# Stage 1 v3, part B: what the 448.5T target requires, what can be supplied, and what is missing

Written by code/report.py from the CSVs in stage1-v3/requirements (rebuilt by code/run_all.sh, which stops on any failed check). Units: bn 2026 international dollars (PPP, today's prices as the constant base) unless a row names its own unit; market US$ = PPP / layer PPP factor (B:gap_by_layer.csv). Labels: [S] sourced, [E] estimated from sourced inputs, [A] assumed. Inputs are read from the repaired v2 files (prefix R:) and the v1 bridge (B:); nothing in stage1-v1, stage1-v2 or stage1-v3/repair was changed (check 16). No companies are proposed.

## Headline

| level | base mix | alternative mix (homothetic) | v2 | row |
|---|---|---|---|---|
| Target: 2 x 2026 | 448,524 [E] | 448,524 | 448,524 | totals_v3.csv TT01 |
| On course 2035, evidence-backed | 275,705 [E] | 275,705 | 265,864 | totals_v3.csv TT02 |
| On course, unbacked (counted in missing) | 2,307 [A] | 2,307 | 2,307 | totals_v3.csv TT03 |
| On-course shortfall (carried, never floored) | 9,957 [A] | 9,957 | 3,021 | totals_v3.csv TT04 |
| - of it closable at precedent (in tier 1) | 9,957 [A] | 9,957 |  | totals_v3.csv TT04a |
| - of it still claimed by an input short at precedent (in tier 2) | 0 [A] | 0 |  | totals_v3.csv TT04b |
| On course net = evidence-backed - shortfall | 265,749 [A] | 265,749 | 265,864 | totals_v3.csv TT05 |
| Missing = target - on course net | 182,775 [A] | 182,775 | 182,659 | totals_v3.csv TT08 |
| **Tier 1**: missing, producible at the fastest precedent | 34,757 [A] | 34,757 | 75,439 | totals_v3.csv TT09 |
| **Tier 2**: missing beyond any precedent | 148,148 [A] | 148,154 | 110,241 | totals_v3.csv TT10 |
| Surplus: maximum at precedent beyond the target (apart) | 130 [A] | 137 |  | totals_v3.csv TT11 |
| Joint maximum (v2: capacity maximum) | 290,550 [A] | 290,550 | 338,283 | totals_v3.csv TT12 |
| Maximum at the fastest precedent = joint maximum + covered shortfall | 300,506 [A] | 300,506 |  | totals_v3.csv TT12a |

- **Counted once, the target splits into on course 265,619, tier 1 34,757 and tier 2 148,148 bn** (identity row TT13 = 0). Both tiers are missing and both go forward. The only on-course output beyond its target is U2 (net purchases abroad, a negative accounting row): 129.9 bn, reported inside the surplus and never summed.
- **Tier 1 fell from 75,439 (v2; the user's 75.5T) to 34,757 bn and tier 2 rose from 110,241 (110.2T) to 148,148 bn.** The repaired joint maximum is 290,550 bn; adding back the on-course shortfall that is closable at precedent gives 300,506 bn, and every layer's target exceeds its maximum at the fastest precedent in both mixes except U2, so tier 1 is each layer's whole room above on course: unbacked on course + further output - the further output the human workforce blocks + the on-course shortfall closable at precedent.
- **Lead decision (2026-11-15), now the base:** the on-course shortfall (9,957 bn) stays carried in as missing, but its tier follows its inputs. Every input that claims it (copper, lithium, saving, construction trades) is covered at its fastest precedent (R coverage_total_at_precedent >= 1, shortfall_tier.csv), so 9,957 bn is tier 1 and 0 bn tier 2. The rule is general: an input below 1 at precedent keeps its claim in tier 2 (check 18 runs a variant with saving at 0.99, which moves 7,282 bn back to tier 2). The previous rule (whole shortfall in tier 2: tier 1 24,801, tier 2 158,104) is kept as error test E4.
- **The mix moves tier 2 between layers, not its total:** the homothetic mix gives tier 1 34,757 and tier 2 148,154 bn. It does change what the target requires (food land, construction, health), section 5.
- **By input, the binding requirement is physical work.** At the best achievable productivity the base target needs 5,903 million worker-equivalents of work; humans at their record employment ratio supply 4,317 million (73%), so 1,586 million must be machine work beyond the on-course fleet. Physical machine labour (K04) needs 1,202 million worker-equivalents against 42.9 million at its fastest precedent: tier 2 of 1,159 million [A], on which 85,376 bn of output depends.
- Other tier-2 inputs at the best achievable ratios: K06 copper 5.6 Mt; K12 building permission 1,127 bn int$ of C13; K14 transformers 1,004 TWh; K15 clinical approval 18,270 bn int$ of C05; K16 machine work in licensed tasks 157 million worker-equivalents; K17 construction trades 75.9 million workers; K21 capital stock 392 tn int$; K22 cropland 592 million ha.
- Tier 1 (proven possible at the fastest precedent, nobody on course to do it): K01 electricity; K02 fuels; K03 machine cognition; K05 AI compute; K07 lithium; K08 rare earths; K09 steel; K10 cement; K11 saving; K13 grid connection; K18 human workers; K20 aluminium. No input is covered by its on-course supply alone. Electricity, for example, needs 74,059 TWh against 44,274 on course.

## 1. The target economy by layer

**This mix is a choice.** Nothing observed today fixes how a 448.5T economy would be composed. The base mix is the v1 income-elasticity split (B:code/bridge.py): the IMF baseline at 2026 layer shares plus the gap to the target spread as share x ((1 + gap/baseline)^eta - 1), with world spend-weighted income elasticities eta (USDA TB-1929; eta = 1 outside household demand). It is applied to the repaired 2026 layer levels (R:baseline/layers.csv, PPP shares) and rescaled to exactly 448,524 bn. The alternative is homothetic: every layer keeps its 2026 share, i.e. doubles.

- Why not the v1 numbers as published: v1 split on share_base (the mean of market and PPP shares), while the repaired 2026 levels the supply side uses are PPP shares. Applied to those levels, the published split makes food at home (C01, eta 0.62) grow 2.12x and built structures (C13, eta 1.00) 1.54x, the reverse of the elasticities: the rank correlation of eta with implied growth is 0.47, against 0.998 for the base mix (mix_choice.csv, check 17). The published split sums to 448,517.9 bn before rescaling. It is kept in target_economy.csv (target_v1_published_ppp_bn) and tested as error E1.

| layer | capability | eta | 2026 | target base | x | %/yr | target alt | v1 published | target base, market US$ |
|---|---|---|---|---|---|---|---|---|---|
| C01 | Nutrition (food and non-alcoholic beverages at home) | 0.62 | 12,590 | 21,196 | 1.68 | 6.0% | 25,180 | 26,745 | 16,453 |
| C02 | Clothing and footwear | 0.97 | 3,389 | 6,517 | 1.92 | 7.5% | 6,777 | 7,969 | 4,800 |
| C03 | Shelter and household utilities (actual and imputed  | 1.06 | 29,600 | 59,177 | 2.00 | 8.0% | 59,201 | 54,054 | 24,938 |
| C04 | Household equipment and upkeep | 1.05 | 3,842 | 7,634 | 1.99 | 7.9% | 7,683 | 9,734 | 6,033 |
| C05 | Health care (household and government) | 1.28 | 23,473 | 51,081 | 2.18 | 9.0% | 46,947 | 47,337 | 22,221 |
| C06 | Mobility (vehicles, fuel, transport services) | 1.15 | 8,609 | 17,805 | 2.07 | 8.4% | 17,219 | 22,525 | 13,886 |
| C07 | Communication | 1.15 | 4,288 | 8,879 | 2.07 | 8.4% | 8,576 | 7,953 | 3,584 |
| C08 | Recreation and culture (household and government) | 1.35 | 5,961 | 13,340 | 2.24 | 9.4% | 11,922 | 17,004 | 10,535 |
| C09 | Education (household and government) | 0.92 | 16,725 | 31,579 | 1.89 | 7.3% | 33,451 | 26,787 | 11,212 |
| C10 | Food service and lodging | 1.31 | 6,028 | 13,288 | 2.20 | 9.2% | 12,056 | 14,945 | 8,464 |
| C11 | Personal, financial and social services (miscellaneo | 1.31 | 15,867 | 34,976 | 2.20 | 9.2% | 31,733 | 39,056 | 21,985 |
| C12 | Collective public services (administration, defence, | 1.00 | 17,486 | 34,091 | 1.95 | 7.7% | 34,971 | 34,955 | 18,251 |
| C13 | Built structures (construction investment) | 1.00 | 54,229 | 105,728 | 1.95 | 7.7% | 108,458 | 83,336 | 31,062 |
| C14 | Productive equipment (machinery and equipment invest | 1.00 | 7,051 | 13,747 | 1.95 | 7.7% | 14,102 | 21,527 | 14,931 |
| C15 | Knowledge and software (other GFCF: R&D, software, d | 1.00 | 7,208 | 14,053 | 1.95 | 7.7% | 14,416 | 19,167 | 12,375 |
| U1 | NOT COVERED: alcoholic beverages, tobacco, narcotics | 1.00 | 3,595 | 7,009 | 1.95 | 7.7% | 7,190 | 7,007 | 4,001 |
| U2 | NOT COVERED: net purchases abroad (AIC minus its com | 1.00 | -137 | -267 |  |  | -274 | -266 | -152 |
| U3 | NOT COVERED: changes in inventories and valuables | 1.00 | 2,258 | 4,403 | 1.95 | 7.7% | 4,517 | 4,402 | 2,513 |
| U4 | NOT COVERED: balance of exports and imports | 1.00 | 2,200 | 4,289 | 1.95 | 7.7% | 4,400 | 4,288 | 2,448 |
| total | | | 224,262 | 448,524 | | | 448,524 | 448,524 | 229,540 |

## 2. Requirements: what the target needs at the best achievable 2035 ratios

Each layer's target output x the input it needs per unit of output (requirements.csv, one row per layer and input; ratio, basis and source on every row, check 10). Input totals are sums of the layer rows (check 11). "Best achievable" is read as the sourced 2035 pathway with the lowest energy per unit of output; where only today's or the baseline ratio is sourced the row says so.

| input | unit | required, base mix | required, alt mix | at the baseline / other ratio | ratio used |
|---|---|---|---|---|---|
| K01 Electricity (generation and delivery) | TWh per year (gross generation) | 74,059.0 [A] | 74,488.0 | 69,351.2 | best achievable 2035 (lowest final-energy path, IEA NZE) + machine lab |
| K02 Fuels for final use (oil, gas, coal, non-electri | EJ per year (final consumption excl. ele | 445.2 [E] | 447.4 | 572.5 | best achievable 2035 (IEA NZE) |
| K03 Machine cognition: cognitive work by AI agents | million worker-equivalents in 2035 | 537.1 [A] | 538.3 |  | best achievable 2035 (fastest sourced productivity) |
| K04 Physical machine labour: robots and autonomous m | million worker-equivalents in 2035 | 1,202.3 [A] | 1,238.1 |  | best achievable 2035 (fastest sourced productivity) |
| K05 AI compute: accelerators in service (chip power) | GW of AI chip TDP in service in 2035 | 146.3 [A] | 146.6 |  | best achievable 2035 (fastest sourced productivity) |
| K06 Copper | million t refined copper per year | 51.0 [E] | 51.8 | 50.2 | pathway-consistent 2035 (GCMO High Demand, closest to NZE); more miner |
| K07 Lithium | kt lithium content per year | 1,208.3 [E] | 1,221.1 | 1,014.0 | pathway-consistent 2035 (GCMO High Demand, closest to NZE); more miner |
| K08 Rare earths (magnet REE: Nd, Pr, Dy, Tb) | kt magnet rare earth elements per year | 213.0 [E] | 215.4 | 193.9 | pathway-consistent 2035 (GCMO High Demand, closest to NZE); more miner |
| K09 Steel | Mt crude steel per year | 2,901.9 [E] | 2,950.7 | 3,076.5 | best achievable 2035 (IEA NZE material efficiency) |
| K10 Cement | Mt cement per year | 5,589.7 [E] | 5,708.4 | 6,095.5 | best achievable 2035 (IEA NZE) |
| K11 Capital: saving to finance investment | bn US$ (market) of gross saving per year | 58,368.1 [E] | 59,875.2 | 58,368.1 | identity (investment = saving at market prices); no ratio to improve |
| K12 Land-use and building permission | bn 2026 int$ of construction (C13) that  | 105,727.5 [E] | 108,457.5 | 105,727.5 | identity |
| K13 Grid interconnection approval | TWh per year of generation connected | 74,059.0 [A] | 74,488.0 | 69,351.2 | best achievable 2035 (lowest final-energy path, IEA NZE) + machine lab |
| K14 Power transformers and grid equipment | TWh per year of generation the delivered | 74,059.0 [A] | 74,488.0 | 69,351.2 | best achievable 2035 (lowest final-energy path, IEA NZE) + machine lab |
| K15 Clinical evidence and approval | bn 2026 int$ of health output (C05) that | 51,080.9 [E] | 46,947.0 | 51,080.9 | identity |
| K16 Professional licensing (machine work in licensed | million machine worker-equivalents in 20 | 1,739.4 [A] | 1,776.4 |  | best achievable 2035 (fastest sourced productivity) |
| K17 Skilled construction and electrical trades (huma | million construction workers (ISIC F) in | 571.0 [E] | 584.3 | 602.7 | best achievable 2035 (fastest sourced productivity) |
| K18 Human workforce (all sectors) | million workers in 2035 | 4,316.6 [E] | 4,316.6 |  | best achievable 2035 (fastest sourced productivity) |
| K19 Output needing a permission current rules do not | bn 2026 int$ of output in 2035 | 53,333.7 [A] | 53,274.4 |  | today (only today is sourced) |
| K20 Aluminium | Mt aluminium per year | 209.3 [E] | 212.0 | 208.1 | best achievable 2035 (IEA NZE) |
| K21 Capital stock (fixed assets in use) | tn 2026 int$ (PPP) of capital stock in 2 | 2,140.1 [A] | 2,144.4 | 1,538.8 | today's ratio held (only today's is sourced) |
| K22 Cropland | million ha of cropland in 2035 | 2,280.8 [A] | 2,485.2 | 2,610.7 | best achievable 2035 (fastest sourced yield growth) |

- **Energy.** Electricity and fuels follow the IEA NZE 2035 path (WEO 2025 Tables A.2c, A.3c, A.5c): per unit of GDP, 1.07x the STEPS electricity and 0.78x the STEPS non-electric fuel, applied to Agent 4's EXIOBASE layer footprints. Machine labour adds its own electricity: 0.98 TWh a year per million physical worker-equivalents (Figure 03: 2.3 kWh for 5 hours, x 2,128 hours) and 3.44 TWh per million AI-agent worker-equivalents (chip GW per worker-equivalent x PUE 1.44 x load 1.0 [A]).
- **Materials.** Steel (0.94x STEPS), cement and aluminium follow NZE material efficiency. GCMO 2026 has no NZE case, so copper, lithium and rare earths take its High Demand path, the one that goes with faster electrification (1.02x, 1.19x and 1.10x STEPS per unit of GDP): the lowest-energy path needs more minerals per unit of output, not fewer. Aluminium (not in Agent 4) uses EXIOBASE aluminium products calibrated to the IEA's 2026 world production.
- **Labour.** Work per unit of output falls at each producing sector's fastest 9-year output-per-worker growth (BL D_PRODMAX, UN AMA / ILO). Total work: 5,903 million worker-equivalents (alternative mix 5,940). Humans are counted up to their fastest-precedent supply (4,317 million, the record employment ratio); the rest is required of machines, split cognitive / physical by Agent 4's high-skill shares. Counted this way, human, cognitive-machine and physical-machine work add up without overlap.
- **Capital.** Capital stock at today's world capital-output ratio (PWT 11.0, 4.72 in 2023), split by layer with EXIOBASE capital consumption [A]. No best-achievable 2035 ratio is sourced; at the lowest world ratio since 1990 (3.39) the target needs 1,539 tn instead of 2,140.
- **Land and permission.** Cropland from EXIOBASE crop footprints and FAOSTAT cropland, with crop output per hectare at its fastest 9-year growth (2.8% a year) [A: crop tonnes per unit of real output otherwise held]. 53,334 bn of the target above on course falls in licensed work or under discretionary building permission (K19, layers_v3.csv perm_share).

## 3. Supply

**By layer** (R:baseline/layers.csv, repaired): evidence-backed on course, unbacked on course, the on-course shortfall (carried in), and the joint maximum at the fastest precedent that respects the binding inputs.

| layer | on course, evidence | unbacked | on-course shortfall | on course net | joint maximum | growth needed %/yr | growth at joint max %/yr | first binding inputs (A4 rerun) |
|---|---|---|---|---|---|---|---|---|
| C01 | 14,261 | 0 | 205 | 14,056 | 15,089 | 6.0% | 2.0% | K04(0.138) > K18(0.922) > K16(0.940) |
| C02 | 4,446 | 0 | 64 | 4,382 | 4,766 | 7.5% | 3.9% | K04(0.138) > K18(0.922) > K16(0.961) |
| C03 | 34,578 | 0 | 497 | 34,081 | 34,211 | 8.0% | 1.6% | K04(0.138) > K16(0.892) > K18(0.922) |
| C04 | 4,942 | 0 | 71 | 4,871 | 6,259 | 7.9% | 5.6% | K04(0.138) > K18(0.922) > K16(0.961) |
| C05 | 30,435 | 0 | 437 | 29,998 | 31,327 | 9.0% | 3.3% | K04(0.138) > K16(0.790) > K18(0.922) |
| C06 | 9,623 | 511 | 146 | 9,477 | 14,677 | 8.4% | 6.1% | K04(0.138) > K16(0.908) > K18(0.922) |
| C07 | 5,363 | 0 | 77 | 5,286 | 5,291 | 8.4% | 2.4% | K04(0.138) > K18(0.922) > K16(0.968) |
| C08 | 6,381 | 1,196 | 109 | 6,272 | 7,729 | 9.4% | 2.9% | K04(0.138) > K18(0.922) > K16(0.958) |
| C09 | 20,178 | 0 | 290 | 19,888 | 22,048 | 7.3% | 3.1% | K04(0.138) > K16(0.790) > K18(0.922) |
| C10 | 7,743 | 0 | 111 | 7,632 | 8,955 | 9.2% | 4.5% | K04(0.138) > K18(0.922) > K16(0.958) |
| C11 | 19,514 | 0 | 280 | 19,234 | 20,836 | 9.2% | 3.1% | K04(0.138) > K16(0.875) > K18(0.922) |
| C12 | 20,220 | 544 | 298 | 19,921 | 22,621 | 7.7% | 2.9% | K04(0.138) > K16(0.865) > K18(0.922) |
| C13 | 67,924 | 56 | 5,513 | 62,411 | 65,575 | 7.7% | 2.1% | K04(0.138) > K16(0.908) > K18(0.922) |
| C14 | 9,674 | 0 | 784 | 8,889 | 11,458 | 7.7% | 5.5% | K04(0.138) > K18(0.922) > K16(0.961) |
| C15 | 12,134 | 0 | 984 | 11,150 | 11,507 | 7.7% | 5.3% | K04(0.138) > K16(0.916) > K18(0.922) |
| U1 | 3,969 | 0 | 57 | 3,912 | 3,912 | 7.7% | 0.9% | K04(0.138) > K18(0.922) > K16(0.961) |
| U2 | -137 | 0 | 0 | -137 | -137 |  |  | none: accounting row with no production, so no input need |
| U3 | 2,258 | 0 | 32 | 2,226 | 2,226 | 7.7% | -0.2% | K04(0.138) > K16(0.892) > K18(0.922) |
| U4 | 2,200 | 0 | 0 | 2,200 | 2,200 | 7.7% | 0.0% | none: accounting row with no production, so no input need |

**By input** (R:agent4_bottlenecks/bottlenecks.csv, repaired Agent 4, except where noted): on-course supply 2035 and supply at the fastest sourced precedent.

| input | on course 2035 | fastest precedent 2035 | precedent |
|---|---|---|---|
| K01 Electricity (generation and delivery) | 44,274.0 [S] | 93,251.9 [E] | China 2002-2011: generation x2.84 in 9 years (Ember), applied to world 2026 supply; world fastest 9 years (OWID, 2001-10) x1.37 is the low case |
| K02 Fuels for final use (oil, gas, coal, non | 374.0 [E] | 548.8 [E] | World fossil-fuel consumption 1965-1974 x1.52 in 9 years (EI via OWID), the post-1965 maximum |
| K03 Machine cognition: cognitive work by AI  | 310.3 [A] | 45,732.6 [A] | AI chip power sold grew x1.87 in 2025 (Epoch), held 2028-2035 on top of the through-2027 fleet |
| K04 Physical machine labour: robots and auto | 32.8 [A] | 42.9 [A] | China electric-car sales 2014-2023 x111 in 9 years (IEA via OWID) applied to humanoid shipments from 50k in 2026 (Counterpoint); industrial robots on  |
| K05 AI compute: accelerators in service (chi | 84.5 [A] | 12,455.7 [A] | AI chip power sold x1.87 in 2025 (Epoch) held 2028-2035; bounded in practice by K01/K14 (power and grid) |
| K06 Copper | 26.5 [A] | 45.4 [E] | World copper mine output 1954-1963 x1.625 in 9 years (USGS via OWID), post-1950 maximum; post-1985 maximum (1995-2004) x1.47 is the low case |
| K07 Lithium | 452.9 [A] | 2,301.2 [E] | World lithium mine output 2016-2025 x7.21 in 9 years (USGS via OWID) |
| K08 Rare earths (magnet REE: Nd, Pr, Dy, Tb) | 136.4 [A] | 709.6 [E] | World rare-earth mine output 1960-1969 x7.97 in 9 years (USGS via OWID) |
| K09 Steel | 2,570.4 [E] | 3,135.5 [E] | World crude steel 2002-2011 x1.68 in 9 years (worldsteel Steel Statistical Yearbook 2012) |
| K10 Cement | 5,428.6 [A] | 7,528.7 [E] | World cement 2002-2011 x2.00 in 9 years (USGS MCS 2004 and 2013) |
| K11 Capital: saving to finance investment | 37,826.3 [A] | 87,530.7 [A] | China gross saving 51.3% of GDP (2008, World Bank) applied to world output at max; on course = world 2015-24 mean saving rate |
| K12 Land-use and building permission | 94,403.4 [A] | 104,600.5 [A] | Auckland 2016 upzoning (+50% consents within 5 years, A4:A05) in the 30% of construction under discretionary permission (A4:A04); outside it current r |
| K13 Grid interconnection approval | 44,274.0 [S] | 87,637.2 [A] | US 2008 queue speed (22 months request-to-operation, LBNL) in queue-bound grids (US+EU share of 2025 generation); STEPS path assumed connected |
| K14 Power transformers and grid equipment | 42,553.3 [A] | 73,055.4 [A] | US generator step-up transformer demand x3.74 in 6 years (2019-25, Wood Mackenzie via pv magazine) used as the supply ramp from the STEPS addition rat |
| K15 Clinical evidence and approval | 30,435.4 [E] | 32,810.9 [E] | Further health output is workforce (2010-19) and productivity (2015-24) precedents achieved under current approval rules (BL C05-F1, C05-F2); new prod |
| K16 Professional licensing (machine work in  | 1,425.5 [A] | 1,582.4 [A] | machine work allowed outside licensed tasks now; fastest reform precedent halves the licensed share (A4:A11) |
| K17 Skilled construction and electrical trad | 380.2 [E] | 495.1 [E] | World construction employment at its fastest 9-year growth (ILO, 2002-2011, 5.36%/yr) from 2026 |
| K18 Human workforce (all sectors) | 4,023.2 [E] | 4,316.6 [E] | WPP 15+ population 2035 x the highest world employment ratio 1991-2025 (BL T19) |
| K19 | - | - | covered by K12, K15 and K16 (supply of permissions) |
| K20 Aluminium | 137.0 [S] | 330.5 [E] | world bauxite mine output 1950-1959 x2.82 in 9 years (USGS via OWID), post-1950 maximum; post-1985 maximum 2004-2013 x1.80 is the low case |
| K21 Capital stock (fixed assets in use) | 1,502.1 [E] | 1,748.4 [E] | world real capital stock at its fastest 9-year growth (1964-1973, PWT 11.0); on course = 2013-23 trend |
| K22 Cropland | 1,628.0 [E] | 1,688.9 [E] | world cropland at its fastest 9-year expansion (1976-1985, FAOSTAT); on course = 2014-24 trend |

- Added here: aluminium (K20; IEA STEPS 137 Mt on course; bauxite's fastest 9-year growth after 1950, x2.82; after 1985 x1.80 gives 211 Mt, just above the 209 Mt needed), capital stock (K21; PWT real capital growth on its 2013-23 trend, 3.8%, and at its fastest 9 years, 5.6%) and cropland (K22; FAOSTAT trend and fastest expansion, 0.64% a year).
- Recomputed against the target, not copied: K12 building permission and K16 licensing. Agent 4 measured them against on course + further output, so its K12 supply capped the 70% of construction outside discretionary permission at capacity, and its K16 on-course supply covered only on-course machine work. Against the target, current rules permit any amount outside the discretionary or licensed share; only that share is missing (tier 1 = the reform precedent, tier 2 = the rest).
- Not counted in supply: the three capabilities R credited outside the baseline (53.3 bn, UNTESTED; R:caps_outside_baseline.csv). They are candidate closers for part C.

## 4. Missing, counted once

**By layer.** missing = target - (evidence-backed on course - on-course shortfall). Tier 1 = the part the layer can produce at the fastest precedent: within its joint maximum plus the part of its on-course shortfall whose inputs are all covered at precedent (layers_v3.csv tier1_room); tier 2 = the rest. Surplus = the maximum at precedent beyond the target, reported apart and never counted toward it.

| layer | target | on course net | missing | tier 1 | tier 2 | surplus | tier 1, market US$ | tier 2, market US$ | output needing a new permission |
|---|---|---|---|---|---|---|---|---|---|
| C01 | 21,196 | 14,056 | 7,140 | 1,238 | 5,902 | 0 | 961 | 4,581 | 832 |
| C02 | 6,517 | 4,382 | 2,135 | 447 | 1,687 | 0 | 330 | 1,243 | 164 |
| C03 | 59,177 | 34,081 | 25,097 | 627 | 24,469 | 0 | 264 | 10,312 | 11,099 |
| C04 | 7,634 | 4,871 | 2,763 | 1,459 | 1,304 | 0 | 1,153 | 1,030 | 213 |
| C05 | 51,081 | 29,998 | 21,083 | 1,767 | 19,316 | 0 | 769 | 8,403 | 8,671 |
| C06 | 17,805 | 9,477 | 8,328 | 5,345 | 2,982 | 0 | 4,169 | 2,326 | 1,404 |
| C07 | 8,879 | 5,286 | 3,593 | 83 | 3,511 | 0 | 33 | 1,417 | 225 |
| C08 | 13,340 | 6,272 | 7,068 | 1,566 | 5,502 | 0 | 1,237 | 4,346 | 484 |
| C09 | 31,579 | 19,888 | 11,690 | 2,450 | 9,241 | 0 | 870 | 3,281 | 4,788 |
| C10 | 13,288 | 7,632 | 5,657 | 1,434 | 4,223 | 0 | 913 | 2,689 | 466 |
| C11 | 34,976 | 19,234 | 15,742 | 1,883 | 13,860 | 0 | 1,183 | 8,712 | 3,858 |
| C12 | 34,091 | 19,921 | 14,170 | 2,998 | 11,172 | 0 | 1,605 | 5,981 | 3,599 |
| C13 | 105,728 | 62,411 | 43,316 | 8,677 | 34,639 | 0 | 2,549 | 10,177 | 16,186 |
| C14 | 13,747 | 8,889 | 4,858 | 3,353 | 1,504 | 0 | 3,642 | 1,634 | 322 |
| C15 | 14,053 | 11,150 | 2,903 | 1,341 | 1,561 | 0 | 1,181 | 1,375 | 320 |
| U1 | 7,009 | 3,912 | 3,097 | 57 | 3,040 | 0 | 33 | 1,735 | 240 |
| U2 | -267 | -137 | -130 | 0 | 0 | 130 | 0 | 0 | 0 |
| U3 | 4,403 | 2,226 | 2,177 | 32 | 2,145 | 0 | 19 | 1,224 | 463 |
| U4 | 4,289 | 2,200 | 2,089 | 0 | 2,089 | 0 | 0 | 1,193 | 0 |
| **total** | 448,524 | 265,749 | 182,775 | 34,757 | 148,148 | 130 | 20,910 | 71,658 | 53,334 |

- Tier 2 sits in the 15 capability layers (140,874 bn) and in the accounting rows U1-U4 (7,274 bn: alcohol and tobacco, inventories, net exports). U3 and U4 have no production of their own and no input need; they grow with the target as accounting remainders. Part C should report them, not name capabilities for them.
- The largest tier-2 layers: C13 34,639, C03 24,469, C05 19,316, C11 13,860, C12 11,172, C09 9,241 bn.

**By input** (physical units; inputs_v3.csv). missing = required - on-course supply (signed); tier 1 = the part the fastest precedent covers; tier 2 = beyond it; surplus = precedent supply beyond the requirement. Output depending = the output (PPP) that cannot be produced without the missing input at the stated ratio, by layer in input_layer_v3.csv. **It is non-additive across inputs: the same output depends on several inputs at once.** Fixed coefficients except copper, lithium and rare earths (CES 0.4, Agent 4); fixed coefficients are an upper bound.

| input | unit | required | on course | missing | tier 1 | tier 2 | surplus at precedent | status | output depending | of which on tier 2 |
|---|---|---|---|---|---|---|---|---|---|---|
| K01 Electricity (generation and delive | TWh per year (gross generation | 74,059.0 | 44,274.0 | 29,785.0 | 29,785.0 | 0.0 | 19,192.9 | tier 1 (closable at precedent) | 178,769 | 0 |
| K02 Fuels for final use (oil, gas, coa | EJ per year (final consumption | 445.2 | 374.0 | 71.2 | 71.2 | 0.0 | 103.6 | tier 1 (closable at precedent) | 71,078 | 0 |
| K03 Machine cognition: cognitive work  | million worker-equivalents in  | 537.1 | 310.3 | 226.8 | 226.8 | 0.0 | 45,195.5 | tier 1 (closable at precedent) | 18,168 | 0 |
| K04 Physical machine labour: robots an | million worker-equivalents in  | 1,202.3 | 32.8 | 1,169.5 | 10.1 | 1,159.5 | 0.0 | tier 2 (beyond precedent) | 86,118 | 85,376 |
| K05 AI compute: accelerators in servic | GW of AI chip TDP in service i | 146.3 | 84.5 | 61.8 | 61.8 | 0.0 | 12,309.4 | tier 1 (closable at precedent) | 18,168 | 0 |
| K06 Copper | million t refined copper per y | 51.0 | 26.5 | 24.5 | 18.9 | 5.6 | 0.0 | tier 2 (beyond precedent) | 1,783 | 206 |
| K07 Lithium | kt lithium content per year | 1,208.3 | 452.9 | 755.4 | 755.4 | 0.0 | 1,093.0 | tier 1 (closable at precedent) | 385 | 0 |
| K08 Rare earths (magnet REE: Nd, Pr, D | kt magnet rare earth elements  | 213.0 | 136.4 | 76.7 | 76.7 | 0.0 | 496.6 | tier 1 (closable at precedent) | 21 | 0 |
| K09 Steel | Mt crude steel per year | 2,901.9 | 2,570.4 | 331.5 | 331.5 | 0.0 | 233.6 | tier 1 (closable at precedent) | 50,782 | 0 |
| K10 Cement | Mt cement per year | 5,589.7 | 5,428.6 | 161.1 | 161.1 | 0.0 | 1,939.0 | tier 1 (closable at precedent) | 12,814 | 0 |
| K11 Capital: saving to finance investm | bn US$ (market) of gross savin | 58,368.1 | 37,826.3 | 20,541.8 | 20,541.8 | 0.0 | 29,162.6 | tier 1 (closable at precedent) | 46,993 | 0 |
| K12 Land-use and building permission | bn 2026 int$ of construction ( | 105,727.5 | 94,403.4 | 11,324.1 | 10,197.1 | 1,127.0 | 0.0 | tier 2 (beyond precedent) | 11,324 | 1,127 |
| K13 Grid interconnection approval | TWh per year of generation con | 74,059.0 | 44,274.0 | 29,785.0 | 29,785.0 | 0.0 | 13,578.2 | tier 1 (closable at precedent) | 178,769 | 0 |
| K14 Power transformers and grid equipm | TWh per year of generation the | 74,059.0 | 42,553.3 | 31,505.7 | 30,502.1 | 1,003.6 | 0.0 | tier 2 (beyond precedent) | 189,097 | 6,024 |
| K15 Clinical evidence and approval | bn 2026 int$ of health output  | 51,080.9 | 30,435.4 | 20,645.5 | 2,375.5 | 18,270.0 | 0.0 | tier 2 (beyond precedent) | 20,646 | 18,270 |
| K16 Professional licensing (machine wo | million machine worker-equival | 1,739.4 | 1,425.5 | 313.9 | 156.9 | 156.9 | 0.0 | tier 2 (beyond precedent) | 23,738 | 11,869 |
| K17 Skilled construction and electrica | million construction workers ( | 571.0 | 380.2 | 190.8 | 114.9 | 75.9 | 0.0 | tier 2 (beyond precedent) | 148,554 | 59,095 |
| K18 Human workforce (all sectors) | million workers in 2035 | 4,316.6 | 4,023.2 | 293.4 | 293.4 | 0.0 | 0.0 | tier 1 (closable at precedent) | 22,091 | 0 |
| K19 | bn 2026 int$ of output in 2035 | 53,333.7 | - | - | - | - | - | information (supply in K12, K15, K16) | - | - |
| K20 Aluminium | Mt aluminium per year | 209.3 | 137.0 | 72.3 | 72.3 | 0.0 | 121.2 | tier 1 (closable at precedent) | 153,577 | 0 |
| K21 Capital stock (fixed assets in use | tn 2026 int$ (PPP) of capital  | 2,140.1 | 1,502.1 | 638.0 | 246.3 | 391.7 | 0.0 | tier 2 (beyond precedent) | 132,512 | 81,357 |
| K22 Cropland | million ha of cropland in 2035 | 2,280.8 | 1,628.0 | 652.7 | 60.8 | 591.9 | 0.0 | tier 2 (beyond precedent) | 127,212 | 115,357 |

- Tier-2 inputs ranked by the output that depends on their tier 2 (non-additive): K22 115,357; K04 85,376; K21 81,357; K17 59,095; K15 18,270; K16 11,869; K14 6,024; K12 1,127; K06 206 bn.
- Electricity: the target needs 74,059 TWh, 29,785 more than on course; China's 2002-11 build rate applied to the world covers it (tier 1). Grid connection follows; transformers fall short by 1,004 TWh at their precedent.
- Lithium (755 kt missing, tier 1) and rare earths (tier 1) rest on their mine ramps (lithium 2016-25, rare earths 1960-69); with substitution little output depends on them (385 and 21 bn); with fixed coefficients see error E5.
- Output depending on a tier-1 input is output that needs the input built at its fastest precedent; on a tier-2 input, output that needs more than any precedent supplies.
- In the requirements table, the "baseline / other ratio" column is the IEA STEPS path for energy and materials, the trend for construction trades, the lowest world capital-output ratio 1990-2023 for capital stock (low case) and the 2014-24 yield trend for cropland (requirements.csv ratio_alt_basis).

## 5. What changes under the alternative mix

- Totals: tier 1 34,757 (base 34,757), tier 2 148,154 (base 148,148). Since every layer's target exceeds its maximum at the fastest precedent in both mixes except U2, the mix only moves tier 2 between layers.
- Layers whose tier 2 moves most (alt - base, bn): C05 -4,134, C01 +3,985, C11 -3,243, C13 +2,730, C09 +1,872, C08 -1,418.
- Inputs whose tier 2 moves (alt - base, own units): K04 +35.8, K06 +0.8, K12 +819.0, K14 +429.0, K15 -4,134.0, K16 +2.0, K17 +13.3, K21 +4.4, K22 +204.4.
- Food is the clearest case: doubling food at home (C01) instead of the income-elastic 1.68x needs 2,485 million ha of cropland instead of 2,281, against 1,689 at the fastest expansion on record.

## 6. Comparison with v2

| change | tier 1 | tier 2 | row |
|---|---|---|---|
| v2 | 75,439 | 110,241 | TT09, TT10 (v2 column) |
| further output at precedent: construction measured by real value, its fastest precedent in the same unit (R repair 1) | -27,748 | 17,907 | DV01, DV06 |
| further output the human workforce blocks (R repair 2) | -19,869 | 19,869 | DV02, DV07 |
| on-course shortfall: v2 put 3,021 bn in tier 1; v3 puts its 9,957 bn closable at precedent in tier 1 and 0 bn in tier 2 | 6,936 | 0 | DV03, DV08 |
| counting by layer (a layer's surplus cannot fill another layer) | 0 | 130 | DV09 |
| v3 | 34,757 | 148,148 | TT09, TT10 |

v2's tier 1 (75,439, the user's 75.5T) was the pool to explain less the part beyond the capacity maximum: further output + unbacked on course + v2's baseline shortfalls. v2's tier 2 (110,241, 110.2T) was the target less the capacity maximum, reported as one unnamed block. In v3 tier 2 is split by layer and by input, and it is larger by 37,906 bn because the maximum now respects the binding inputs and measures construction the way real GDP does.

## 7. The five likeliest errors, tested in code

Verdict rule (code/errors5.py, checked by check 14): material if the test moves total tier 1 or tier 2 by more than 1% of target A, moves any layer's tier by more than that, changes an input's status, moves any input's output-depending figure by more than 1% of target A, or changes the top five inputs ranked by output depending.

| id | likely error | test | tier 1 | tier 2 | largest layer shift | status changes | verdict |
|---|---|---|---|---|---|---|---|
| E1 | The target mix uses the repaired 2026 PPP layer shares, while v1 published its split on share_base (mean of market and PPP shares); the published spli | rerun with the v1 split as published (share_base x baseline + required_base, rescaled to target A) | 34,757 -> 34,757 | 148,148 -> 148,147 | C13 22,392 | K10: tier 1 (closable at precedent) -> covered on course; K12: tier 2 (beyond precedent) -> tier 1 (closable at preceden | **material**: layer C13 tier moves 22,392 bn; input status changes; output depending on K17 moves 68,663 bn; top-5 inputs by output depending change |
| E2 | Energy and materials at the best achievable path (IEA NZE for energy, steel, cement, aluminium; GCMO High Demand for minerals) instead of the baseline | rerun with every energy and material ratio at the IEA STEPS 2035 path (Agent 4 rerun intensities) | 34,757 -> 34,757 | 148,148 -> 148,148 | C01 0 | K02: tier 1 (closable at precedent) -> tier 2 (beyond precedent); K14: tier 2 (beyond precedent) -> tier 1 (closable at  | **material**: input status changes; output depending on K02 moves 83,060 bn; top-5 inputs by output depending change |
| E3 | Labour at each sector's fastest 9-year output-per-worker growth everywhere at once, a rate no economy has sustained in all sectors together | rerun with output per worker on the 2014-24 sector trends (Agent 4 rerun intensities) | 34,757 -> 34,757 | 148,148 -> 148,148 | C01 0 | none | **material**: output depending on K17 moves 15,575 bn; top-5 inputs by output depending change |
| E4 | The on-course shortfall is credited to tier 1 because every input it claims is covered at its fastest precedent (lead decision); but economy-wide cove | rerun with the previous rule: the whole on-course shortfall in tier 2 (tier-1 room = joint maximum - on course | 34,757 -> 24,801 | 148,148 -> 158,104 | C13 5,513 | none | **material**: total tier moves > 1% of target; layer C13 tier moves 5,513 bn |
| E5 | Output depending on copper, lithium and rare-earth shortfalls is computed with substitution (CES 0.4, Agent 4) rather than fixed coefficients | rerun with fixed coefficients for copper, lithium and rare earths | 34,757 -> 34,757 | 148,148 -> 148,148 | C01 0 | none | **material**: output depending on K07 moves 277,498 bn; top-5 inputs by output depending change |

- E1: target moves (bn): C13 -22,392, C14 +7,781, C01 +5,549.
- E2: required (base -> test): K01 74,059 -> 69,351, K02 445 -> 573, K06 51 -> 50, K07 1,208 -> 1,014, K09 2,902 -> 3,076, K10 5,590 -> 6,095, K20 209 -> 208.
- E3: required (base -> test): K01 74,059 -> 74,698, K03 537 -> 635, K04 1,202 -> 1,511, K17 571 -> 603.
- E4: tier 1 34,757 -> 24,801; tier 2 148,148 -> 158,104.
- E5: output depending (base -> test): K06 1,783 -> 213,618, K07 385 -> 277,882, K08 21 -> 159,960.

Reading (computed): material E1, E2, E3, E4, E5; minor none. Tests that leave the tier totals unchanged move what layers and inputs need (E1 moves C13 by 22,392 bn). Status changes under the baseline energy and material path (E2): K02: tier 1 (closable at precedent) -> tier 2 (beyond precedent); K14: tier 2 (beyond precedent) -> tier 1 (closable at precedent). E4 restores the previous rule (the whole on-course shortfall in tier 2, as the joint maximum subtracts it): tier 1 changes by -9,957 bn. E5 decides where minerals rank in part D's list.

## 8. Limits, and the cheapest test that would settle each

- NZE as "best achievable": the IEA path is drawn for an economy growing about 3% a year; a doubling economy turns its capital over faster (efficiency could be faster) and builds more (materials could be higher). Test: IEA World Energy Balances energy intensity of the fastest-growing economies (China 2002-11) against their output growth.
- Minerals on the GCMO High Demand path (no NZE case in GCMO 2026). Test: the IEA Critical Minerals Data Explorer NZE series.
- Machine-labour electricity rests on assumed draw and hours (RQ_A01, RQ_A02). Test: metered kWh per shift of deployed humanoids; metered energy of a GB300 NVL72 rack.
- Capital stock at today's capital-output ratio: the largest single judgement among the inputs; at the 1990-2023 low the requirement is close to on course. Test: capital-output ratios in the fastest-growing decade of each large economy (PWT).
- Cropland assumes crop tonnes per unit of real output change only with yields (RQ_A05); higher-value food per tonne would lower it. Test: FAO food production index against real household food consumption 2014-24.
- Output depending uses fixed coefficients for every input but copper, lithium and rare earths, an upper bound (E5 shows how much substitution matters). Test: price elasticities of demand for steel, aluminium and electricity.
- K15 (clinical approval) equals C05 capacity under current rules (Agent 4 definition), so its tier 2 repeats C05's; it is not an independent permission shortfall.
- U3 and U4 (inventories, net exports) are accounting rows that grow with the target; their tier 2 is not producible by any capability. U2 is negative; its on course exceeds its target.

## Files

- target_economy.csv, requirements.csv, inputs_v3.csv, input_layer_v3.csv, layers_v3.csv, totals_v3.csv (file formats of the brief; every numeric column has _label and _src).
- mix_choice.csv (why the base mix), mix_compare.csv (alternative minus base), shortfall_tier.csv (each layer's on-course shortfall by input, its precedent coverage and tier, and the test variant), model_constants.csv, errors5.csv, sources.csv.
- code/: run_all.sh, exio.py (aluminium and crop footprints), landcap.py (PWT, FAOSTAT, WDI series), inputs.py, model.py, build.py, errors5.py, sources.py, report.py, hashes.py, check.py; work/ intermediates; raw/ faostat_landuse and uptime_survey2025 (new downloads; every other raw file is read in place from stage1-v2).
- Source prefixes: R stage1-v3/repair, B stage1-v2/v1/bridge, BL stage1-v2/baseline, A4 stage1-v2/agent4_bottlenecks, A4v1 stage1-v2/v1/agent4_bottlenecks, M2 stage1-v2/merge; bare ids are this folder's sources.csv.
