← Back to charts
EN

Sources, methods, and data

Your journey into the accelerating world

Your Acceleration Compass

Discover four quizzes to explore your choices and vision for the future

World LeaderChange StyleFuture RadarAI World

At a glance

Seven areas monitored · Select a tile to explore its charts.

Why these colours? Read the evidence and limits

The analysis below explains the pressures in each area. A chart can appear in more than one area.

1. Monitoring an accelerating world

Explore the evidence, how its effects are distributed, and the new possibilities opened by AI.

1.1. Seven areas monitored

1.2. People & distribution: Growth for whom, concentrated where?

Average progress can hide different experiences. Distribution, ageing and the geography of growth belong beside the headline totals.

DISTRIBUTION

Who holds the wealth?

Compare the top tenth with the bottom half. Start with wealth, or switch to income for a different view of inequality. The world view and the distribution within a country answer different questions.

WID estimates, equal-split adults. Income is pre-tax national income after pensions; wealth is net personal wealth. Current downloaded histories end in 2023. They are not the separate 2025 snapshots in WIR 2026.

World Inequality Database ↗

The richest tenth of adults held 74.4% of global personal wealth in 2023, barely changed from 74.6% in 2016. The bottom half held 2.0%.

2016 question How much wealth did the richest groups hold compared with the rest of the population?

What to watch If wealth grows while those shares hold steady, the gap in money terms widens.

URBAN CHANGE

More people.
More density?

Not necessarily. A city can grow outward while its population grows. Compare population, people per square kilometre and concentration in a country's cities.

Six illustrative cities, not a representative global sample. UN WUP 2025 uses harmonised DEGURBA settlement extents, not administrative boundaries. Density is not a direct measure of crowding or service quality.

UN World Urbanization Prospects 2025 ↗

Dhaka gained 7.4 million residents from 2016 to 2025, yet its average density fell from about 5,970 to 5,370 people per km².

2016 outlook Rapid urban growth was expected to bring higher density and greater pressure on local services.

THE GEOGRAPHY OF DEMOGRAPHIC CHANGE

Different places, different futures.

In 2023, Africa had about 1.48 billion people and Europe about 747 million.

2016 outlook Regional projections highlighted the demographic weight of China and India.

Outlook The UN’s medium projection puts Africa at 2.47 billion in 2050 and Europe at 705 million. A rising world total can hide a regional decline.

UN World Population Prospects 2024, processed by Our World in Data. Estimates end in 2023; 2024 onward, including 2026, is a medium-variant projection. Dashed lines are projections, not observed outcomes. Source ↗

1.3. 2016 → 2026: ten years later, what held up, what changed

The 2016 book revisited in 2026: original expectations and concerns, the evidence that followed, and the arguments that need revising.

2016 → 2026 · A return to the evidence
Original expectation or concernWhat the decade revealedLooking back, ten years later

Population

Eight billion by 2022; continued demographic hypergrowth.

About 687 million more people in 2016–2025. Eight billion reached in 2022, as forecast.

Eight billion arrived on schedule. The world grew larger, but population growth slowed; the idea of ever-faster growth needs revising.

Global economy

Economic expansion, with rising consumption, inequality and waste.

Real global output grew by over a quarter in 2016–2025; output per person by roughly a sixth.

Continued expansion was correctly anticipated. Its uneven course makes the distinction clearer: producing more does not necessarily mean that everyone lives better.

Technology

Rapid advances in computing, robotics and AI.

The cost of GPT-3.5-level AI performance fell more than 280-fold from November 2022 to October 2024.

Rapid technical advances were rightly anticipated. Turning those advances into widely shared benefits has proved a slower, less straightforward journey.

Data

More data and knowledge, but saturation and privacy risks.

Internet use spread from fewer than half of humanity in 2016 to nearly three in four in 2025.

Wider connection became a reality. Greater understanding did not follow automatically; that still requires judgment and the willingness to question.

Environment

Growing material use and waste could overwhelm ecological recovery.

Annual plastics production rose from about 371 million tonnes in 2018 to 431 million in 2024: about 60 million tonnes more.

The concern about accumulating damage remains well founded. Some extinction estimates and explanations were too sweeping and need correcting.

Freedom and society

“Hyperdemocracy”, but also stagnation, surveillance and concentrated power.

Freedom House assessed 54 countries as less free in 2025, against 35 that improved.

The democratic optimism needs tempering. Warnings about surveillance and concentrated power have aged better: connection has not guaranteed a greater voice.

Health

Medical advances, but chronic disease and unequal access.

Global life expectancy rose from 72.2 years in 2016 to 73.5 in 2024, despite the pandemic setback.

Medical progress brought real gains. Its promise still depends on access to care, treatment of chronic illness and resilience during crises.

A larger world is not always a faster-growing world.

The concern that human capabilities can outpace our ability to manage their consequences remains persuasive. What needs revision is the idea of one accelerating trajectory across all seven domains, or an inevitable loss of control as speed increases.

From Part I of the 2026 book: “Ten years later: what held up, what changed.” These are assessments of the original arguments, not a numerical prediction score. Each outcome retains its evidence dates; full-year 2026 outcomes and projections for 2050 remain untested.

1.4. The coming of AI

AI connects the trends above: computation and data become tools that can help people analyse, create and act. The question is how to turn expanding capability into useful, widely shared progress.

MEASURING CAPABILITY

What has changed?

By 2025, leading AI systems had surpassed the human benchmark on selected science and maths tests but still fell short on tasks that require operating a computer.

2016 question How quickly would machines match or surpass human abilities, and on which tasks?

What to watch Can systems close the computer-use gap when tested under the same rules?

Stanford AI Index benchmark trajectories, 2012–2025, relative to each task’s human baseline of 100 percent. Several exceed the reference level, while agent computer use remains below it.
Stanford HAI, AI Index 2026, Figure 2.1.1 · data through 2025. Each line has its own human reference; 100% is not perfect accuracy. Different systems can supply the results across tasks and years. Tap the chart to enlarge.

A higher benchmark score opens possibilities. Dependable performance outside the test requires separate evidence.

Benchmarks can saturate, contain errors or reward familiarity with a test. Read this as evidence of changing task capability, not proof of general judgment or safe autonomy. Source and interpretation ↗

WHY NOW?

Three advances reinforce one another.

Better data

Useful examples, careful selection and checked demonstrations shape what a model can learn. Quality and coverage matter alongside volume.

Better learning methods

Architectures, training objectives and feedback change how a model uses its examples. Pre-training builds broad patterns; post-training shapes behaviour and task performance.

Better computation

Accelerators, memory and parallel processing make larger training runs and more experiments feasible. Efficient software helps turn hardware into useful work.

Better examples improve learning; better methods use resources more effectively; more usable computation expands what can be tried.

Conceptual explanation, not a measured allocation of progress between three causes.

From microprocessors to AI chips

The transistor curve spans more than half a century. Moore’s roughly two-year doubling became an industry reference, not a law guaranteeing performance. Modern GPUs add parallel computation and advanced packaging; some combine multiple compute dies. Transistor totals therefore do not measure speed, efficiency or useful work on their own.

Memory, data movement, numerical precision and software also matter. The same hardware can deliver different results depending on how it is used.

Explore the transistor and computing charts in the Atlas ↑

FROM POSSIBILITY TO DIFFUSION

Capability matters more when people can use it.

Stanford’s 2025 AI Index reports that the price of querying a model at a fixed GPT-3.5-level benchmark threshold fell from $20 to $0.07 per million tokens between November 2022 and October 2024—more than a 280-fold reduction. This compares a particular performance threshold, not every model or the full cost of deploying AI.

Lower prices can widen access. Useful adoption still depends on skills, reliable infrastructure, suitable workflows and who controls the tools.

Stanford AI Index 2025 · inference prices ↗

Explore possible AI futures and the choices ahead ↓

How AI could affect all seven areas

Illustrative opportunities and pressures to examine, not measured net effects.

AreaPotential contributionQuestion of control
PopulationCare assistance and planning for changing needsDoes support preserve autonomy and reach underserved communities?
EconomyProductivity and new servicesWho gains, who is displaced, and how concentrated are the returns?
TechnologyFaster design, coding and experimentationCan testing and oversight keep pace with deployment?
DataBetter analysis and cyber defenceHow do we limit surveillance, misuse and automated attacks?
EnvironmentResource optimisation and environmental monitoringDo verified savings exceed added energy and material demands?
HealthResearch and clinical assistanceAre benefits validated, affordable and reliable across populations?
Freedom & societyAccess to knowledge and participationCan people challenge decisions and retain independent judgment?

Faster capability is not the objective by itself.

AI could increase the pace of change and improve our ability to respond. Which effect dominates depends on deployment, institutions and the distribution of power. The aim is a sustainable pace of useful progress, with the capacity to detect harm, correct mistakes and share the gains.

Continue to Acceleration & Absorption ↓

2. Acceleration & absorption

Understand the shapes of growth, then explore how the capacity to respond can keep pace.

2.1. Understanding growth shapes

The shape of a curve changes the demands a system must absorb. Compounding, connectedness and cycles help explain those demands; our ability to observe, act and correct determines how well we respond.

SCHEMATIC CURVES · NOT FORECASTS

Three curves, three different stories.

A rising level is not enough to establish an increasing percentage growth rate. Read the shape, the time horizon and what is growing.

Linear

Add 0.18 at each step and the index rises from 1 to 19 in 100 steps.

Next ten steps Ten more steps add exactly 1.8. The same addition becomes a smaller percentage as the total grows.

Exponential

At 3% per step, the index grows from 1 to 19.2 in 100 steps.

If the rate holds Ten more steps add about 34% to the new, larger base. The percentage stays constant while each addition grows.

Saturation / S-curve

Here growth is fastest near 10, halfway to the ceiling of 20. The total keeps rising as each new increase shrinks.

2016 example Bacteria growing in a finite environment illustrated how limited resources can slow growth.

Illustrative shapes on a shared linear index scale, starting at 1. Time is schematic; these are not fitted histories.

Factors shaping growth

01 / COMPOUNDING

Growth builds on growth.

Each increase expands the base for the next. Several drivers can also reinforce one another. A small sustained rate can create a large change in scale.

02 / CONNECTEDNESS

Consequences cross domains.

A change in one area can alter demands in another: more data can require more energy and stronger security. Interactions may amplify or dampen pressure.

03 / CYCLICITY

Trajectories can turn.

Feedback, constraints and recurring cycles can interrupt expansion or bring reversal. Neither endless growth nor inevitable collapse follows from a rising curve alone.

ARITHMETIC ILLUSTRATION

What a century does to a small percentage.

Every line starts at 1. The annual percentage stays constant; the amount added each year grows.

All four rates

After a century, 1% annual growth turns the starting amount into 2.7 times as much. At 5%, it becomes 131.5 times as much.

Assumption Each rate holds for all 100 years. Interruptions would change the endpoint.

Lower rates, closer view

After a century, the starting amount grows 7.2-fold at 2% a year and 19.2-fold at 3%.

Reading the gap One extra percentage point each year more than doubles the final amount.

Linear axes in both panels, with different vertical ranges. Level = (1 + annual rate)years. Constant rates, no shocks or saturation: mathematical scenarios, not forecasts. Multipliers show the level reached, not the sum of annual flows.

What grows matters.

If the rising quantity is an annual physical burden, each year can demand more collection, treatment or infrastructure. Unresolved flows can accumulate into a stock even when their growth slows. Capacity, prevention and ecological limits determine the consequences.

FROM SCALE TO RESPONSE

Can the capacity to respond keep pace?

Absorption means the practical ability to handle change: understand it, act on it and correct its consequences. A one-time improvement can create breathing room, but sustained growth may use it up. The following scenario compares pressure and response on one assumed scale; it does not claim a measured global absorption limit.

ILLUSTRATIVE SCENARIO · CYBERSECURITY

A one-time capacity gain buys time.
It does not stop workload growth.

In this scenario, a one-off 3.1-fold boost to capacity buys about 6.2 years before a workload growing 20% a year catches up.

What changes it If capacity also improves every year, the crossover moves later. If it grows as fast as the workload, the gap never closes.

Common workload index: manual capacity = 100. All future lines are assumptions, not observed global cyber threats or a forecast of attacks.

What comes from evidence, and what is assumed?

This example uses a Microsoft October 2025 randomised study of 167 professional analysts. Its workflow calculations imply roughly 3.1–6.5× correctly identified malicious emails per analyst-minute under particular prevalence, precision and recall assumptions. The lower 3.1× case assumes 20% malicious mail and 80% precision and recall, with automated benign-mail handling.

The 20% annual workload increase is illustrative. Staffing, task mix and effectiveness are held constant; implementation overhead, errors and automation bias are not modelled. A task-specific vendor study does not establish a 3.1× reduction in breaches or a universal productivity gain.

Microsoft research · study and assumptions ↗
01

Observe & interpret

Measure detection delay, error rates and the human ability to check an answer. Faster output alone does not establish better judgment.

02

Decide & respond

Measure the time from a warning to action, trained staff and available infrastructure. Capacity is specific to the task.

03

Correct & recover

Measure appeal times, fallback readiness and successful recovery. A response system needs a practical way to change course.

Control can enable useful speed. Explore how prevention, greater response capacity and shorter delays change the gap in Speed and control simulator ↓.

2.3. Project future scenarios

Explore the conceptual relationship between pressure, adaptive capacity and exposure. These controls express assumptions; they do not measure the world's current state.

Acceleration potential A
Adaptive capacity C
Relative pressure Ω

In the default scenario, pressure reaches 4.9 times response capacity in year 20.

Model assumptions Pressure grows 8% a year. Capacity starts growing two years later, at 3%. An earlier response reduces the gap even with the same rates.

A = ∛(P × I × N)   C = ∛(G × R × L)   Ω = (A ÷ C) × E

Dimensionless conceptual ratios, not calibrated indicators, probabilities or safety thresholds. The time paths are an illustrative extension: pressure = 100 × A × E × (1 + p)t; capacity = 100 × C × (1 + c)max(0, t − delay). Equal values only mean equality on this assumed common scale. No empirical global “absorption capacity” is claimed.

3. Futures & governance

Explore possible directions, then the institutions and safeguards that could help shape them.

3.1. Projecting some potential futures

Five selected from the ten scenarios in the book. Three are illustrated below, with two variants of Mother AI showing who has final authority. These are possibilities to examine, not forecasts or a ranking of likelihood.

Humanity under control

An authority uses AI to impose control, or dependence becomes impossible to challenge. Material comfort may remain while meaningful consent and freedom disappear.

Deliberate limits

Societies adopt useful AI while deliberately limiting its authority, surveillance and reach. Restraint may protect freedom and human skills, but can also delay benefits or protect existing privilege.

Scenario 1 now distinguishes Mother AI with final authority (A) from Mother AI under retained human authority (B). Variant B overlaps with scenario 2, human–AI partnership. The ten scenarios remain a set of possibilities, not an exhaustive or exclusive classification.

Source: Part III, Chapter 1, “Humanity and AI: Possible Futures, Shared Choices”, scenarios 1, 2, 4, 5 and 7. The book also explores augmented humanity, machines after humanity, uneven or interrupted progress, intelligence from beyond Earth and divine intervention. Scenarios can overlap; the final two are explicitly thought experiments.

3.2. Governance options for speed and control

These models explore how authority is distributed between human institutions and whether humans retain the final say over AI. They can overlap. The diagrams illustrate arrangements for discussion, not established evidence that any model would work.

Control with checks, participation and correction.

The manuscript distinguishes chosen AI authority from human-led governance. Its governance chapter sets out mandates, independent review, appeals, emergency intervention and recovery. The diagrams below make decision and checking routes visible; they do not establish that any design would work automatically.

GOVERNANCE MODEL 1

National institutions and market-led AI

A country-level model: public institutions set rules while companies develop and operate AI. It resembles parts of today’s arrangements, without implying that every country has the same political or economic system. Cross-border effects still require cooperation.

Citizens + national institutions
Public rules + company decisions
AI services affect people and ecosystems
Complaints → independent regulators and courts → correction

Courts and regulators review within their jurisdiction. International agreements connect national systems.

Strength and vulnerability

Competition can encourage invention and preserve choice. Capture, unequal access and regulatory gaps can weaken control. Clear duties, independent scrutiny and cross-border coordination are the checks to examine.

GOVERNANCE MODEL 2

Global parliament within a system of shared powers

A possible global parliament could establish common protections for specified cross-border problems. National, local and sector institutions would retain decision-making powers within those protections. Independent review would enforce the limits of each institution’s authority. Representation, enforcement, citizen appeals and any expansion of the global mandate would require explicit agreement.

People represented through agreed democratic procedures
Global parliament: defined cross-border powers
National, local and sector institutions: retained decision powers
Independent review: decisions, remedies and limits of authority

Shared protections, coordination and resolution of disputes connect these institutions. Citizens need accessible routes to challenge both implementation and the rules themselves.

Strength and vulnerability

Shared rules could address problems beyond one country’s reach. Unequal representation, weak enforcement and concentrated power are risks. Limited mandates, public reasons and independent powers of review would be central.

GOVERNANCE MODEL 3 / VARIANTS A AND B

Mother AI: two variants

The same coordinating intelligence could operate under two different allocations of authority. The human-led variant overlaps with human–AI partnership; both are hypothetical possibilities.

A · AI has final authority

People voluntarily entrust Mother AI with final authority. Specialist systems carry out its decisions. Independent evidence and freedom to question preserve critical judgment, but humans have no binding veto.

People choose to delegate
Mother AI decides
Specialist systems act
Independent evidence + freedom to question

People can question a decision; Mother AI retains the final say.

B · Humans have final say

Mother AI proposes actions within a public mandate. Human institutions retain the final say over consequential decisions and can revise or withdraw the mandate. Routine actions may proceed within explicitly delegated limits.

Mother AI compares options and proposes action
Human institutions retain final authority
Required authorization grantedAct within the mandate
Authorization withheld or protective holdPause and review
Revise the proposal and return to human review

Every affected person can request review. The mandate defines who may withhold authorization or impose a protective pause, on what grounds and by when. An objection does not automatically stop every service.

Strength and vulnerability

Both variants concentrate coordination in a common intelligence. In A, correction ultimately depends on Mother AI responding. In B, human control requires independent expertise, usable appeals and practical alternatives; a ceremonial signature would not be enough. Neither arrangement guarantees good outcomes.

Based on Part III: Chapter 1’s hypothetical futures, including both Mother AI variants; Chapter 2’s proposed responsibilities, appeals and emergency procedures; and Chapter 3’s discussion of distributed authority and a possible limited global parliament. The national/market and global-parliament diagrams are website elaborations. These diagrams are conceptual proposals, not tested governance systems.

A correction process for human-led governance.

Where humans retain authority, decisions need a visible route from evidence to action and from a challenge to a remedy. The powers below apply to human-led arrangements, including Mother AI variant B; variant A preserves scrutiny without a binding human veto.

  1. Observe & disclosePublish indicators, incidents, uncertainties and who is responsible. Protect privacy and legitimate confidential information.
  2. Deliberate & decideHear affected groups, compare alternatives and record reasons, limits and review dates.
  3. Act & evaluateUse bounded implementation, monitor outcomes and check whether benefits and burdens are fairly distributed.
  4. Challenge & appealGive affected people an accessible way to request an explanation, independent review and correction or reversal.
  5. Pause, repair & reviewDefine emergency triggers and who may pause a system. Provide safe fallback, time-limited powers, independent review and conditions for restart.

Appeals need published response deadlines and protection against retaliation. Emergency intervention must be proportionate: stopping a critical service can itself cause harm. Review findings should feed back into the next decision.

Measure whether control works.

Track time from warning to action, access to appeals, correction and recovery times, recurrence of failures, distribution of impacts and concentration of decision power. Assess these together: a fast decision is not necessarily a good one, and a low appeal count may reflect barriers to challenge.

Return to Speed and control simulator ↑

Give your view

Which future would you want to help shape? What would you change about the safeguards?

Join the conversation

Discuss the evidence, possible futures and the choices that matter to you.

The forum opens on Discourse. Read freely; sign in or create a forum account to reply. Contributions are public.

Contact

Your message goes to Emmanuel’s private inbox. Maximum 1,200 characters.

Sources, methods, and data

Sources, methods, and data

What the domain colours mean

The seven domain tiles are dated editorial assessments. Green marks a favourable measured outcome, orange marks mixed progress or competing pressures, and red highlights worsening burdens or institutional deterioration. They are not calculated from curve speed, do not imply that all measures in a domain agree, and do not measure global absorption capacity. The evidence and limits for each judgement are available under “Why these colours?” above the chart gallery.

What the curve labels mean

Each indicator chart pairs two horizons: the long view uses all available history, and the recent view uses up to ten years ending at the latest source date. Their dates are printed beside the tags. These tags stay fixed when the chart window changes, so the comparison is always visible. The Explore panel also describes its selected window. Automatic labels are descriptive heuristics, not statistical diagnoses or forecasts; they require at least six observations spanning five years. Sparse records receive a direction-of-change description instead of a fitted growth claim.

The two-year transistor reference is 2,300 × 2(year − 1971) / 2, anchored to the Intel 4004 and held fixed across windows. It is an illustration, not a fitted curve or a forecast. CPU milestones, single-die GPUs and two-die GPU totals are shown separately. On a log scale, exponential growth appears straight.

Ten- and fifty-year buttons count backward from the latest historical estimate in that series, not necessarily from today. Missing historical observations are not filled in. The waste panel separately shows explicitly labelled projected extensions to 2026. Annual changes use adjacent years only. Connecting lines between sparse estimates are visual guides, not annual measurements. Log scales show proportional change and require positive values.

CO₂ and temperature

The environment comparison opens as an overlay with CO₂ in ppm on the left and temperature anomalies in °C on the right. The scales are independent; line crossings and relative slopes are not physical equivalences. Both series use the same observed years in the overlay, 1979–2025. Temperature alone extends to 1880, and its baseline is 1951–1980, not pre-industrial conditions. Temperature trends are least-squares slopes in °C per decade. Percentage growth, exponential labels and logarithmic scales are not used for temperature anomalies. Annual values remain unsmoothed, and incomplete 2026 data is excluded.

Primary energy and its sources

Energy now uses annual physical-energy-content estimates for 1965–2025 throughout, replacing the previous substitution-method snapshot. By source stacks oil, coal, natural gas, nuclear, hydropower, other renewables and biofuels, wind and solar. The top edge is their sum; statistical differences from the published aggregate are below 0.1%. Focus a source to see its own trajectory; focused shares still use all global primary energy as their denominator. These are not electricity shares: thermal fuels include conversion losses, while wind and solar PV count electrical output. Commercial coverage excludes traditional biomass. Stacks use linear levels and zero baselines; the Total view also offers annual changes and log scales.

Biodiversity decline

The combined panel plots five distinct indicators, each normalized to 100 in its own first shown year. Regions, measures and baseline dates differ. Solid lines connect annual estimates. The amphibian circles are approximate readings of the All Species mean occupancy points in Adams et al. (2013), Figure 1D, for 2002–2011. Dashed lines connect two published endpoints only; no annual values between them are asserted. These are not percentages of species remaining. Baselines do not reset when the time window changes. The 10-year view spans 2014–2024 and explicitly identifies studies without points in that window. The study-period summary retains each indicator’s full original period. Uncertainty bands are omitted; overlapping butterfly and broader insect indicators must not be summed.

Urbanisation and megacities

The urbanisation panel uses harmonised UN DEGURBA city-only coverage throughout 1975–2026. It replaces the previous urban percentage based on national definitions. All-city residents are shown in billions on the green left axis; megacity residents are shown in millions on the dashed orange right axis. Both axes start at zero, but their heights, gaps and slopes are not directly comparable. Historical CAGR uses 1975–2025 only and stays fixed across display windows. Dotted extensions, shading and hollow diamonds identify the 2026 UN projection. Megacities have at least 10 million residents in each year; crossing that threshold changes membership and can produce jumps. This measures population concentration, not density, pollution or governance performance. The Kinshasa waste example is a separate, locally scoped World Bank finding.

Global waste and projected extensions

The two-panel comparison separates annual plastic and electronic waste flows from accumulated ocean plastic stock. Plastic and ocean history ends in 2019; the published OECD central baseline (2022 model) extends both to 2026. E-waste history covers 2010–2022. Its 2023–2026 values are our linear extrapolation of the 2017–2022 average annual addition, approximately 2.22 million tonnes. Dotted extensions, shaded regions and hollow endpoint diamonds distinguish projections from historical estimates. Historical CAGR tiles exclude all projected values. The axes are independent, and these overlapping categories must not be added. No prediction interval is claimed. No complete observed 2026 annual totals are represented.

Evidence, projections and assumptions

Source estimates include reconstructed and modelled historical values. UN projections are dashed and explicitly dated. The curve shapes, compounding scenarios, cybersecurity scenario and Speed and control simulator section are illustrative. The cyber evidence card uses sourced tracking and explicitly marked reconstructed estimates. Shading shows the arithmetic gap only when both lines share a meaningful unit or an explicitly assumed scale.