The Method

Exactly how the numbers on this site are made — the sources, the rules, the math, and the mistakes we've corrected. Check us, don't trust us.

What this is

Many Minded tracks one question daily: is the world moving toward a future where every basic need is available and effectively free — and how fast? We follow 11 needs (energy, health, food & water, shelter, knowledge, connection, mobility, goods, care, security, the commons) and compound the day's verified developments into a running index. It is a coverage-momentum instrument built from curated news, not an official statistic and not a measure of how much the world changed — and this page exists so you can judge exactly how much weight to give it. The measured half of the picture, in people rather than points, is the value ledger.

Where the signal comes from

Twice a day (13:53 and 01:00 UTC), we gather headlines from 58 live-validated sources — science wires, sector trade press, world news, and regional editions across India, Africa, Southeast Asia and China — interleaved so no outlet dominates. From ~500 candidates, an AI editor (Anthropic's claude-opus-4-8, run with the editorial charter below) selects the most significant developments and assigns each:

• sector — which of the 11 needs it most affects;
• direction — forward (advances toward available/free) or friction (sets it back);
• rubric — a two-axis score: magnitude if true (1–5) and an evidence grade on the claim itself; the stored weight is derived, never assigned (see the rubric below);
• status — released (it happened) or coming (announced/in trials), with the shortest timeframe the source supports — never an invented year;
• why it matters — the consequence for abundance, not a restatement of the headline.

The charter's standing rules: landmark breakthroughs must never be crowded out by one loud sector; war, sanctions and export controls count as friction only when they materially hit a need (rated by supply/cost effect, not drama); de-escalation that restores supply counts as forward; hype, listicles and routine product PR are excluded; stories covered in recent days are not re-selected without a material new development. If the AI editor is unavailable, a crude keyword scorer takes over and we treat that as a degraded day — our own monitoring emails us when it happens.

Hosted briefs. Some signals carry a "read the brief" link to an article hosted here, produced by a three-stage clean room. First, an extractor reduces the source's reporting to a fact sheet — bare claims, numbers, dates, and the source's own caveats, in flat wording of its own (facts are free for anyone to state; the source's expression is not). Second, a separate writer composes the brief from the fact sheet alone — it never sees the source's prose, so it cannot copy or paraphrase it; briefs whose material is too thin are skipped rather than padded, and no figure, date, or quote is ever invented. Third, a fact-checker verifies the finished brief claim-by-claim against the source text: clean briefs are labeled AI-verified on the page, failures are labeled flagged with the reason, and briefs whose source text we could not fetch are labeled source-only — the label is always shown, including when the answer is unflattering. The original reporting is always named and linked from every brief, every brief's image is our own generated art, never the source's photograph, and each brief shows the live index measurements for the needs it touches. The brief is our read of why a development matters for abundance — it is not a substitute for the source's reporting, and we say so on the page.

How the index is computed

Every selected story applies its signed impact to its sector(s). A story repeated within 14 calendar days counts only once — the first time. Each sector's day-move is net points × 0.6%, clamped at ±14%; days carrying more than the ~18-story slate the scale was calibrated on are normalized down to slate-equivalent, so publishing more stories per day (a pipeline choice) cannot masquerade as the world moving faster. The overall index moves by the average of all 11 sector moves, damped ×0.6 — so breadth moves the needle and a single busy sector can't. It compounds from a base of 1000.

Because the feed deliberately surfaces mostly genuine advances, the index drifts upward almost every day. The informative number is therefore the pace versus typical: we compare any window against the median of all equal-length windows on record (sparse retrospective days excluded so history can't flatter the present). "Faster than usual" is news; "up" alone is not.

The scoring rubric

Every curated signal is scored on two independent axes, and the weight that moves the index is derived from them — never assigned directly:

• mag (1–5) — magnitude if true for the tracked needs, judged as if the claim fully holds. 5 = would bend a tracked curve; 1 = minor or incremental.
• ev — the evidence grade of the claim itself, graded on the site's standing scale.

The stored weight is w = min(mag, cap[ev]):

An unverified revolution caps low no matter how grand the press release — the cold-fusion / LK-99 base rate is the reason. Each card shows its signed points, its evidence-grade chip, and the rubric's one-line justification (the scoring receipt). Items curated before the rubric existed render as plain points. On days when the AI editor is unavailable, the keyword fallback is forbidden confidence: every fallback item is capped at w ≤ 2 and labeled "keyword-classified (no editor on duty) — provisional".

Three tiers, three different questions

Nothing on this site is scored on one number, because no one number can answer the three questions we ask. Each tier is measured differently, published differently, and fails differently.

Why the second tier had to exist. On 2026-08-05 five stories tied at weight 4: 1.9M Indian households put on a zero electricity bill — realized, permanent, compounding every day since — and PJM asking FERC for permission to curtail data centres, a filing that has changed nothing for anybody. Salience rated them the same, correctly: both deserved the day's attention. But the day then netted +47 forward against −41 friction as if the filing cancelled the households, and read as falsely neutral. A bounded attention ranking cannot carry value, so we stopped asking it to.

How reach is recorded. A story gets a reach figure only when the source states a real countable number of people — or their households, homes, vehicles, jobs — affected. Never estimated, never modelled, never inferred from market size or population. Every figure carries its source and a one-line derivation, printed on the card. Each is staged on a realization axis:

Units are never added together: households, people and patients are separate totals, and a story repeated across days is counted once. Reach records the figure; the rubric records how much to trust it — a number a company states about itself is recorded here and still carries the vendor grade. Currency figures appear only where they are arithmetic over two fetched numbers with the assumption printed alongside.

Most stories carry no reach figure, and that absence is signal, not a gap. Roughly nine in ten never get one, and whole days of genuinely good news carry none at all — a peer-reviewed paper, a model release, a price curve bending: real progress with no countable humans attached. PJM-asks-FERC gets none, and should get none. An empty ledger day is the tier refusing to manufacture a number; those days are read by salience and the coverage index instead. If empty ever started to look like failure, the pressure would be to invent figures, which would destroy the tier outright.

Why forward and friction are not commensurable. The index compounds both because coverage momentum is a real thing to measure. But most friction is potential and reversible — a filing, a proposal, a warning, an export control that may be lifted — while the biggest forward stories are realized and permanent. Subtracting one from the other loses exactly that distinction, so the dashboard now draws both series separately and the net is labelled as what it is: coverage momentum, not a measure of how much the world changed. Measured friction — a death toll, a count of people cut off — is recorded in the value ledger too, in its own column, and is never subtracted from anyone's gains.

The flare protocol

A capped 1–5 scale cannot express tail events — "unlimited free energy" would be worth more than a 5, if true. A claim whose magnitude-if-true exceeds the scale (a candidate regime break in a tracked curve: room-temperature superconductivity, net-gain grid fusion) is marked as a flare and climbs a public 4-stage ladder:

• 1 · claimed — the claim exists; news can light this stage.
• 2 · replicated — an independent second group reproduces it; news can light this stage too.
• 3 · measured — a verified observation is admitted to the cost-curve engine; only the engine's own pipeline can light stages 3–4.
• 4 · refit confirms — the curve refit shows the break.

Until stage 3, a flare's index contribution stays at its evidence-capped weight — the loudness is visual, not numerical. The flare card also shows the conditional impact, computed live from the engine's fits ("if this verifies: this gate moves from ~2031 to now"), labeled speculative. Most flares will die; the ladder and its misses stay published.

Two tiers of "days" — where dates come from

We used to print "≈X days closer" per story from a fixed 1.5-days-per-point exchange rate. That number was invented, and it's gone. The doctrine now:

• Daily signals are evidence, not date-movers. Feed cards show signed points and an evidence grade — never days. Headlines accumulate under a milestone as evidence.
• "Days" come only from engine refits. Each horizon milestone shows the fitted 80% crossing window of its tracked gates and the latest measured move — "window moved −183d at the 2026-07-25 refit" — computed from consecutive rows of the public prediction ledger, an append-only history of every changed crossing window. The ledger is the receipt: a date moves only when an observation changes a fit, and every move is dated and public. Fits resolve in whole years, so refit deltas land in ~half-year steps — that coarseness is the real resolution of the fits, not a bug.

What the percentages mean

Every "% toward free" figure on the site — the eleven needs, their sub-needs, the milestone cards, the deck's rings — is computed in one place, js/abundance.js, and read by every page. The rule below was written before anything was recomputed (2026-10-28 audit) and is applied the same way to every need. No figure was aimed at.

• Reach — the measured share of people for whom the need is actually met, or who actually use or have it. Where the need is a physical system, the measured share of the physical quantity that meets the condition (electricity generated low-carbon). Signal or infrastructure coverage is not reach, and neither is a count of countries. Those stay on this page as labeled context and never enter a %. The UHC service-coverage index is used, but it is an index, not a share of people, and says so.
• Cost — how far the price of the good itself, to the user, has fallen. A $0 price at the point of use scores 100. Otherwise it is log10(first price ÷ latest price) ÷ 2, clamped to 0–100, so a 100× drop scores 100. These do not count: the price of an input or component (memory is not a phone), a provider's own cost (disk storage is not the reader's price), and a spending series (spend per person is quantity × price). Every cost series shows its first year, because where a series happens to start is arbitrary and changes the score.
• One input per axis. Each sub-need uses the single most direct eligible input on each axis. Nothing is averaged, because averaging let extra rows pad the old scores. A $0 good delivered over the internet is in hand for the people who are online, so its reach is internet use. That pairing is the same everywhere it applies.
• Score. With both axes: reach × cost ÷ 100, the share of people for whom it is both in hand and near-free. With one axis: that axis, labeled reach-only or cost-only. With neither: the old editorial placeholder, labeled editorial. Those placeholders are unsourced and are kept, not tuned.
• Need and composite. A need is the unweighted mean of its sub-needs, and the composite is the unweighted mean of the 11 needs. No source justifies any weight, so there are none. Each is also shown measured-only: the mean of its measured sub-needs, or nothing if it has none.
• Milestones. These use the same rule. Capability (the METR task horizon) and API token prices are context and never enter a %. A window is labeled engine only when the engine fitted it on a series whose gate is the milestone's end state and which measures what the % measures. Every other window is editorial, hand-set, and is marked ✎ wherever it shows. Fits are deterministic (no random draws), so the same data always gives the same window.
• Rounding happens once, at display.

The audit — every input, every %

These tables are rendered from js/abundance.js and the engine as the page loads, so they cannot drift from what the site shows. Roles: reach and cost enter the score, context is shown and sourced but never scored, and editorial marks an unsourced placeholder.

Milestones

The 11 needs and their sub-needs

Inputs that were wrong or ineligible in the previous revision

Back data

The index history before 2026-06-21 is retrospective: 401 landmark events of 2025–2026 (dates and sources verified after the fact) merged under the live record. Live data always wins; retrospective days are marked in the data. They exist so the chart carries real history — not to manufacture a trend, which is also why they are excluded from the "typical pace" baseline.

Verification & corrections

Every displayed number is re-verified against primary sources on a rolling basis. The 2026-07-10 pass (three independent research agents) confirmed most figures and corrected these, live the same day:

2026-10-28 — the percentages were placeholders. Until today the "% toward free" figures were hand-typed estimates (our own research notes called them "illustrative … placeholders, not claims"), and FREE INTELLIGENCE showed a literal 62% on a hand-set 2026–2030 window. They are now computed from sourced reach and cost data (see "What the percentages mean"); the intelligence window is the engine's own 80% band, and a new FREE INFORMATION milestone tracks the measured ITU offline share.

2026-10-28 (later) — information measured as effective access. Information access had been the mean of coverage, personal internet use (73.6%), literacy, open-access share (50%) and storage cost: 82%. FREE INFORMATION was gated on "≤5% offline" (2043–2050). Use and paywall share measure uptake and the scholarly record, not whether a person can read the open record for free. Information access is now coverage + literacy + the $0 price + storage cost: 96%. The window is the engine's fit of the uncovered share. FREE INTELLIGENCE counts coverage and the $0 free tiers instead of internet use (79% → 90%, window unchanged). KNOWLEDGE and CONNECTION measure their formerly editorial sub-needs the same way.

2026-10-28 (audit) — one rule, applied to every need. An audit of the previous revision found that the figures for information, intelligence, knowledge and connection had been pushed up by the method. Signal coverage (96%) was counted as reach. Every $0 price added a 100 inside an arithmetic mean. A count of economies and a count of countries were counted as shares of people. Component prices (DRAM, disk, battery packs), a spending series (health spend), a generator's cost (solar LCOE) and a robot's price were counted as the price of the need. The estimate-grade AI price series and the METR capability series entered a %. The rule is now the one above, written before any recompute. Before → after: FREE INFORMATION 96% → 74%, window 2027–2033 (uncovered share, rule of thumb) → 2043–2050 (offline share to ITU's ≥95% "universal" bar). FREE INTELLIGENCE 90% → 74%, window 2027–2029 (API price + METR) → 2043–2050 (offline share). FREE ENERGY 53% → 49%. LABOR, OPTIONAL 35% → 36%. KNOWLEDGE 94% → 77%. CONNECTION 93% → 76%. HEALTH 30% → 35%. SECURITY 38% → 30%. Composite 43% → 39% (measured sub-needs only: 60%). Every input, its role and its source are in the audit tables above. The full record is research/audit/abundance-audit-2026-10-28.md.

When we're wrong, the correction ships and gets listed here. Known-stale items on the watch list right now: the 2026 Fusion Industry Association report (superseding our 2025 investment figures) and the mid-2026 Khanmigo RCT results.

Rules we hold ourselves to

• Friction is counted, displayed, and moves the index down — a tracker that only goes up is marketing.
• Self-reported claims are labeled ("84% of fusion companies say grid power before 2040 — their own survey answer, not a measured curve").
• Exponential cost-collapse is claimed only where measured (solar, batteries, compute, intelligence) — never blanket-extrapolated.
• Speculative windows are labeled speculative; no invented ETAs, ever.
• Political floors are named as political (you can't 3D-print away a zoning code).

The 58 sources

• New Atlas
• IEEE Spectrum
• Ars Technica
• Ars Technica · Science
• TechCrunch
• The Verge
• ScienceDaily
• MIT Tech Review
• Quanta
• Hacker News
• Guardian · Science
• Phys.org
• Science News
• Nature
• Singularity Hub
• ScienceAlert
• The Decoder
• The Register
• Tom's Hardware
• Google DeepMind
• CleanTechnica
• pv magazine
• Heatmap
• Canary Media
• Utility Dive
• EIA · Today in Energy
• Carbon Brief
• KFF Health News
• Fierce Biotech
• STAT
• Nature Medicine
• Lifespan.io
• AgFunder News
• Food Dive
• Green Queen
• Electrek
• SpaceNews
• The Robot Report
• Phys.org · Materials
• Construction Dive
• EFF Deeplinks
• Krebs on Security
• BBC World
• Hechinger Report
• Rest of World
• Mongabay
• Manufacturing Dive
• HumanProgress
• SCMP Tech
• CnEVPost
• Google News · Friction
• Google News · Zoning
• Google News · Care
• Google News · Shelter
• Google News · Frontier
• Google News · India
• Google News · Africa
• Google News · SE Asia

Feeds are health-checked weekly (HTTP, parseability, freshness) by automated monitoring; dead or stale feeds get replaced, not silently dropped.

The data is open

The cost-curve engine is public: the cost curves — live fitted price curves with intervals, gate crossings, the prediction ledger, calibration, and our backtested misses.

Every day's curated signals — sector, direction, impact, status, source link — are publicly readable, and the whole dashboard is open without an account. If you want the raw daily data for research or a build of your own, come inside and ask on the line.