D8. The demand sizing model

What this appendix contains. The method, assumptions and full results of the model behind Demand sizing. The model is tools/demand_model.py (standard-library Python). It reads data/demand_assumptions.csv and the Part IV balance model outputs (data/balance_outputs.csv, S-BASE population and meat consumption) and writes data/demand_outputs.csv and data/demand_sensitivity.csv. Running python3 tools/demand_model.py from the package root reproduces both files exactly.

#D8.1 Structure and equations

Common inputs, for each year: population P and meat consumption M (kt carcass weight) from S-BASE; meat protein MP = 0.15 x M (the balance model's factor); meat protein per person-day m = MP / P / 365.

Totals: delivered = sum of routes; displaced = sum of routes; displaced share = displaced / MP. Comparison rows: Part IV S-ALT food protein (1%, 3%, 5% and 10% of meat protein in 2030, 2035, 2040 and 2050) and S-ALT microbial feed protein. Ingredient product for R1 to R4 is delivered protein / 0.60, the protein content of textured-protein line output (50 to 70% as sold, TPP-01), which is lower than the 0.65 average of the R1 import pool because that pool includes gluten (75%) and isolates (90%); line equivalents = ingredient product / 7 kt a year per 1 t/h line. Value = delivered protein x USD 3 per kg (order of magnitude only).

What "delivered by domestic or novel protein" means. Only R1 applies an explicit domestic share. R2 to R6 count all plant or novel protein used in those routes, whatever its origin; much of it would be imported soy or pea unless a domestic maker wins the business. R7 counts protein exported. The line equivalents treat all R1 to R4 ingredient product as textured-line output, although much of the R1 pool is gluten and isolates, which need other plants. D8.3 gives the domestic reading and the soy-line count.

Requirement translation (2030 only). For each of R2 to R5, the model solves for the level that would displace S-ALT's 2030 volume on its own: hybrid adoption at 30% replacement; share of all canteen and school meal protein; regular buyers; and extra chay days per person per month across the whole population.

#D8.2 Assumptions

Sources for the assumptions are listed in the source_ids column of demand_assumptions.csv; the main ones are ^{1,2,3,4,5,6,7,8,9,10,11,12,13,14,15}.

• DMA-004 (R1 pool). Monthly trade shows no chay season in the 2024 jump in HS 3504 imports, which is priced at less than half China's average for the code. If the post-2023 Chinese increment is not food, food-grade 3504 is about 20% of the 27.0 kt, not 40%, and the pool is about 35 kt of product rather than 40 kt, the lower half of the 30 to 50 kt range ^16,17,1 VN-directThấp suy luận. The textured-type part a soy extrusion line can serve was 12,832 t in 2025 (HS 2106.10 from China, India and Serbia) ¹⁸ VN-directTrung bình bộc lộ.
• DMA-024 (chay keepers). No probability survey asks about chay, but the Diet Quality Questionnaire (Gallup, 1,007 adults, 13 November to 12 December 2021) found 7.9% (5.3 to 11.0%) ate no meat, poultry or fish on the previous day. That is a ceiling for chay-day person-days; the central 2.5% uses about a third of it ^19,20,21 VN-directTrung bình bộc lộ.
• DMA-059 (processed share). National diet surveillance counts processed-meat days (18.1% of adults on a given day) but its item omits giò and chả (pork rolls and loaves) and counts days, not grams, so it cannot test R3 ^19,20 VN-directTrung bình bộc lộ.
• DMA-060 to DMA-077 (hybrid adoption and replacement). Cost does not limit R3: soy extension saves 14 to 16% of raw-material cost at 20% replacement at every hog price since 2019 (our calculation) ^22,18,23 VN-directThấp suy luận. Adoption is limited by product identity, sensory ceilings (about 10 to 20% of the meat in fine emulsions, 30% in coarse mince, which is the evidence ceiling for DMA-077) and trust: no Vietnamese brand sells a declared blend, and buyers read hidden extension (độn, filler) as cheating ^24,25,26 VN-directTrung bình bộc lộ. D-STRETCH's price-parity condition therefore does not apply to R3; its 15% by 2035 assumes a positive blend frame not yet observed in Vietnam. Processors would mostly extend with imported soy, which enters duty-free, so R3 is plant protein of any origin ²⁷.
• DMA-083 (protein per dish). Keep 20 g. School caterers' sheets give 85 to 95 g raw meat, 67 to 70 g fish or 80 g tofu per dish, and 31.8 g of protein per lunch including rice (17.5 g animal); factory caterers' main dishes weigh 90 to 120 g ^28,29,30 VN-directTrung bình bộc lộ.
• DMA-099 to DMA-114 (schools and meal displacement). Decision 1340's animal-protein target ended in 2020. MOH Decision 3958/QĐ-BYT (25 December 2025) sets no animal share and asks for plant protein (legumes, tofu) at least twice a week; in 48 audited school-weeks only 9 met that rule, the plant slot was filled with tofu beside meat, and no dish used textured soy ^31,32,29 VN-directCao bộc lộ. School protein sold into the plant slot should count as delivered but not displaced (factor 0, like existing chay days); only blends in minced-meat and fish-cake dishes (28.5% of audited days) or added plant days displace meat. The low school adoption values stand.
• DMA-079 to DMA-098 (canteens). Five caterers' published menus lead with pork (33.1% of protein dishes), fish (25.6%) and chicken (13.4%), with tofu in 8.7% (plant-only 2.9%) and no textured soy ^{33,34,35,36,30} VN-directTrung bình được nêu. The D-BENCH canteen share implies 36.6 million meal-equivalents a year by 2035, far above what diet-change programmes deliver (D7. Global benchmarks).
• DMA-168 (export protein share). Vietnam's textured-protein exports to the EU, the UK, Japan, Korea and Australia were 0 to 73 t each in 2024 to 2025, the Vietnam-made wrapped plant foods found abroad carry 4 to 5.4 g of protein per 100 g, and the high-protein dry chay exported today is made from imported isolate, which R7 does not count as domestic protein ^37,38 VN-directThấp bộc lộ. A share of 0.05 to 0.12 is as defensible as 0.25.

#D8.3 Results by route and scenario

Protein delivered by domestic or novel protein, kt of protein a year VN-directThấp ước tính suy luận.

Meat protein displaced, kt of protein a year, and as a share of meat protein demand, against Part IV S-ALT VN-directThấp ước tính suy luận.

Context and capacity VN-directThấp ước tính suy luận.

How much is made in Vietnam, D-BENCH 2035 (our calculation from demand_outputs.csv) VN-directThấp ước tính suy luận:

So the routes create demand for about 19,000 t of protein a year by 2035, of which about 11,000 to 17,000 t would be made in Vietnam for the home market. Within R3, the domestic part at a 25% domestic share is about 0.20 of 0.79 kt. With R7 at 0.05 to 0.12, the model total would be 17.1 to 17.8 kt.

Soy lines supported. The model's 3.5 line equivalents in 2035 convert all R1 to R4 ingredient product (24.7 kt) at 7 kt per 1 t/h line. For soy extrusion only, the textured-type pool (12,832 t of product in 2025) grows to about 23,000 t by 2035 at 6% a year; a 25% domestic share is about 5,700 t, which is one 1 t/h line or two 0.5 t/h lines. The rest of the R1 pool is gluten and isolates, which need wet-milling or isolate plants (our calculation) ^1,18 VN-directThấp ước tính suy luận.

#D8.4 What S-ALT 2030 would require from each route alone

To displace S-ALT's 2030 volume (11.8 kt of meat protein) on its own VN-directThấp ước tính suy luận:

#D8.5 Sensitivity

One assumption at a time set to its low and high bound; all else at D-BENCH values; outcome is total meat protein displaced in 2035 (central value 2.35 kt) VN-directThấp ước tính.

No single assumption closes the gap

Meat protein displaced in D-BENCH 2035, kt, with one assumption at its low and high value

No single assumption moves the 2035 D-BENCH result by more than about 0.6 kt, against an S-ALT assumption of 39.3 kt for 2035. The conclusion that the food side of S-ALT is a stretch above every documented route does not depend on any one assumption.

Corrections outside the model (D-BENCH 2035; our calculation, not run through tools/demand_model.py) VN-directThấp ước tính:

None changes the headline: displacement stays at about 0.2% of meat protein, and delivered protein stays in the 17 to 19 kt range.

#D8.6 Limits

• The three most influential inputs are unmeasured in Vietnam: the share of people keeping chay days, the share of meat eaten as processed products and protein per canteen meal (Demand to frontier lists the cheap ways to measure them).
• Displacement factors come from studies outside Vietnam; Vietnamese compensation on added chay days is unknown.
• The model has no prices or elasticities inside it; scenarios encode price conditions through adoption shares. For hybrids, price is not the binding condition (D8.2 notes).
• R2 to R6 count protein of any origin, and only R1 applies a domestic share; D8.3 gives the domestic reading.
• One displacement factor (0.8) covers canteens and schools; school protein sold into the plant slot replaces tofu, not meat (D8.5).
• Funder units (D8.7) are computed outside the model and do not move with the assumptions.
• Feed demand for novel protein is not modelled here; it is in the balance model (F4. Balance model).

#D8.7 Funder units: animals spared, CO2e avoided and meals

Welfare, climate and market-shaping funders count animals, tonnes of CO2e and meals, not tonnes of meat protein. We converted the model's D-BENCH and D-STRETCH outputs for 2030, 2035 and 2050 with a standard-library script (working-papers/wave7/lines/L3-funder-units/calc_funder_units.py), which writes demand_funder_units.csv and impact_per_tonne.csv VN-directThấp ước tính suy luận.

Method.

• Species split. Added chay days (R2), canteens and schools (R4) and household analogues (R5) take the national consumption mix of the Part IV S-BASE path. Hybrids (R3) take our assumption of 75% pork, 15% poultry and 10% ruminant, informed by the Vissan label audit (pork first-listed in 50 of 80 non-chay products, beef in 18; SKU counts, not volumes) ²³.
• Animals. Displaced protein / 0.15 kg protein per kg carcass = carcass displaced; carcass / FAOSTAT carcass weight per animal for Viet Nam, 2022 to 2024. Counts are slaughter avoided per year, including animals raised abroad for imported meat; they are not welfare-weighted.
• CO2e. Gross = carcass displaced x GLEAM life-cycle intensity for East and Southeast Asia, humid zone (reference year 2005). Net = gross minus the replacing protein at 8.36 kg CO2e per kg protein (a pulses proxy). The high bound uses per-protein means from Poore and Nemecek where they exceed GLEAM's upper values.
• Meals. A meal-equivalent is one protein dish of 20 g of protein supplied by plant or novel protein instead of meat (DMA-083).
• Canteen meat mix. We read five caterers' published menus: of meat-led protein dishes, 62.6% are pork, 26.4% poultry and 11.0% beef. At that mix a tonne of protein through canteens spares about 745 animals (51 pigs, 691 birds, 4 cattle and buffalo) rather than about 1,090 at the national mix ^33,30,40 VN-adjacentThấp suy luận. We report the two as a range.

Conversion factors ^{40,41,42,43,44,45}:

Results, displacing routes R2 to R5 (our calculation) VN-directThấp ước tính suy luận:

The national-mix values and their carcass-yield ranges are in demand_funder_units.csv; the caterer-mix values replace R4's national-mix animals with 931 animals per tonne of meat protein displaced (745 / 0.8). Precision is arithmetic, not accuracy. Import substitution (R1) spares no animals and avoids about no CO2e, because soy protein replaces soy protein; its value is the import bill. On D-BENCH in 2035, by route: added chay days spare about 1.2 million birds, 49,000 pigs and 56 kt CO2e; hybrids 366,000 birds, 54,000 pigs and 45 kt; canteens and schools 0.60 to 0.89 million animals and 36 kt; household analogues about 116,000 animals and 2 kt.

Per tonne of protein delivered, 2035 (impact_per_tonne.csv; our calculation) generalThấp ước tính suy luận:

Negative values mean the delivered protein adds emissions because it removes no meat. The R3 CO2e figure rests on the 10% ruminant assumption; at 0% ruminant it would be about 28 t.

Per meal. At 20 g of protein and 0.8 net displacement, a canteen meal shifted spares about 0.015 animals at the caterer meat mix (0.010 to 0.029 across five caterers), about 0.020 at the national mix, 0.0016 for a pork dish and 0.055 for a chicken dish, against about 0.134 in ACE's estimate for Sinergia Animal's programme ⁴⁶ (D7. Global benchmarks). Upland school lunches and dinners, 84% pork-led among meat-led meals, give about 0.009.

Limits of the conversion. Everything inherits the demand model's Low confidence. Species mixes are assumptions; GLEAM intensities refer to 2005 and no Vietnamese cradle-to-carcass LCA was found ⁴⁷. The package's 0.15 kg protein per kg carcass and its poultry carcass basis could each shift head counts by 15 to 25%. Fish and eggs are not counted, so R2's animal count is a floor; fish leads 26% of canteen protein dishes. Head counts are not welfare-weighted.

Related: Demand sizing, F4. Balance model, D7. Global benchmarks.