← Back to Lifeprint

Lifeprint Methodology

30 September 2026 · R. Christian Smith

Scope and claims

Lifeprint estimates the effects of one person's way of living across four areas — on themselves, on people alive today, on future generations and on the living world — from about 60 questions, and shows each as a range against references the person chooses. It is free, non-commercial, and runs entirely in the browser: no answer leaves the device.

This document describes tier 1, version 74, the version published on 30 September 2026. Tier 1 asks only what a person can answer from memory in five to thirty minutes. Tier 2, which would read bank statements and barcode scans, is not built.

What it produces. 45 measure cards, of which 39 compute from tier-1 answers and 6 wait for data that does not exist yet or for tier 2. Of the 39, 9 are modelled through published factors, 24 are what the person reported, 2 are risk screens, 2 are Lifeprint's own constructions, and 2 are context about the person's employer rather than about the person.

What it deliberately does not do.

• No single score. The measures are in different units and rest on different quality of evidence. Adding them would require weights, and the weights are the argument, not the arithmetic. Where Lifeprint does rank changes across measures, it uses weights the person sets themselves and shows them.
• No judgement. Cards state what was measured and against what. The suggestions attached to each card are options with their effects, re-run through the same calculation, not instructions.
• No hidden uncertainty. Every figure is a range. Where a reference has no dataset behind it, it is drawn as a placeholder and labelled as one.
• No attribution of an organisation's emissions to an employee. Where a person works is shown as context about the sector, never converted into a share of harm.

What it is not. It is not an audit, not a certification, and not a legal or financial instrument. Several figures are first-pass approximations, listed in full under Known limitations. It has not yet been reviewed by anyone outside its author.

How a measure is made

Every card follows the same path: an answer is mapped to a physical quantity, the quantity is multiplied by a published factor, and the result is widened into a range.

1. Answer to quantity. Answers are bands, not numbers. “a few times a week” becomes 182 servings a year; “60–100 m²” becomes 0.8 of an average dwelling. The band midpoints are set once, in code, and are listed with each model below.
2. Quantity to effect. The quantity is multiplied by a factor from a named dataset — DEFRA for travel, Poore & Nemecek for food, EXIOBASE for anything bought with money.
3. Effect to range. The central estimate is multiplied by a low and a high multiplier that reflect how much the factor and the band could be wrong. These multipliers are judgement, not computed confidence intervals, and Uncertainty below says how they were chosen.
4. Range to card. The range is drawn on a shared axis against the references the person ticked. A skipped question leaves its card blank rather than assuming a value.

Evidence types. Each card is labelled with the kind of claim its number is, because a self-reported hour and a modelled tonne are not the same sort of fact.

The 24 self-reported cards carry no model at all: the answer is the measure. Their only modelling decision is the band midpoint, and their only real uncertainty is whether people report themselves accurately, which Lifeprint cannot check.

Country. The person picks a country to compare against. That choice changes the national averages, the currency, the diet mix, the grid intensity and the spending baseline — not the method.

Where the numbers come from

Every dataset below is publicly available without payment. That was a constraint, not a preference: Lifeprint holds no licences, so anything behind a paywall was excluded even where it was better.

The per-country reference notes in the page footer give the exact year and instrument for every national average, and each card's Source and caveat names what that specific figure rests on.

The spending model

Five cards — carbon, land, wages, materials and pollution — depend on an estimate of what the person spends in a year, because EXIOBASE gives effects per euro. Tier 1 does not ask for spending, so it is inferred from income band and household size.

where spend_pp is national average household spending per person from EXIOBASE 2022, income_mid is the midpoint of the chosen income band, n is household size, and ȳ is mean national net income per household member.

The exponent is 0.6, an income elasticity of spending: spending rises more slowly than income, because higher earners save a larger share. Until version 70 it was 0.7, a value calibrated to remove a bias and with no source behind it. It is now bounded by data. Eurostat's household budget survey (hbs_exp_t133, mean consumption expenditure by income quintile, purchasing power standards per adult equivalent, 2020) gives a top-to-bottom quintile consumption ratio of 2.11 in Spain and 2.19 in France. The income quintile share ratio for the same year (ilc_di11) is 5.77 and 4.42. Taking logs of each pair implies an elasticity of 0.43 for Spain and 0.53 for France.

Those are floors rather than estimates, because budget surveys under-report spending at the top of the distribution, which flattens the measured consumption gradient. 0.6 sits above the floor to allow for that and well below the old 0.7. It is still a judgement, now with data on one side of it, and a proper estimate needs household micro data rather than published quintile means.

Non-essential spending. One question asks how much the person spends beyond essentials compared with others on a similar income. The five answers scale only the non-essential part of each spending-based measure, by 0.5, 0.75, 1, 1.3 and 1.7. The non-essential share is taken from EXIOBASE as restaurants, furniture and household goods, recreation, vehicles, and clothing and electronics where not answered separately: 15.1% of spending in Spain, 23.4% in France, 34.7% in Ireland, 14.9% in the United States.

EXIOBASE books the retail margin on non-essential goods under retail trade rather than under the goods themselves, so the non-essential share is understated and this lever moves the measures less than it should.

Where the answers override the model. Clothing (about €30 an item) and devices (about €500 each, spread over three years) are taken from the person's own answers and subtracted from the residual, so they are not double-counted. Food, home energy, fuel and flights are excluded from the spending-based part entirely, because they come from physical answers instead.

EXIOBASE: processing and reconciliation

Five cards rest on EXIOBASE, so how it was processed matters more than any other methodological choice here.

Which table. EXIOBASE 3.10, IOT_2022_pxp — the 2022 product-by-product table. Not the 2024 table, although it exists and is more recent. The 2024 tables are projections rather than measured years, and they failed the checks below: the air-emissions table returned either blanks or a world total of 216 Gt CO₂e, five times the real figure, and the land and water tables were empty. Measured years are preferred over projections throughout.

Processing. For each of the four countries, the household final-demand column of the Y matrix was taken — so every figure reflects what households in that country actually buy, domestic and imported in real proportions, rather than the domestically made version of each product. The Leontief inverse was computed locally from the Z matrix and gross output. Extension tables were read positionally rather than through their multi-row headers, because trusting the header parsing is what produced the broken 2024 read: a row with any numeric cell is a data row, and blank cells in a data row count as zero.

Reconciliation. Nothing was used until its world total was checked against an independently published figure. These are the checks that passed:

Because f-gases are excluded, the carbon card understates a full consumption footprint by roughly the f-gas share, and a person on the national average sits somewhat below the published consumption-based average. The country average tick on that card is consumption-only so that it matches this scope.

One correction worth recording. “Private households with employed persons” is excluded from the supply-chain wages card. Ireland's row in that sector is misallocated to Latin America in the underlying data, which would have put a large share of an Irish household's apparent wage footprint in the wrong continent. Paid domestic work is measured directly instead, on its own card.

What this method cannot do. EXIOBASE is monetary: two products in the same sector at the same price have the same footprint. A person who buys one expensive, durable, locally made coat and a person who buys one cheap coat at the same price register identically. Product-level distinctions are exactly what tier 2 would add.

Diet, land and water

Four questions — red meat, poultry, fish, dairy and eggs — drive three cards. Frequencies map to servings a year: daily 365, a few times a week 182, weekly 52, rarely 12, never 0. Servings are 0.15 kg for red meat and poultry, 0.14 kg for fish. Fish is split between wild-caught and farmed: EU apparent consumption is 71% wild by weight (EUMOFA, The EU Fish Market 2025, 2023 data), and the United States is taken at about half and half. The farmed part is 80% fish and 20% prawns on Poore & Nemecek means; the wild part uses Parker et al. 2018, 2.2 kg CO₂e per kg landed, with no cropland and no irrigation. Treating all fish as farmed, as Lifeprint did until version 70, overstated the emissions, land and water behind fish by roughly a factor of two to three for European users. A plant base of 0.25 kg grains or potatoes, 0.3 kg vegetables, 0.2 kg fruit and 0.05 kg pulses is added every day regardless of the answers, as are 0.25 kg milk, 0.03 kg cheese and 0.04 kg egg per day for those eating dairy.

Red meat is split by what each country actually eats, because a kilogram of red meat means something very different in Spain and in the United States. Beef from beef herds carries roughly eight times the emissions of pork per kilogram, so the pork–beef–lamb mix dominates the result.

Shares come from Delgado et al. 2022 and MAPA for Spain, FranceAgriMer and INRAE for France, CSO and Teagasc for Ireland, USDA ERS and the Beef Checkoff for the United States. Per-serving factors then follow: Spain 3.81 kg CO₂e, France 5.89, Ireland 4.54, the United States 7.83; poultry 1.48, fish 2.28, dairy and eggs 1.69, the plant base 1.86.

Known bias. Poore & Nemecek give global means including deforestation, not European production. An average Spanish diet comes out around 2 t CO₂e a year, above the 1.2–1.6 t that Spanish studies using European data report. The card says so.

Land adds three parts: diet land converted at about 1.3 global hectares per hectare, the home's footprint by floor area divided by household size, and land behind goods and services from EXIOBASE, converted with Global Footprint Network equivalence factors of 2.52 for cropland, 0.46 for pasture and 1.29 for forest. Most of the goods-and-services part is forest, for paper, packaging, furniture and timber.

The national land references are land only — 1.39 gha for Spain, 1.46 for France, 2.24 for Ireland, 2.60 for the United States — computed from the same EXIOBASE tables rather than taken from published ecological footprints. This matters: a published footprint includes the forest needed to absorb CO₂, which Lifeprint counts on the carbon card instead. Mixing the two scopes was an error caught before publication. The world biocapacity reference of 1.48 gha does include that carbon land, so it overstates the room left, and the card says so.

Water counts freshwater withdrawals for irrigation and processing only, from the same Poore & Nemecek tables: not rain falling on pasture, and not weighted for local scarcity. Cheese, nuts, rice and farmed fish dominate per kilogram. A litre withdrawn in a wet country and a dry one count the same, which is the card's main weakness.

Food waste is a fourth diet-adjacent card. National averages come from Eurostat env_wasfw for 2024, household activities only: Spain 24 kg per person per year, France 62, Ireland 41, with the United States at 73 from the EPA Wasted Food Report's 2019 residential tonnage. All include inedible parts such as peel and bone. Until version 70 Lifeprint used 77, 30, 77 and 100, mixing national sources on incompatible scopes — France's figure was edible food only — which made three of the four wrong by roughly a factor of two and put the countries in the wrong order. Because a four-option answer cannot carry an absolute weight, the bands are now set relative to each country's average rather than as fixed kilogram values.

Home energy and travel

Home. Household energy use per occupied dwelling comes from Eurostat 2024 divided by the occupied dwelling count, with each country's own space-heating share. Floor-area bands scale it by 0.55, 0.8, 1.25 and 1.8 around an average dwelling of about 100 m², and the result is divided by household size.

Heating fuel factors are gas 0.20, oil 0.27, district heating 0.15 and wood 0.03 kg CO₂e per kWh, wood being counted as biogenic under DEFRA's convention. A heat pump divides heating demand by three before applying the grid factor. A green tariff or own panels multiplies electricity by 0.25 rather than zero, because a tariff does not by itself remove the generation. Non-heating electricity is the remainder at the grid factor.

The grid intensities are Ember's 2024 lifecycle figures, covering upstream fuel supply, supply chain and manufacturing, on one metric for all four countries. National regulators publish combustion-only figures that run lower — RTE reports 21.7 g direct and 30.2 g lifecycle for France against Ember's 40.5, and EPA Ireland 226 g against Ember's 271 — and the gap is not a constant, so the two sets cannot be mixed. The US dwelling figure is EIA's 2024 residential energy divided by occupied housing units, with the heating share from the 2020 Residential Energy Consumption Survey; the 45% used until version 70 was an artefact of dividing a per-heating-household average by a per-all-households total. Spain's and France's are low enough that heating choice dominates the card; Ireland's and the United States' are high enough that electric heating without a heat pump can be worse than gas.

Flights. DEFRA 2025 economy factors with radiative forcing: 0.126 kg CO₂e per passenger-km short-haul and 0.117 long-haul. Return distances are assumed at 2,000 km short-haul, roughly Barcelona to London, and 14,000 km long-haul, roughly Europe to the Americas or Asia. Flight counts map to 0, 1.5, 4, 8 and 12 returns a year.

These assumed distances are the single largest source of error on the flights part. Someone who takes six short domestic hops and someone who takes six European returns are counted the same.

Cars. DEFRA 2025 average-car factors: petrol 0.163, diesel 0.173, hybrid 0.128 kg CO₂e per km. “Petrol or diesel” is taken as 0.168. An electric car is 0.05 kg/km on the Spanish grid at 0.21 kWh/km including well-to-tank. Distance bands are 30, 100, 225 and 350 km a week. Someone with no car but who answers that they mostly get around by car is given 0.8 t a year, and by motorbike 0.3 t.

Running costs, used for the money line under each suggestion, come from the EU Oil Bulletin and national price sites for fuel (September 2026), and Eurostat nrg_pc_204 and nrg_pc_202 for household electricity and gas in the second half of 2025. Litres per 100 km are derived from the DEFRA per-km factors divided by DEFRA per-litre factors: 7.9 petrol, 6.7 diesel, 6.2 hybrid; electric cars at 21 kWh per 100 km charged at the household price.

Wages, materials and pollution along supply chains

These three cards take the estimated spending from above and multiply it by EXIOBASE intensities per euro, weighted by what households in that country buy from each country of origin.

Whose work you depend on. Of every €100 spent in Spain, about €45 is paid as wages somewhere along the chain: €3.90 to people with only basic schooling, €24 with secondary, €18 with higher education. About €36 is profit; the rest is taxes and depreciation. The basic-schooling wages behind a Spanish basket are earned in Spain (69%), the rest of Africa (7%), China (4%), the rest of Asia-Pacific (4%) and the UK (3%).

Two limits matter. Skill here is education level as EXIOBASE defines it (ISCED), not pay or status. And these are totals, not rates: the card says where the work is, not whether it was paid fairly. Whether it was paid fairly is a different question, approached on the clothing living-wage card and properly only in tier 2.

Raw materials extracted uses EXIOBASE's domestic-extraction-used extension across 62 categories, split into metal ores, fossil fuels, non-metallic minerals and biomass. Clothing carries about 1.01 kg per euro in Spain and electronics 0.56, against 0.497 for average spending, so those two are taken from the person's own answers rather than from the residual. This is extraction, not waste: most of the tonnage is sand and gravel for buildings and roads, which is worth saying because a figure of 8 t a year otherwise reads as rubbish produced.

Air and water pollution leads with fine particulates because they do the most harm to health, and reports nitrogen oxides, sulphur oxides, ammonia, and nitrogen and phosphorus to water alongside. Ammonia and sulphur are flagged on the card as upper bounds, for the reconciliation reasons above. Emissions from the person's own boiler and car exhaust are not included here — those sit on the carbon card — so this card is supply chains only. The food part uses the country's average diet rather than the person's, because the EXIOBASE food basket is not re-weighted by the diet answers.

Devices are the one place where a spend-based figure was judged too poor to use alone. Price is a weak guide to a device's footprint, so the estimate is the midpoint between the spend-based figure and manufacturer life-cycle assessments of about 200 kg CO₂e, spread over three years.

Screens, self-reports and indicators

These are the weakest cards in the tool, and they are labelled as such on the page.

The forced-labour and child-labour screens count how many categories in a person's answers carry a flag somewhere in their supply chains, using the US Department of Labor's 2024 list of 204 goods from 82 countries. Buying any clothing raises the garments-and-cotton flag; any devices raises electronics; coffee or cocoa without fair-trade raises that; any fish raises seafood.

This is a count of risks, not of people, and not evidence about the person's own purchases. Tier 1 never learns which brands or origins were involved, so the screen cannot distinguish a person buying from a scrutinised supplier from one buying from the worst in the sector. The list also over-represents China and under-covers the United States and its close trading partners, which means the screen partly measures where investigators have looked. Both limits are printed on the cards.

Self-reported measures — 24 of the 39 that compute — apply no model beyond a band midpoint. Sleep, care given, hours with family, people you could call, worthwhile hours, repairs, giving, voting, civic time, declared income. For these, the only question is whether people report themselves accurately, and Lifeprint has no way to check. Social desirability almost certainly inflates the good ones.

Two cards are Lifeprint's own constructions, with no dataset behind the scale, and say so:

• Interest, collective ability, inquiry counts three conditions that bear on whether a person can act at work — whether pay depends on volume, whether a collective route exists, whether the person has looked into what their work does. Each scores 0, ½ or 1 and the three are added. Two of the three come from Iris Marion Young's account of structural responsibility; the third is a duty of inquiry. No dataset backs the scale. It exists to put the three conditions side by side, not to grade anyone.
• People you manage: voice at work rests on a single question about whether a complaint channel exists and has changed something.

Decision reach at work is context, not a measure of the person. It reports the sector's emissions and employment, and emissions per job, from Eurostat — and then explicitly declines to convert that into a share of harm attributable to the individual. The reasoning is Young's: connection to a structure creates a responsibility to act, not a measurable slice of past damage. This was a deliberate refusal to produce a number that would have been easy to compute and impossible to defend.

References and positions

A measure alone means little, so each card draws comparisons on a shared axis. Lifeprint distinguishes three kinds, and marks them differently.

Observed figures (grey) are what is: a national average, a minimum wage, a median wealth. They carry a source and a year.

Positions (orange rings) are what could be, on a stated basis. They are arguments about fairness, not facts, so each one carries its reasoning and the person can switch it off. All five are on by default.

Placeholders (dashed rings) are references with no dataset behind them yet. They are drawn so the card is not empty, and labelled “placeholder, no dataset yet” so nobody mistakes them for evidence. Eight remain, listed under Known limitations.

A card records which of its references are sourced for each country. The same reference can be sourced in one country and a placeholder in another — the carbon card's income-group tick is sourced for Spain, France and the United States but not Ireland, where only a 2006 study by decile was available and was judged too old to use.

One scope warning. On the carbon card the country-average tick is consumption-only, matching Lifeprint's scope. A second tick adds investment-attributed emissions from the World Inequality Database, and the income-group figures use that wider scope, so they sit higher by construction. The two are not comparable with each other, and the card says so.

Uncertainty

The ranges are judgement, not statistics. Each modelled measure is multiplied by a low and a high factor chosen to reflect how wrong the band midpoint and the published factor could jointly be. They are not confidence intervals, they were not computed from variance in the source data, and they should not be read as if they were. This is the honest position and it is worth stating plainly rather than dressing the numbers up.

The multipliers in use:

Where no central estimate is drawn. Some cards are marked coarse: the range is shown as a band with no dot, because the model has no reason to prefer the middle. This happens for land, water, materials, pollution, wages, food waste and animals, and for anyone who skipped the income question.

What the ranges do not cover. They cover factor and band uncertainty only. They do not cover: whether the person answered accurately; whether the national average used as a baseline fits that person's region or housing; f-gases and public services, which are outside scope entirely; or the possibility that a whole modelling choice is wrong. A range of 5–9 t does not mean the true value is 95% likely to be in that interval. It means the model's inputs could plausibly produce anything in it.

Rounding. Effects that round to zero are suppressed rather than shown as “−0.0”, so a suggestion with a real but tiny effect on a card simply does not appear on it.

Known limitations and open items

This list is meant to be complete. If something is wrong with Lifeprint and it is not here, that is a failure of this document. Version 70 closed most of what stood here before; what follows is what remains.

Six cards are blank because the data to fill them does not exist at tier 1:

Five cards now draw no comparison line at all, because the honest answer was that no trustworthy statistic exists: property beyond your home, where your money sits, repair rather than replace, things repaired last year, and people seen in person last week. Each says so on the card and gives the nearest published figures in its note, with their definitions, rather than a single invented average. Two placeholders remain: worthwhile hours, for which no national benchmark exists anywhere, and the reference on people seen in person.

Specific weaknesses that remain, roughly in order of how much they matter:

1. The income elasticity is still partly judgement. It is now 0.6 rather than 0.7, bounded below by a derivation from Eurostat data (see The spending model), but the derivation is a crude two-point calculation and budget surveys under-report spending at the top. A proper estimate needs household micro data. It drives five cards.
2. Wild-caught fish are costed on a different basis from everything else. The fish split uses Parker et al. 2018 for the wild part, which counts fishing-vessel fuel only — not refrigerants, gear, processing or transport — while the farmed part uses full life-cycle figures from Poore & Nemecek. The wild figure is therefore a floor, and the fish estimate is conservative. Poore & Nemecek does not cover capture fisheries at all, so no like-for-like figure exists.
3. The animal count does not distinguish wild from farmed. A kilogram of wild catch is many more individual animals than a kilogram of farmed fish, because small pelagic species dominate by number. The count therefore understates, probably substantially. The available estimate was too weakly sourced to use.
4. Assumed flight distances of 2,000 km short-haul and 14,000 km long-haul return still dominate the error on what is, for many people, their largest single item. Eurocontrol reports an average flight of 1,140 km in 2023 across all flights including intercontinental, which is consistent with the short-haul assumption but does not confirm it.
5. The United States falls back to EU figures for accidents at work by sector, because no equivalent series is loaded.
6. Accident rates are not comparable between countries. France's rates run about twice the EU average in almost every sector, which is a reporting artefact rather than a real difference in safety, and a zero may be a real zero or a count too small to publish.
7. Where a sector spans several NACE sections, the accident figure is a range across them rather than an employment-weighted average, because section-level employment is not loaded.
8. Ireland remains the weakest country. Its carbon average is consumption-based CO₂ only rather than full CO₂e, its income-group figures come from a 2006 study and are not shown, and its EXIOBASE figures are distorted by the scale of foreign-owned production booked there.
9. The diet model runs high against national studies, for the deforestation and global-mean reasons above.
10. Spain's food-waste figure may be method rather than behaviour. At 24 kg per person it is far below the EU average of 69, and it comes from a consumer diary panel rather than waste-composition analysis.
11. Ammonia and sulphur run about 2× and 1.5× above inventories and should be read as upper bounds.
12. The non-essential lever is understated, because EXIOBASE books retail margins under retail trade.
13. The welfare-label reference covers eggs only. It is the share of laying hens kept outside cages, which says nothing about meat, dairy or fish, and counts hens kept rather than eggs bought.
14. The US heating share is from 2020, the most recent Residential Energy Consumption Survey with consumption data; the next is due in 2027.
15. “A household like yours” was deliberately not built. It would have repeated the income-based estimate back as though it were observed data.

Not yet done: no external review and no tester round.

Versions and how to check this

Version 74, 30 September 2026. Tier 1. English, Spanish, Catalan and French.

The page this describes is Lifeprint, which carries its version number and a link to this document in its footer. Version 69 added that marker and that link. Version 70 closed most of the open validation gaps listed below and changed several figures; the changelog records which. Version 72 moved both onto their own site: this document is now published at app.rchristiansmith.com/lifeprint/methodology/ and the page links to it there rather than to a Claude artifact.

Versions are the published versions of the page itself. The changes that altered a number, rather than the interface, are:

The process rules adopted after the version 50 withdrawal, and followed since:

1. Reconcile every dataset against an independent published total before using it.
2. Prefer measured years over projections.
3. Local scripts self-check and print their reconciliation.
4. Warn about risky sources before starting, not after.
5. Run the pre-publish audit before every publish.

The pre-publish audit builds a profile and compares every modelled card against the national averages, flagging anything outside 0.5–2×. For a near-average profile the sourced measures land between 0.75× and 1.35×. Cards that flag are self-reported ones where the test profile genuinely differs from the national average, not model errors; the distinction is checked by diffing audit output between versions.

How someone could audit this. The page is a single HTML file with no build step and no server; every constant is a literal in the source with a comment naming its origin. Each card's Source and caveat names its dataset and year. The EXIOBASE processing is a separate Python script that prints its reconciliation as it runs, so its world totals can be checked against the table above without rerunning anything.

What would most improve confidence, in order: an external reviewer on the diet and spending models; a tester round to find where the interface misleads; replacing the calibrated elasticity with a sourced one; and loading the Eurostat accident tables properly.

Sources

The dataset table above links each source. Per-country reference notes, with the exact instrument and year for every national average, are printed in the page's own footer and change with the country selected.