Exclusive study — wave 1
The reckoning on the electricity conversion factor
Nearly 400,000 EPC-assessed homes have left the energy-sieve bracket with no works done — and 600,000 that nothing will take out of it
On 1 January 2026 the primary-energy conversion factor for electricity used by France’s energy performance certificate — the A-to-G rating required whenever a home is sold or let — moved from 2.3 to 1.9 (order of 13 August 2025). No reckoning on that reform has been published since. We have done it, and we did not model it: 392,948 EPC-assessed homes left the energy-sieve bracket, the F and G classes, through that change alone — without a kilowatt-hour saved and without a euro of works. ADEME, the French environment agency, kept both ratings, the old one and the new one; we counted them, one by one, across 14,938,757 certificates. That is more than one EPC-assessed energy sieve in five: 22.86%. A draft order would move the same factor to 1.7 on 1 January 2027; if it were published as it stands, a further 161,479 EPC-assessed homes would leave the bracket, on a target perimeter of 9,625,591. And 604,669 that no factor will ever take out of it: it is not their consumption that rates them, it is their boiler.
Study published on · Figures frozen on · Source: OneDpe.fr
Key figures
- counted392,948
- EPC-assessed homes that left the energy-sieve bracket through the move from 2.3 to 1.9
- across the 14,938,757 certificates for which ADEME kept both states — that is 22.86% of the 1,718,766 energy sieves under the 2.3 factor, which fall to 1,325,818
- counted, then reclassified against the statutory table604,669
- EPC-assessed homes that their emissions alone keep in the bracket whatever the factor
- 58.52% of the 1,033,290 energy sieves that would survive under the 1.7 factor on the target perimeter of 9,625,591 — of which 95.46% on gas or oil and 95.44% built before 1975
- reclassified against the table, and simulated for 7.8% of cases161,479
- further exits if the draft 1.7 order were published as it stands
- 13.52% of the energy sieves remaining under the 1.9 factor, on the target perimeter of 9,625,591; 147,117 counted, 14,362 simulated
The check that could have failed
0 out of 714,457exits among energy sieves whose published carbon class is F or G
This figure measures nothing of the subject: it measures our chain. A home rated a sieve by its emissions cannot be released by a factor that only touches energy — so had our measurement been wrong, this number would have been positive. It is zero, across all 714,457 cases concerned.
« 604,669 EPC-assessed homes that this factor will never release, because it is not their consumption that rates them, it is their boiler. A conversion factor does not decarbonise a boiler. »
The copy, the method, the data
The full analysis does not cut well from the bottom. Rather than let it be trimmed badly, the copy is here — free to reuse, including commercially, with the credit “Source: OneDpe.fr”. For anyone who would rather check the count than reuse it, the method note gives the perimeter, the keys, the conventions and the limits in full. And for anyone who would rather do their own local angle, the data are served as a spreadsheet file, area by area.
- Data by department, 97 rows (CSV, French headers)
The 95 mainland departments — Corsica aggregated under code 20 — plus two off-ranking rows without which the columns would not reconcile: certificates with no usable geocode, and overseas France aggregated. Per department: certificates, energy sieves under the three factors, exits of the completed step and of the conditional step, combined span, combined rate and second-step rate, ranks out of 95 with the number of ties, and sieves blocked by carbon. Three columns reconcile exactly with the figures on this page: certificates (9,625,591), sieves under the 2.3 factor (1,588,358) and combined span (555,068). The other five fall short by overseas France alone, which the canonical dataset publishes on those three columns only — by 197, 158, 99, 39 and 84 units, each bounded by its 296 sieves. Serving overseas line by line would reproduce a rate the floor forbids. No rate below that floor of 1,000 sieves under the 2.3 factor: those cells carry a text token, never a blank that would read as “zero”.
- Data by department, 97 rows (CSV, English headers)
The same file with English column names and tokens, for an international desk or an English-configured spreadsheet.
- Press release (French)
The copy as it goes to newsrooms: headline, standfirst, the mechanism in three paragraphs, the carbon ceiling, the regulatory status of the draft 1.7 order, and the attributable quotes. Free to reuse, including commercially, with the credit “Source: OneDpe.fr”.
- Press release (English)
The same release in English. Its publishable sentences are English originals, not free translations.
- Full method note (French)
Everything needed to redo the count without us, in French: the enforceable perimeter and why it differs from the full database, the six populations not to be confused, what each nature of figure covers, the rate definitions, the checks that could have failed and what they returned, the transposition to the national stock with its criterion written before the measurement, and the declared limits.
- Full method note (English)
The same method note in English.
Infographics
Free to reuse, including commercially, with the credit “Source: OneDpe.fr”. Click a thumbnail to enlarge and download — PNG 3200 px, editable SVG, greyscale proof.
Nothing was renovated: only the factor changed
A conversion factor does not decarbonise a boiler
The mechanism and the carbon floor: the same fact, seen twice
The criterion was written before the measurement
The other figures in the file
- reclassified against the statutory table156,067
- of the 604,669 homes held in the bracket by their emissions alone (25.81%) are energy sieves on carbon alone, against the statutory threshold that applies to themon the energy axis they are already E or better; 432,306 (71.89% of the 601,313 measurable) exceed that threshold by one class only
- counted, reclassified and simulated555,068
- EPC-assessed homes that leave the bracket across the full span 2.3 → 1.7target perimeter of 9,625,591 — combined rate 34.95%. A different population from that of the 392,948: on this perimeter the same first step yields 393,589 exits (24.78%), and 393,589 + 161,479 = 555,068 to the unit
- transposed — this is not a count275,000
- principal residences, if our rates are transposed to the national stock: the only figure on this page that does not count certificatesrange 248,000 – 294,000. We do not estimate: we transpose — our exit rates, measured by heating energy, applied to the energy-sieve headcounts published by the ministry’s statistical service. The denominator is theirs; only the rates are ours. The method was tested on the 2.3 → 1.9 step, already in force and whose official result is known: there it gives 650,000 against 694,000, that is −6.3%, where a rule of three misses by +29% (figure 4). The official consultation summary puts the figure at 300,000, above our upper bound: we corroborate the order of magnitude, we do not reproduce it exactly
The data
The stock of EPC-assessed energy sieves, at the three factors (target perimeter)
| Factor | Energy sieves | Share of the base | Exits at this step |
|---|---|---|---|
| 2.3 — before 1 January 2026 | 1 588 358 | 16,50 % | — |
| 1.9 — in force | 1 194 769 | 12,41 % | 393 589 · taux ① 24,78 % |
| 1.7 — draft order | 1 033 290 | 10,73 % | 161 479 · taux ② 13,52 % |
Target perimeter: 9,625,591 EPC-assessed homes. Rate definitions — rate ① = exits 2.3 → 1.9 over energy sieves under the 2.3 factor; rate ② = exits 1.9 → 1.7 over energy sieves under the 1.9 factor, a denominator already thinned by the first reform; combined rate = span over energy sieves under the 2.3 factor, that is 34.95%. The 13.52% therefore does not read against the 24.78%: the second reform applies to an already skimmed stock. Check: 1,588,358 − 1,033,290 = 555,068, to the unit. The 1.7 row is conditional — the draft order has not been published in the Journal officiel.
The mechanism: exit rate by share of electricity (counted subset, 8,871,468 certificates)
| Share of electricity in final consumption | Energy sieves under 2.3 | Rate ① | Rate ② | Share of those remaining locked by carbon |
|---|---|---|---|---|
| below 25% | 682 565 | 2,31 % | 1,16 % | 85,56 % |
| 25 – 50% | 33 708 | 25,70 % | 18,03 % | 17,84 % |
| 50 – 75% | 77 472 | 32,57 % | 28,25 % | 0,74 % |
| 75 – 99% | 19 574 | 36,16 % | 31,37 % | 0,40 % |
| 100% — all-electric | 633 451 | 45,04 % | 33,37 % | 0,35 % |
The gradient is monotonic across the five bands and at both steps: nearly one all-electric energy sieve in two leaves the bracket, one low-electricity sieve in forty-three — a ratio of 19.5. The right-hand column is the bridge to the carbon ceiling: among low-electricity homes, 85.6% of the sieves that remain are held there by their carbon. The mechanism and the carbon ceiling are not two results, they are the same fact seen twice. One qualification that cuts against us: the axis is built on a quantity that overstates the electricity actually carrying the factor, which dilutes the top band — the 45.04% is a floor, and the real gradient is steeper than the one we publish. These rates do not recompose into a national rate: they are measured on the counted subset alone.
Where homes leave and where they do not — combined rate 2.3 → 1.7 by département
| Département | Combined rate 2.3 → 1.7 | of which all-electric among its energy sieves under 2.3 |
|---|---|---|
| Hérault | 66,61 % | 73,67 % |
| Aude | 62,09 % | 61,85 % |
| Bouches-du-Rhône | 59,94 % | 69,16 % |
| Var | 59,24 % | 63,76 % |
| … (intermediate ranks, not published) | … | … |
| Ardennes | 19,22 % | 20,78 % |
| Moselle | 18,37 % | 19,28 % |
| Creuse | 18,19 % | 20,10 % |
| Haute-Saône | 15,40 % | 14,53 % |
The rate varies by a factor of 4.3: in the Hérault, two EPC-assessed energy sieves in three leave the bracket across the full span; in the Haute-Saône, one in six and a half. We do not explain this map by climate, and we publish no correlation: the two columns are two series set side by side, never linked — no coefficient, no scatter plot. The mechanism is demonstrated inside one and the same pool, never between territories. Publication floor: 1,000 energy sieves under the 2.3 factor. 96 mainland départements are retained, served as 95 rows — Corsica is aggregated under code "20", counts summed and rate recomputed, at 55.58%. The eight overseas codes are set aside (138 exits in all). To that is added a mass that is not a territory and that we keep out of every ranking: 156,262 certificates with no geocoding, carrying 8,023 exits. The 96 départements cover 98.53% of the span, not all of it.
The factor takes flats out; what stays blocked is a house
| Segment | Energy sieves under 2.3 | Span 2.3 → 1.7 | Combined rate | Cohort blocked by carbon |
|---|---|---|---|---|
| Flat | 825 252 | 361 563 | 43,81 % | 245 879 |
| House | 762 980 | 193 439 | 25,35 % | 358 781 |
On our population of certificates, 65.1% of the exits across the span are flats and 34.8% houses — a split never to be written without its population: the ministry’s statistical service gives 46% / 54% on a modelled stock of principal residences, and that base declares itself biased towards sales and new lettings, two markets where flats are over-represented. But the cohort blocked by carbon flips: 59.34% houses, against 41.9% of houses in the perimeter. The "building" segment is not served here: 126 energy sieves under the 2.3 factor and 66 exits, that is 0.01% of the span — count given, rate withheld, below every publication floor.
Methodology
Published in full so anyone can check, replicate or challenge it.
Scope, definition, sources and limits
Scope
Target perimeter, worded so that a third party can replay it: EPC-assessed homes in category A, 3CL-2021 method, ADEME extract frozen at 2 June 2026, excluding superseded certificates, excluding certificates generated from a whole-building EPC, and excluding rows whose measured conversion factor does not fall back on one of the two statutory values (2.3 or 1.9) — that is 9,625,591 EPC-assessed homes, after removing 4,754,030 rows (33.1%). The headline figure rests on a DIFFERENT population: the 14,938,757 certificates for which ADEME kept both primary-energy states, that is 82.06% of the 18,203,808 rows in the dataset and 99.50% of live certificates issued between July 2021 and December 2025. On the target perimeter, the 1.9 → 1.7 step is counted on 92.165% of rows (8,871,468) and simulated on the remaining 7.8%. The departmental map retains 96 mainland départements, served as 95 rows — Corsica is aggregated under code "20", counts summed and rate recomputed — above a floor of 1,000 energy sieves under the 2.3 factor; it covers 98.53% of the span.
What the figure means
The primary-energy conversion factor for electricity is the number by which the kilowatt-hours of electricity a home consumes are multiplied to obtain the primary energy that decides its rating. A home is an energy sieve through its energy, through its emissions, or through both: the rating retains the worse of the two classes. Four natures of figure are distinguished. On this page they are written out in full next to each value; in the figures and in the method note they carry a glyph with its legend. ⬛ observed, “counted”: a count of two states kept by ADEME itself, with no reconstruction, no threshold table and no assumption. ◨ reclassified against the statutory table: a count on quantities published by ADEME, reclassified against the statutory threshold table — one reconstruction only, saying which side of the threshold the value falls; that table reproduces the published carbon class to 99.97% and the published F/G energy class to 99.98%. ⬜ simulated: reserved for the 7.8% of certificates for which ADEME never produced a state at 2.3; where both routes coexist, the simulation over-counts by 0.56%. ◇ transposed: reserved for the single national-stock figure, which is not a count — our exit rates applied to the headcounts published by the ministry’s statistical service, a method tested on the step already in force before being applied to the next one. None of these four natures applies to an amount in euros. Rate definitions: rate ① = exits 2.3 → 1.9 over energy sieves under the 2.3 factor; rate ② = exits 1.9 → 1.7 over energy sieves under the 1.9 factor; combined rate = span over energy sieves under the 2.3 factor.
Sources
- Public database of energy performance certificates, ADEME (Open Licence) — extract frozen at 2 June 2026, certificates issued since July 2021, 3CL-2021 method, category A
- Order of 31 March 2021 on the energy performance certificate — 3CL-2021 method, annex 3 (electricity conversion factor). Consolidated version as at 1 January 2027 served by Légifrance (JORFTEXT000043353335), consulted on 17 August 2026: it carries "+ 1.9 for electricity"
- Order of 13 August 2025, NOR ATDL2519132A, Journal officiel no. 0197 of 26 August 2025, text no. 9 — the most recent text amending the conversion factor (2.3 → 1.9)
- Order of 25 March 2024, annex 1 (JORFARTI000049446339), replacing annex 5 of the order of 31 March 2021 — threshold table modulated by floor area and altitude, checked cell by cell (528 cells, no divergence)
- Draft order moving the factor to 1.7, NOR VLOL2617030A — public consultation from 9 July to 6 August 2026, summary published on 13 August 2026; examined by the Conseil supérieur de l’énergie, France’s higher council for energy, on 23 July 2026 (item 18 on the agenda; only the examination is established, no vote is public). Neither the draft nor its explanatory memorandum carries a number of homes, a perimeter or a method — checked on the documents themselves
- The French ministry’s statistical service (SDES), "housing stock at 1 January 2025" release published on 12 November 2025, and the opinion of the Conseil supérieur de la construction et de l’efficacité énergétique of 24 July 2025 — official estimates for the 1.9 reform
- BDNB data published by CSTB, 2025-07-a vintage — DVF × EPC matching, single-lot transactions, for the price-per-square-metre gradient on houses only; the Notaires de France 2024 study as external benchmark
- Full method note — opposable perimeter, six populations, rate definitions, checks that could have failed, declared limits, primary sources and consultation dates: attached to the press release and available on request
Limits — what this study does NOT say
- Certificates ≠ homes, and deduplication is only partial. We remove 1,196,586 superseded certificates, including 1,153,887 proven replacements; but two independent certificates on the same home — a sale, then a new letting — are joined by no replacement chain and both remain counted. That residue is not measured. This is why we write "EPC-assessed homes" throughout, and why our figures are floors.
- The simulated share over-counts, and we do not claim to have corrected it. The quantity used to reconstruct the state at 2.3 overstates the electricity actually carrying the factor (ratio 0.9842 at the median, 0.6905 at the 5th percentile); on the 2.3 → 1.9 step the simulated route over-counts by 9.87%. That bias now affects only the 7.8% of uncovered rows. Nowhere do we write that the chain is corrected: we write that it is bounded.
- Uncovered certificates are not neutral, and no rule of three is allowed. The 3,265,051 rows outside the headline population show 15.02% of published energy sieves against 8.88% among the covered ones, and 57.46% are superseded. No scaling by dividing by the coverage rate is defensible, and we perform none.
- No comparison of headcounts with an official figure is defensible — including where it would suit us. Four simultaneous gaps: population, date, status (ex post measurement against ex ante estimate) and geography. The honest comparison is in rates, and it does not run the way one might think: the official simulation takes 17.7% of the energy sieves out among principal residences, and we count 22.86%. An adjacent and unavoidable limit: the certificate does not record occupancy status, so we cannot restrict to principal residences.
- No amount in euros outside the house segment. The price-per-square-metre gradient is measured on 825,719 house sales, and our own source forbids applying it to flats: outside Paris and the inner suburbs, a G-rated flat sells per square metre 6.2% above a D-rated one — the gradient flips. That would not be an approximation, it would be a sign error. Three amounts, calculated and correct, were withdrawn from this file on that ground alone. The gradient is not a controlled regression: it describes an observed price difference between two classes, it does not say what a change of rating would produce.
- No correlation between départements, no coefficient, no scatter plot. The share of electricity by département is measured and published as a side-by-side description. Reason: on this corpus, a territorial bias has already flipped a correlation coefficient of 0.82 into −0.32. No municipal figures either: we work only from INSEE codes, never from postcodes.
- No confidence interval, no test, no "±". Our figures are exact counts, for which these are moot; an inference would require clustering on the building, since up to 51.9% of a département’s certificates derive from a whole-building EPC. No mean emissions either: the distribution tail is pathological. We publish only medians, percentiles and counts.
- The reconstruction margin, quantified before anyone finds it for us. On the check of the first reform, 1,164 exits in the tested group are not confirmed crossing by crossing — an upper bound of 0.30% on the 392,948. 2,031 rows (0.022%) are not reproduced as F or G by our threshold table, all at 40 m² or under. And the first official ex post assessment does not yet exist — the "housing stock at 1 January 2026" release has not appeared: it may not say the same thing as we do.
Quotes
Attributable and ready to publish. For an interview, see the contact below.
« A reform took the energy-sieve rating away from nearly 400,000 EPC-assessed homes without a single wall being insulated, and nobody had done the count. We did not model it: ADEME kept both ratings, the old one and the new one, and we counted them one by one. The check that could have proved us wrong is the one on the 714,457 energy sieves that their emissions were blocking: not one left, and had our measurement been wrong, that number would not have been zero. But the most important result is not that one, it is the one that stays: 604,669 EPC-assessed homes that this factor will never release, because it is not their consumption that rates them, it is their boiler. A conversion factor does not decarbonise a boiler. We are not saying whether the reform is good; we are saying what it has done, and what it will not do. »
« The official figure and ours do not measure the same thing: theirs is an ex ante estimate on a modelled stock of principal residences, ours a count of certificates, after the fact, overseas included. The two are not comparable in headcount, and we will not compare them. What does compare is rates. The official simulation of the first reform takes 17.7% of the energy sieves out among principal residences, and we count 22.86%. That gap is not a difference in responsiveness, and we have corrected our own reading of it: at equal main heating energy our exit rates are slightly lower than the official ones — 43.42% against 45.40% on electricity. It comes entirely from composition: our energy sieves are more electric than those of the national stock. »
About OneDpe
OneDpe.fr is an independent platform of analysis tools for the French energy performance certificate (DPE). It offers free tools — certificate consistency checks, discount estimation, 3CL simulation, 2026 reform simulation, renovation works estimation — built on the analysis of several million certificates from the ADEME open data, cross-referenced with French property transaction data (DVF). OneDpe is published by RIANN SAS (Paris).
Going further
Journalist? Detailed figures, interviews and custom breakdowns: presse@onedpe.fr — reply within 24 hours.
Steven Annonziata, co-founder of OneDpe (RIANN SAS) and author of the study — available to comment, in French or English, within 24 hours.