What solar displaced: addition at world scale, substitution only where demand stopped growing

Reachability swept first: yes — semantics as well as status, seven of the nine ingest routes run through fetchOwidPoints rather than reasoned about (two curl-checked only, and flagged below), and two independent routes to the same data cross-checked. All three are lessons paid for on the previous two sheets. Vintage checked: yes — Ember (2026) and other sources via OWID, lastUpdated 2026-06-30, nextUpdate 2027-06-30. A full year of stability, which makes this the safest of the three sheets to queue deep.

Segment: What did solar displace?

• Source theme(s): Batch 54 #4
• Spine role: main story
• The framing people carry: that wind and solar are replacing fossil generation — every record for renewables read as a matching retreat for coal.
• Hypothesis (from the bank): renewable growth accompanied by roughly flat fossil generation in fast-growing systems ("addition rather than substitution"), genuine fossil decline only where total demand was flat or falling, coal and gas trading places more than the fossil block shrank, and at least one prominent success decomposing into a fuel switch.
• Verdict: CONFIRMED on every sub-claim tested, and stronger than hypothesised on the first. At world scale fossil generation did not stay "roughly flat" — it grew in every window tested. Displacement happens only where demand is flat or falling. The US success is 70% a coal-to-gas switch. One of the bank's four sub-claims (the same country resolving differently by decade) was not tested.

The numbers (every row sourced — no exceptions)

All rows: Ember (2026) and other sources via OWID, lastUpdated 2026-06-30, nextUpdate 2027-06-30. Tier A throughout. Deltas are 2010 → 2023 unless stated.

The texture (what narrative flattens)

• Addition, not substitution, wherever demand is growing. World fossil generation grew in every window tested — by roughly as much as wind and solar over 2010–2023, and by 48–71% as much in later windows ending 2023 (see Open questions: the "as much" framing is window-specific). China's fossil generation grew by nearly twice its wind and solar over 2010–2023; India's by four times.
• Displacement is real — and it lives where demand stopped growing. The US (total +3.4% over thirteen years, +140 TWh), Germany (−117) and the UK (−90) all cut fossil generation. None of them did it while demand grew materially — the US was near-flat, Germany and the UK falling. Set that against China's +125% and India's +105%.
• The best-known success is mostly a fuel switch. US coal fell 1,172 TWh; gas rose 818 of it. That is a real emissions cut — gas burns at roughly half coal's carbon — and it is uncomfortable for both camps: not the renewables story its advocates tell, and not a story in which fossil held its ground.
• Germany's renewables mostly replaced nuclear, not fossil. Wind and solar +154 TWh against nuclear −133. The Energiewende's generation gains were largely spent covering the reactors it closed.
• The direction is turning — and this holds WITHOUT the provisional years. Fossil growth was roughly equal to wind and solar over 2010–2023, but in every window starting 2013 or later and ending in the settled year 2023 it runs at 48–71% of theirs. Windows ending in provisional 2024 agree. Wind and solar now outrun fossil growth globally — but fossil has still not fallen at world scale in any window tested. (2020–23 reads 0.89, inflated by rebound from the COVID fossil trough; treat it as the anomaly.)

Datoid artifact (Tier-A only)

• Nothing here is ingested yet, and the route is clean. RUN through fetchOwidPoints (126 points each): electricity-coal, electricity-gas, solar-energy-consumption, wind-generation, nuclear-energy-generation, hydropower-consumption, electricity-generation — seven. Checked by curl only, NOT run through the connector: electricity-fossil-fuels and electricity-renewables, whose index-3 columns are fossil_generation__twh and renewable_generation__twh. Run those two before relying on them. Most 301 to electricity-mix.csv?…&source=<X> and land the right single column at index 3, the same mechanism the Kaya carbon-intensity series uses.
• Two independent routes to the same data agree EXACTLY — the per-source slugs against the stacked grapher, to the hundredth of a TWh, for all six sources in World 2023. One producer throughout, which is why the identity closes here and did not on either previous sheet.
• DO NOT INGEST THE STACKED GRAPHER. electricity-prod-source-stacked has nine value columns and parseOwidCsv reads cols[3] — verified, that is other_renewables_generation__twh (geothermal, wave and tidal), 88.4 TWh for World 2023. It would ingest cleanly and chart the least relevant column in the dataset under a title promising all of them.
• No VALUE_DIVISORS entries exist for these slugs, so all would store in TWh as published — consistent with each other, which is the property #75 needs for a shared axis. Check this still holds when the seed is written; the previous sheet found three separate divisor traps.

Story justification (decide here — do NOT defer)

• Does the argument have a reveal that needs sequencing? YES — BUILD.
• Because: the reveal is withholding the fossil series. The reader first sees wind and solar climbing — the celebratory framing, and true — and only then sees fossil climbing alongside it. Shown together from the start, the two lines are a legend-reading exercise. There is a second reveal available in the US fuel switch, where the "success" decomposes into gas.

Open questions / where I'm guessing

• THE "EXACTLY EQUAL" LINE IS A WINDOW CHOICE — do not let it become the headline. Fossil ÷ wind+solar is 1.01 for 2010–2023, which makes a memorable sentence, and 0.48–0.71 in later windows ending in the settled year 2023. The durable claim is directional: fossil kept growing globally in every window, while wind and solar have recently begun outrunning it. The previous two sheets were both caught quoting the window that gave the cleanest number; this is the same temptation and it is marked here so it is not repeated.
• 2024 and 2025 may be provisional. The data runs to 2025, and Ember's most recent years are routinely revised. The 2010–2023 deltas are the conservative basis; anything leaning on 2025 should say so.
• The bank's third sub-claim is NOT tested — that "the substitution question resolves differently for the same country in different decades". It needs a per-country split by decade that was not run.
• "Displaced" is a claim about generation, not about causation. The arithmetic shows fossil falling alongside flat demand in the US, Germany and the UK; it does not show that renewables caused the decline rather than coinciding with demand saturation, efficiency, offshoring of heavy industry, or gas prices. Say "fell while", never "was replaced by", except for the within-fossil US switch, which is a same-system identity.
• solar-energy-consumption serves electricity GENERATION, despite its name. It 301s to electricity-mix?…&source=solar and its column is solar_generation__twh. Recorded because the name invites the reverse of the co2-intensity error — and because OWID also publishes solar primary energy consumption, which is a different quantity under the substitution method.
• Bioenergy may be folded into other categories before 2000, per Ember's own chart note. Irrelevant to the 2010–2023 window; relevant to any long view.