on the soy in the filling, so they give domestic textured soy no export customer ⁴² VN-directHigh (Export demand).

For school-age children protein is not the gap. Stunting is 18% (2023) and 32% among ethnic minority children ⁴³ VN-directHigh, but it is set before age two. Upland boarding-school lunches and dinners carried meat, fish or egg in 240 of 242 published meals, and weighed boarding diets met protein and zinc; the documented shortfalls are breakfast, calcium and vitamin A ^44,45 VN-directMedium revealed. A plant product is not the answer to a school protein gap that does not exist (Protein diet).

Only four of the original nineteen demand moves have a payer. The largest new market-shaping fund excludes whole-food plant-based work, culture change and individual diet change, and prioritises Japan and Korea ¹ generalHigh stated; the regional diet-change fund excludes Vietnam ⁴⁶ VN-directMedium; the leading animal charity evaluator's grants now exclude diet change ⁴⁷ generalMedium. In funder units, the benchmark path spares about 2.4 to 2.7 million animals a year by 2035, nearly all chickens and ducks, and avoids about 0.14 Mt CO2e; import substitution spares none (our estimate) VN-directLow estimate inferred. 4 of 19 about 2.4 to 2.7 million (Demand sizing)

The regional firm record favours familiar foods sold to others. Of 24 alternative-protein firms in Vietnam and the region since 2019, 7 are active, 4 pivoted (all towards private label, co-manufacturing or non-food), 2 exited and 11 are dormant; none of the 4 Vietnamese meat-analogue startups is active, while all 3 makers of traditional formats are ^48,49 VN-directLow revealed. No founder has yet tried selling an ingredient to incumbents, so that route is the hypothesis to test.

#29.4 What each actor should take away

• Funders and field-builders: fund what fits your unit and your rules. Welfare funders: canteen menus with Humane World, the one implementer with Vietnam staff ⁵⁰ VN-directMedium. Market-shaping funders: procurement of a canteen protein ingredient and the naming and GMO rules. R&D funders: ingredient function and Vietnamese sensory tests. National survey budgets, not philanthropy, should carry the chay-day survey question. See International.
• Investors: treat domestic textured protein as debt-financed import substitution with a flour-price kill test, and price blends against soy. Pitch import substitution to development finance on the import bill, not on climate. See Investors.
• Startups: sell a familiar food with little capital, aim first for ingredient, private-label or co-packing supply, and bring a chay maker or distributor in as co-owner or anchor buyer. See Startups.
• Manufacturers and retailers: you already own the cheapest extender and the private-label route; test a protein upgrade through a co-packer, and design plant milks at or below 5 g sugar. See Manufacturers.
• Policy makers: use the School Health Programme to put protein quality, zinc and iron (not an animal share) into the menu software; enforce the label rules that exist; rule on plant milks and the excise before 2027. See Policy makers.
• Research bodies: measure protein quality and zinc in tofu, textured soy and blends; run an equal-saving blind test of a fungal extender against soy. See Research.

#29.5 What the desk could not answer

The questions that now decide most need a call, a quote, a trial or purchased data, not more reading. The full list is in R1. Open questions (OQ-270 onwards) and M4. Actor check and waves.

#29.6 Where we stopped

Wave 7 ran six lines and all six changed something. Wave 8 ran six and five did. Wave 9 ran four: none changed a conclusion in Part III, one corrected a wave 8 proposal, one added a constraint and two were clean negatives. Every remaining candidate was either a deeper version of a question already answered or needed a method other than desk research. The signal came, as in earlier rounds, from asking different kinds of question rather than the same question harder. Details are in M4. Actor check and waves.