# Method note — The reckoning on the electricity conversion factor (OneDpe)

**Document for the press.** Figures established on the public database of energy performance
certificates published by ADEME, the French environment agency — **extract frozen at 2 June 2026** ·
Primary sources consulted on **17 and 18 August 2026** — each source carries its own date in § 13 ·
Contact: Steven Annonziata — presse@onedpe.fr

> This note exists for two reasons. A figure without its denominator is not verifiable, and **seven
> populations** coexist in this file: they are not interchangeable. And because we give our own
> bounds — that is what makes the rest believable.

*Quotations from French orders, consultation documents and official statistical publications are
OneDpe's translations. The French original is given alongside every quotation that carries an
argument, because in at least one case the point turns on which word is* not *in the sentence.*

---

## 1. The English lexicon, frozen

This note and the release that accompanies it are a **mirror of the French file, not a translation
of it**. The French source sheet imposes some twenty formulations, several of which are the safety
mechanism of the whole dossier rather than matters of style. It has no English column. This section
is that column. **One translation per term. Do not introduce synonyms.**

Terms marked ⚓ are inherited unchanged from the house style sheet frozen on 6 August 2026
(`2026-08-06_media-qa-and-style-sheet-EN.md`) and are already in use in four earlier OneDpe
releases. Changing one of them now would split our own published record.

| French | English — use this | Never |
|---|---|---|
| passoire énergétique / thermique | ⚓ **energy sieve** — at first mention, gloss as *the F and G classes* | thermal sieve, energy colander, draughty home, energy-inefficient home, "worst-performing property" |
| logements diagnostiqués | **EPC-assessed homes** · long form at first mention: *homes holding an EPC* | homes, dwellings, properties, housing, "French homes", "the stock" |
| sort du statut de passoire / quitte le statut de passoire | **leaves the energy-sieve bracket** · variant: *drops out of the energy-sieve bracket* | is renovated, is upgraded, is improved, is retrofitted, **is no longer an energy sieve** |
| statut de passoire | **the energy-sieve bracket** | energy-sieve status — *permitted only inside the translation of an official French quotation, where it renders their wording, not ours* |
| nous dénombrons | **we count** · *we counted them, one by one* | we estimate, we model, we project, we assess, we reckon |
| dénombrement | **a count** | an estimate, a projection, a modelling |
| dénombrement exhaustif | **a full count** — gloss once: *every row tested, no sampling* | a survey, a sample, an extrapolation |
| reconstitué au barème réglementaire | **reclassified against the statutory threshold table** | simulated, estimated, predicted, modelled, "reconstituted" |
| recalculé au seuil qui est le sien | **recomputed against the threshold that applies to it** | recalculated at the 70 threshold |
| simulé | **simulated** — and only for the 7.8% of certificates ADEME does not cover | using it for the file as a whole |
| barème réglementaire | **the statutory threshold table** | the scale, the grid, the benchmark, the schedule |
| au barème 2,3 / 1,9 / 1,7 | **under the 2.3 / 1.9 / 1.7 factor** | at the 2.3 scale, on the 2.3 grid *(collides with the threshold table)* |
| au seuil réglementaire qui est le sien | **against the statutory threshold that applies to it** | above 70 kg, the 70 kg threshold, "the 70 benchmark" |
| écart de prix associé à l'étiquette | **the price gap associated with the rating** | value created, value unlocked, what the market will pay, worth, gain, uplift |
| déclaré « confort d'été insuffisant » par son propre DPE | ⚓ **rated "insufficient" for summer comfort by its own EPC** | measured overheating, will overheat, uninhabitable in summer, any temperature |
| situé en tissu urbain compact | **located in compact urban fabric** | will be X °C hotter, "heat trap", any temperature |
| coefficient de conversion en énergie primaire de l'électricité | **the primary-energy conversion factor for electricity** at first mention, then **the conversion factor**, then **the factor** | the PEF unglossed, "the coefficient" |
| projet d'arrêté | **the draft order** | the order (unqualified), the forthcoming order, the 2027 order |
| si le texte est publié en l'état | **if the text were published as it stands** | when the order comes into force, "the order of 1 January 2027" |
| arrêté du 13 août 2025 | ⚓ **the order of 13 August 2025** | decree, act, statutory instrument |
| la réforme des petites surfaces | **the small-floor-area reform** | the small-surface reform, "the EPC reform" |
| la réforme du coefficient | **the conversion-factor reform** | "the EPC reform" in the singular — *conflating the two is the commonest mistake on this subject* |
| voie exacte (S4) | **the counted subset** | the exact route, the exact path, the hard data |
| arc / arc complet 2,3 → 1,7 | **the full span 2.3 → 1.7** · variant: *the two steps combined* | the arc, the trajectory, the journey |
| pas ① / pas ② | **step ① / step ②** | phase, wave, tranche |
| sortie(s) | **exit(s)** — *homes leaving the bracket* | escapes, releases, removals, "wins" |
| taux de sortie | **the exit rate** | the success rate, the hit rate |
| périmètre cible | **the target perimeter** | the sample, the scope, the catchment |
| assiette | **the base** | the basis, the pot |
| cohorte GES | **the carbon cohort** — gloss once: *carbon here means greenhouse-gas emissions* | the CO₂ cohort, the emissions group |
| étiquette | **the rating** · *rated F*, *rated G* | the label, the sticker, the badge |
| classe | **class** · *class E*, *by one class* | grade, band, tier |
| plancher *(our figure is one)* | **a floor** | a minimum, a conservative estimate, a lower bound |
| plancher de publication | **the publication threshold** | a floor *(reserved for the sense above)* |
| non géocodé | **not geocoded** | unlocated, missing, "unknown *département*" |
| maille résiduelle | **a residual bucket** | a region, a zone, an "other" category |
| redevient louable à la relocation | **becomes lettable again when it is next let** | tenants' leases will change on 1 January, "can be rented out again from January" |
| le gel des loyers est levé pour l'avenir | **the rent freeze is lifted for the future only** | the freeze is lifted, back rent can be recovered, "a catch-up is allowed" |
| estimation ex ante sur parc modélisé | **an ex ante estimate on a modelled housing stock** | the government got it wrong by X homes |
| le gaz et le fioul verrouillent | **gas and oil lock** the energy sieves that remain | the government forgot carbon |
| travaux | ⚓ **works** | renovation works, refurbishment, "a retrofit" *(when rendering the legal and cost category)* |
| sans un kWh économisé, sans un euro de travaux | **without a kilowatt-hour saved and without a euro of works** | — *(factual, authorised, headline included)* |
| DPE | ⚓ **EPC** / energy performance certificate | DPE unglossed |
| diagnostiqueur | ⚓ **assessor** | surveyor *(a UK surveyor is a chartered surveyor — a different profession)* |
| ADEME | ⚓ **ADEME, the French environment agency** (first mention) | ADEME unglossed |
| département | ⚓ ***département*** (italic, kept) | county, department |
| France métropolitaine | ⚓ **mainland France** | metropolitan France |
| outre-mer / DROM | ⚓ **the French overseas *départements*** | overseas territories |
| maison / appartement | **house** / **flat** | single-family home, apartment |
| énergie primaire / finale | **primary energy** / **final energy** | delivered energy, end-use energy |
| Md€ | **€X billion** | X bn€, X Md€, X B€ |
| transposer / transposition au parc | **transpose** / **the transposition to the national stock** — gloss once: *the official denominator borrowed, our rates lent* | extrapolate, gross up, scale up, project, estimate, "our estimate of the stock" |
| banc d'essai | **the back-test** — gloss once: *the step whose official result is already published* | validation, calibration, sanity check |
| règle de trois | **a rule of three** — *our overall rate multiplied by the national stock* | simple scaling, pro rata *(too neutral: this is the method that fails)* |
| énergie principale de chauffage | **main heating energy** | fuel type, heating fuel |
| notre panel n'est pas plus réactif, il est plus électrique | **our panel is not more responsive, it is more electric** | "29% more responsive" *(false — an artefact of composition)* |
| parc / parc national | **the national housing stock** — always modelled, always labelled as such | "the stock" *(reserved: never for our counted population)* |

### 1.1 The eight that are the safety mechanism, and what each one holds

1. **"EPC-assessed homes"**, never "homes". It is this qualifier that carries the whole
   *certificates ≠ homes* limit (§ 11, limit 1). English has no everyday word meaning "a home that
   has been the subject of a certificate", so the compound has to be built and then held. The long
   form *homes holding an EPC* is used at first mention and in the opposable perimeter definition.
2. **"leaves the energy-sieve bracket"**, never "is upgraded" and — the trap specific to English —
   never **"is no longer an energy sieve"**. That last phrasing is the most natural English sentence
   available and it is exactly the one the French lexicon bans, because it describes the building
   rather than its classification. *Bracket* is chosen over *status* deliberately: an English reader
   hears the sense of a tax bracket, a band you move between without yourself changing, which is
   precisely the fact being reported.
3. **"we count"**, never "we estimate". This is the dividing line of the entire dossier. Every time
   the verb slips, the headline figure becomes a model, and the one thing that distinguishes it
   disappears.
4. **"reclassified against the statutory threshold table"** for ◨, **"simulated"** for ⬜, and never
   one for the other. ◨ applies a public rule to public data; ⬜ rebuilds a quantity ADEME never
   published. Letting "simulated" spread beyond the 7.8% contaminates the other 92.2%.
5. **"against the statutory threshold that applies to it"**, never "above 70 kg". 70 is the
   threshold only for homes over 40 m² outside the altitude regime: **20.20%** of the perimeter has
   a modulated threshold (§ 8).
6. **"the price gap associated with the rating"**, never "value created" and never "what the market
   will pay". The gradient is not causal. It says at what median price per square metre the houses
   in each class have sold; it does not say what a change of rating would produce.
7. **'rated "insufficient" for summer comfort by its own EPC'**, never "overheating". That indicator
   is a declared field of the certificate, not a measurement. *(It carries no figure in this release
   — the block is measured but out of scope here. The wording is frozen now so that it cannot drift
   if an annex ships later.)*
8. **"energy sieve"**, held exactly. This is the one calque with a published record behind it: it
   appears in four earlier OneDpe releases in English. A newly invented variant would not merely
   read oddly, it would fork our own corpus — and a badly chosen calque has contaminated a whole
   press chain before now.

### 1.2 Six places where English forced a choice French did not

Declared here rather than discovered by a reader comparing the two files.

1. **"Barème" splits in two.** French uses one word for the *statutory threshold table* (the
   surface- and altitude-modulated thresholds) and for the *conversion factor in force* ("passoires
   au barème 2,3"). English cannot: the first is **the statutory threshold table**, the second is
   **under the 2.3 factor**. Collapsing them would make the sentence "reclassified against the
   statutory threshold table under the 2.3 factor" unreadable — and, worse, would suggest the
   threshold table itself changed with the factor. It did not.
2. **"Voie exacte" becomes a population, not a route.** English has no comfortable noun for
   *voie* here. **"The counted subset"** names the population instead of the channel, which is in
   fact more precise: what defines S4 is that both states exist, so the step is counted.
3. **"Arc" becomes "span".** English *arc* reads as narrative. **"The full span 2.3 → 1.7"**, or
   "the two steps combined".
4. **"Plancher" disambiguates — and this is a gain.** French uses one word for *our figure is a
   floor* and for *the publication threshold below which no rate is served*. English separates them
   (**a floor** / **the publication threshold**), and the mapping is fixed above so the gain is not
   thrown away by a synonym.
5. **A French verbatim cannot survive translation.** The source sheet's most exacting point is that
   the official summary says *"estimé à 300 000 logements"* and that the word *environ* ("around")
   **is not in that sentence** — it sits elsewhere in the file, attached to the factor. In English
   both would translate to "estimated at" and "around", and the distinction evaporates. Every
   quotation is therefore labelled a OneDpe translation, and the French original is printed
   alongside the ones that carry an argument (§ 13).
6. **The regulatory conditional needs more scaffolding in English.** French marks it
   morphologically ("*ferait sortir*"). English "would" alone starts to read as prediction after two
   or three sentences, so the explicit clause **"if the text were published as it stands"** is
   repeated more often here than in the French file. That is not padding; it is the same guard
   carried by different means.

### 1.3 Numbers and typography

British English. **Decimal point, comma thousands separator** — `392,948`, `22.86%`, `1.211`. No
space before `%`. Dates as `25 August 2026`. Currency symbol first: `€190.00`, `€1.95 billion`,
`€470/m²`. Units as `kWh/m²/year` and `kg CO₂/m²/year`. The three objects of this note are written
**2.3**, **1.9** and **1.7** — never 2,3 / 1,9 / 1,7.

---

## 2. The perimeter, in one opposable sentence

Two perimeters are in use, and only one is the "target" perimeter. Here it is, drafted so that a
third party can replay it:

> *Homes holding an EPC in category A, 3CL-2021 calculation method, ADEME extract frozen at 2 June
> 2026, excluding superseded certificates, excluding certificates generated from a building-level
> EPC, and excluding rows whose measured conversion factor does not fall back on one of the two
> official values (2.3 or 1.9).*

The last clause is the one people forget to declare. Without it, a contradictor replaying the
definition does not find our headcount. It is in the code, so it is in the definition.

**What the perimeter removes, and why:**

| Exclusion | Rows | Reason |
|---|---:|---|
| Per-vector sum not reconciling with the published primary energy | *(outside the base)* | 11.8% of category A; **328,210** of these rows have **final** energy above their **primary** energy, which is physically impossible |
| Superseded certificates | **1,196,586** | A replaced certificate is no longer the home's EPC. **1,153,887** have their replacement in the same base: the double count is proved, not assumed |
| Certificates generated from a building-level EPC | **3,557,444** | The energy-sieve rate of the removed stratum is **3.83%** against 17.61% elsewhere: we are removing the **least** energy-sieve sub-population |
| **Total removed** | **4,754,030 (33.1%)** | Published base: **9,625,591** |

**On superseded certificates, this is not our choice, it is ADEME's.** It regenerated the state at
2.3 for only **3.0%** of superseded certificates, against **99.50%** of live ones. And article 3 of
the draft 1.7 order covers only certificates "in force". On the exclusion of certificates generated
from a building-level EPC, we align with the ministry's statistical service, which itself excludes
from its cleaned base the unvisited flats derived from building-level EPCs (36.1% of that base).

⚠️ **A reconciliation we have not done, and which we declare**: the territorial tables are
established on the ADEME extract frozen at 2 June 2026; **they have not been reconciled with the
database served by our online tools**.

---

## 3. The seven populations

| Code | Population | Headcount | What it is for |
|---|---|---:|---|
| **S1** | Certificates for which ADEME kept **both** primary-energy states | **14,938,757** | The **headline figure** (392,948) and nothing else |
| S2 | The complete dataset | 18,203,808 | The coverage denominator: **82.06%** |
| **S3** | **Target perimeter — the published one** | **9,625,591** | The span, the 1.9 → 1.7 step, the carbon cohort, every cross-tabulation |
| S4 | S3 ∩ S1 — the "counted subset" | 8,871,468 = **92.165%** of S3 | The **counted** part of the 1.9 → 1.7 step, the mechanism, the value block |
| S5 | Admission variant of the target perimeter | 10,102,806 | **Declared sensitivity** (§ 9). Never a headline, never a figure |
| S6 | Wide base | 14,379,621 | Validity checks only |
| S7 | Residential certificates **in force at 18 August 2026** (category A, 3CL-2021 method, mainland) | **16,264,575** | **Coverage of the national stock**, and nothing else — never a denominator for a result |

**No other surface exists** in this file.

⚠️ **What our base covers of the national stock — our measurement, not a borrowing.**
**16,264,575** residential certificates (category A, 3CL-2021 method, mainland) are **in force at
18 August 2026**, against about **30.9 million principal residences** (the ministry's statistical
service): a raw ratio of **53%**. That ratio is an **upper bound**, and it has to be written as
one: **several certificates can bear on the same dwelling**, so the share of the stock actually
covered is **lower**. The wording we use is therefore: *at best one dwelling in two, and probably
fewer*. It is also the structural reason why no rule of three on our sample is defensible (limit 6,
§ 9 bis).

⛔ **Two borrowings we refuse.** *(a)* The sentence "fewer than one dwelling in two holds a valid
EPC" circulates attributed to the Cour des comptes, France's supreme audit institution: it is **not
verifiable** in that report, which gives a monthly volume of certificates and «*plus de 32 millions
de logements privés en métropole, dont 5,5 millions de passoires*» — "more than 32 million private
dwellings in mainland France, of which 5.5 million energy sieves" — **not a coverage rate**
(checked on 18 August 2026). We publish **our** measurement, and we attribute it to no one else.
*(b)* An approximate deduplication by (municipality × floor area × method) was attempted: its
result **is not a count of dwellings** — the key is **municipal**, it is not an address, and the
method variable does not carry the calculation route. It is published nowhere, and it must not
be.

⚠️ **The confusion not to make.** **392,948** bears on **S1** (14,938,757 certificates): that is the
reform **already in force**. **555,068** bears on **S3** (9,625,591 EPC-assessed homes): that is the
full span 2.3 → 1.7, the reform in force **plus** the draft order. The two numbers do not subtract,
do not add and do not compare without their population.

---

## 4. The three markers, and their exact scope

This is the principal asset of the work, and it is typographically visible in all our publications.
Three levels, never two.

| Marker | What it means | What we allow ourselves to write |
|---|---|---|
| ⬛ **OBSERVED** | A count of two states **kept by ADEME itself**. No reconstruction, no threshold table, no assumption | "we count", "ADEME kept both states" |
| ◨ **RECLASSIFIED** | 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. Reproduces the published class to **99.97%** (carbon) and **99.98%** (F/G energy) | "reclassified against the statutory threshold table", "recomputed against the threshold that applies to it" |
| ⬜ **SIMULATED** | Reconstruction of the consumption at 2.3 from the declared electricity. Bias measured, **never nil** | "simulated" — and **only** for the 7.8% of certificates ADEME does not cover |

⛔ **Scope rule.** These markers qualify **quantities read or recomputed on the ADEME data**: counts,
the rates derived from them, medians, percentiles. They apply **to no amount in euros** (a
`floor area × Δ(€/m²)` sum is not a reading of data), **to no comparison with an external source**,
and **to no projection**. Putting ⬛ on anything else would import the credibility of the count onto
a quantity that has not earned it.

### 4.1 The counted part, and the simulated part

| Quantity | **Counted** part **of rows** | **Simulated** part **of rows** |
|---|---:|---:|
| Headline figure (392,948, reform 2.3 → 1.9, population S1) | **100%** ⬛ | 0% |
| Step 1.9 → 1.7 on the target perimeter (161,479) | **147,117** ◨, produced by the **92.165%** of covered rows | **14,362** ⬜, produced by the **7.8%** that are not |

⚠️ **92.165% and 7.8% are shares of ROWS in the perimeter, not shares of exits.** The two series do
not divide into one another, and this row is not a percentage breakdown of the 161,479 exits: the
counted exits are those the covered rows produce, the simulated exits those the others produce.

**The gap between the two routes is measured, not assumed**: where they coexist — a full count
across the 8,871,468 rows of S4 — the simulated route finds 147,941 exits against 147,117 counted,
that is **+0.56%**. The simulation **over-counts**. Our span is therefore a floor on that side too.

---

## 5. Rate definitions, to be copied as they stand

- **rate ①** = exits (2.3 → 1.9) ÷ energy sieves **under the 2.3 factor**
- **rate ②** = exits (1.9 → 1.7) ÷ energy sieves **under the 1.9 factor** — a denominator **already
  lightened** by the first reform
- **combined rate** = the full span ÷ energy sieves **under the 2.3 factor**

⚠️ Never set 24.78% and 13.52% side by side without these definitions: a hurried reader will
conclude that the second reform "does half as much". It does not — it applies to a stock already
skimmed.

---

## 6. The span, at the three factors (target perimeter S3)

| Quantity | Value | Composition |
|---|---:|---|
| Perimeter stock | 9,625,591 | — |
| Energy sieves under the **2.3** factor | **1,588,358** | 16.50% of the base |
| Energy sieves under the **1.9** factor | **1,194,769** | 12.41% |
| Energy sieves under the **1.7** factor | **1,033,290** | 10.73% |
| Exits ① (2.3 → 1.9) | **393,589** | 342,027 ⬛ + 51,562 ⬜ · rate ① **24.78%** |
| Exits ② (1.9 → 1.7) | **161,479** | 147,117 ◨ + 14,362 ⬜ · rate ② **13.52%** |
| **Combined span (2.3 → 1.7)** | **555,068** | combined rate **34.95%** |

**Three integrity checks pass**: the stock of energy sieves is **strictly decreasing** across the
three factors; `energy sieves at 2.3 − energy sieves at 1.7 = the combined span` **to the unit**
(1,588,358 − 1,033,290 = 555,068); and across 9,139,965 rows tested, **exactly one** "entry" into
the bracket is observed — identified, documented as a data anomaly, and kept as it stands. It is
that single row which proves the check is not tautological.

---

## 7. The checks that could have failed

**These carry the argument, not the absence of error.** A check that cannot fail is not a check: it
weakens the others by contagion.

| Check | Result | Marker |
|---|---:|:--:|
| **Zero exits among energy sieves blocked by their carbon.** Of the 1,718,766 energy sieves under the 2.3 factor, those whose **published** carbon class is F (523,010) or G (191,447) — **714,457** — record exactly zero exits | **0 / 714,457** | ⬛ |
| **The 1.9 → 1.7 step is not a tautology**: **7,397** energy sieves do cross the E/F boundary on the energy axis between 1.9 and 1.7 and **do not leave** the bracket, held by their carbon alone | 7,397 | ◨ |
| **The headline figure is invariant to the arrival state**: 392,948 measured against the published rating **and** 392,948 measured against the arrival column ADEME kept, 0 entries in both cases | **0 divergence across 14,938,757 rows** | ⬛ |
| **Our threshold table reproduces the carbon class published by ADEME** | **615,900 / 616,093 = 99.9687%** | ◨ |
| **Our threshold table reproduces the published F/G class on the energy axis** | **99.9778%** (2,031 residuals, all at ≤ 40 m²) | ◨ |
| **The rule "the rating is the worse of the two classes"**, as the text requires | **14,938,757 / 14,939,336 = 99.9961%** | ⬛ |
| **The factor applies to electric kilowatt-hours only**: the extra primary energy is exactly 0.9 times the electricity — that is, 1.9 − 1 — at the 5th, 50th **and** 95th percentiles | **0.9000** across 603,948 homes | ⬛ |
| **The 579 rating anomalies are outside the headline figure**: 579 rows carry the arrival state without the departure state; they are exactly the 579 rows with a divergent rating | **0 within the 392,948** | ⬛ |

⚠️ **What is NOT a check.** No downgrade is observed: no home enters the energy-sieve bracket because
of the factor. That is what arithmetic requires — primary energy can only fall when the factor
falls, and emissions do not move — and what the data confirm without exception. We put it that way,
and never as "a perfect check".

### 7.1 The robustness check on contamination by the small-floor-area reform

Two reforms follow one another within the period, and conflating them would be the major error on
this subject. We separate them, and we have tested that separation three ways.

1. **On the population that is uncontaminable by construction** — certificates issued from 1 July
   2024, born under the new small-floor-area scale — the exit rate is **24.12%** (135,325 / 561,043),
   **higher** than the 22.25% of the earlier period.
2. **The "clean figure" we refused to publish.** Removing the 1,506,571 certificates covered by both
   sets of columns gives 287,940. That is an **over-correction**: those rows are small floor areas
   (1,708,196 of 1,708,611 are ≤ 40 m², **none** above), therefore structurally more electric and
   more responsive. Excluding them removes 105,008 exits, of which **101,489 (98.87%)** are confirmed
   crossing by crossing at constant thresholds. That number stays here, in the method note; it is
   not a result.
3. **The before/after July 2024 gap is not a method effect.** The same lift appears in the floor-area
   bands the reform never touched (41-60 m²: 23.41 → 25.98%; 61-100 m²: 17.05 → 18.66%; > 100 m²:
   12.78 → 14.04%). The two direct standardisations converge: floor-area structure explains only 0.3
   to 0.4 of a point. It is a **period effect**.

**What the small-floor-area reform did, separately**: at least **69,049** EPC-assessed homes left the
bracket before the conversion factor came into play — a floor, measured on the 1,506,571 certificates
covered twice over. ⛔ That number **does not compare in headcount** with the "160,000 **flats in
principal residences that left the bracket through the July 2024 revision of the small-floor-area
thresholds alone, that is 62% of the exits in that segment**" published by the ministry's statistical
service: different base, different perimeter, different segment. *(The full qualification is
mandatory: the same document carries a second "160,000", which designates the private rental stock,
all causes combined.)*

---

## 8. The carbon block: statutory threshold table, never a flat threshold

The cohort of 604,669 homes blocked by their emissions is **not** obtained by comparing emissions to
70 kg CO₂/m²/year. **70 is the threshold only for homes over 40 m² in the lowland regime.**

- The threshold is **modulated by floor area** below 40 m² (70 → 72 → 74 → 76 → 84 → 90 kg for the
  E/F boundary, from 40 down to 8 m², by linear interpolation as the text prescribes) **and by
  altitude** (above 800 m in climate zones H1b, H1c and H2d: 80 kg for E/F, 110 for F/G).
- **20.20% of the perimeter has a modulated threshold**: 1,856,501 certificates at ≤ 40 m² and
  128,340 in the altitude regime. Within the carbon cohort itself, **95,734 homes (15.83%)** do not
  have 70 as their threshold — 82,167 by floor area, 13,567 by altitude.
- The threshold values were **verified against the text** (annex 1 of the order of 25 March 2024):
  528 cells re-compared with the source, **0 divergence**.

⚠️ **Our queries follow the text, not our own engine — and this needs saying before anyone finds
it.** The order prescribes **linear interpolation** between the tabulated floor areas; the engine we
publish produces a **rounded** variant. A third party replaying the calculation from our code would
find **155,890** instead of 156,067, and **579** residual disagreements with the class ADEME
publishes instead of **193**. We publish the **interpolated** values, which follow the text, and we
give the others here.

**Two denominators coexist in this block, and we label them systematically.**

| Denominator | Headcount | What it carries |
|---|---:|---|
| The carbon cohort | **604,669** | The shares by energy vector, age and segment, and the 156,067 |
| The base for distances to threshold | **601,313** | The **median emissions of 87.9 kg**, the 432,306 (71.89%), the median distance of 17.0 kg, the shares under 5 and 10 kg |

The gap of **3,356** rows is an internal inconsistency in the dataset: their published carbon class
is F or G while their emissions are **below their own threshold**. We keep them in the cohort — it is
ADEME that classes them so — and we set them aside from the distance calculation, which would be
meaningless on them.

The implicit emission factors rebuilt from the data fall back, to the ten-thousandth, on annex 7 of
the order of 31 March 2021 as updated on 11 April 2025: gas 0.2270 · oil 0.3240 · LPG 0.2720 ·
coal 0.3850 · electricity 0.0790.

---

## 9. Perimeter sensitivity, declared and quantified

| Perimeter | Stock | ES 2.3 | ES 1.9 | ES 1.7 | ① | ② | **Span** | Combined rate |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
| **S3 — published** | 9,625,591 | 1,588,358 | 1,194,769 | 1,033,290 | 393,589 | 161,479 | **555,068** | **34.95%** |
| S5 — variant | 10,102,806 | 1,673,554 | 1,264,316 | 1,096,393 | 409,238 | 167,923 | 577,161 | 34.49% |

*(ES = energy sieves.)*

**Gap: +3.98% on the span, −0.46 of a point on the rate.** It comes **exclusively from the base**:
**zero rows** change classification between the two admission rules, across the 14,379,621 rows in
common. Variant S5 is technically better on one point — it does not require measuring a conversion
factor in order to decide membership of the perimeter — and that is why it appears here. The
**published** perimeter remains S3.

---

## 9 bis. The transposition to the national stock: method, back-test, and what it cannot do

**What we do, in one sentence.** We apply **our** exit rates, measured by **main heating energy**,
to the **energy-sieve headcounts published by the ministry's statistical service** for the same
energy (table 2 of the "stock at 1 January 2025" vintage, § 13). The denominator is **official**;
only the rates are ours. The verb is **transpose**: we do not gross our sample up towards the
national stock, we **borrow a denominator** and **lend** nothing but rates, one stratum at a time.
⛔ This is not an estimate, and the word "estimate" has no place beside these figures.

**The criterion was written before the measurement, and it was not reopened** —
`2026-08-18_critere-go-nogo-transposition.md`, dated 18 August 2026, prior to the calculation. It
required **two simultaneous conditions**: *(1)* the transposed total for step ① falls within
**625,000 – 763,000**, that is 694,000 ± 10%; *(2)* the gap on the decisive stratum —
**electricity**, which carries 84% of the official movement — stays **below 3 points**. Both are
met. Had either failed, this section would not exist and the file would have stayed with the count
alone.

### 9 bis.1 The back-test — a step whose official result is already known

The **2.3 → 1.9** step is **already in force** and the administration has published its result. The
method can therefore be tested **in public**, against a known target, before being applied to the
next step, whose result is not known.

| Item | Energy sieves at 2.3 *(official, thousands)* | Official rate | **Our rate** | Gap | **Transposed** |
|---|---:|---:|---:|---:|---:|
| Electricity | 1,282 | 45.40% | **43.42%** | **−1.98 pt** | 557 |
| Oil | 1,098 | 1.37% | **1.13%** | −0.23 pt | 12 |
| Gas | 993 | 4.33% | **4.12%** | −0.21 pt | 41 |
| Wood and other | 543 | 10.13% | **7.37%** | −2.76 pt | 40 |
| **All** | **3,915** | **17.73%** | *(not applicable)* | — | **650** |

Official target: **694–695 thousand** — 694 by differencing the two overall columns, 695 by summing the four heating-energy rows; the one-unit gap is a rounding of the source table. Our transposition: **650**, that is **−6.3%** on the 694 reading and **−6.5%** on the 695 one. ⚠️ **We publish −6.3%**, the reading closest to the overall line; the other is given here so that both can be reproduced.

⭐ **The four strata agree in the same direction — none offsets another.** That is what rules out
being "right by cancellation", a trap this file has met once already: check ② came out at
**+0.56%** in aggregate while **4.11%** of row-by-row disagreements cancelled each other. A total
that is right by cancellation is worth less than a total 6.5% low whose four strata all lean the
same way.

**Three methods, one back-test** — it is this comparison that gives the result its value, not the
result on its own:

| Candidate method | Step ① transposed *(thousands)* | Gap to target |
|---|---:|---:|
| Rule of three — our **overall** rate × the national energy-sieve stock | *(not published — see ⛔ below)* | **+29%** |
| Post-stratification by **segment** (house / flat) | *(not published)* | **+26%** |
| **Transposition by main heating energy** | **650** | **−6.5%** |

⛔ **The number we refuse to publish, and why we now know it is wrong.** A rule of three on our
overall rate (22.86%) applied to the national energy-sieve stock would give **about 890,000**
homes. The back-test says exactly how it is wrong: that method misses a **known** result by
**+29%**. It would be **our composition bias set in bold**. It stays barred from every publication
of ours. *(It appears here, in the method note, as a refused and dated figure; it is not a result,
and it is carried nowhere else.)*

### 9 bis.2 What the measurement corrects in our own file

Our drafts explained the gap between **22.86%** and **17.73%** by a greater "responsiveness" of our
panel — *29% more responsive in relative terms*. **The stratum-level measurement refutes that**: at
equal main heating energy our rates are **slightly lower** than the official ones, across all
**four** items. The overall gap comes **entirely from composition** — our energy sieves are more
electric than those of the national stock. The correct formulation is: *we are not more responsive,
we are more electric*. That correction has been carried everywhere the sentence appeared (release,
this note, objections sheet).

### 9 bis.3 Step ②, whose result is not known

Same method, same source, denominator = the "after (cep = 1.9)" column of table 2.

| Item | Energy sieves at 1.9 *(official, thousands)* | **Our rate** | **Transposed** |
|---|---:|---:|---:|
| Electricity | 700 | **32.59%** | 228 |
| Gas | 950 | **2.04%** | 19 |
| Wood and other | 488 | **4.45%** | 22 |
| Oil | 1,083 | **0.56%** | 6 |
| **All** | — | — | **275** |

**275,000** if the text were published as it stands, within a range of **248,000 to 294,000**. The
lower bound applies the criterion's ±10% tolerance downwards (247,500, rounded to 248,000); the upper bound corrects for the −6.5% bias
measured at the back-test (275 / 0.935 = 294,000). The official announcement says "**around
300,000**", **with no published method**: neither the draft order nor its explanatory memorandum
carries a headcount, a perimeter or a method (§ 13).

⛔ **The conclusion, and its imposed wording.** Our transposition **corroborates the official order
of magnitude** on the 1.7 step. Nowhere do we write that the government is wrong: we write that we
know how to **check** its order of magnitude, and that the method producing theirs has not been
published. The regulatory conditional applies throughout: the text has not been published.

**The differentiator transposes too.** Among the energy sieves that **would remain** after a move
to 1.7, **89.5%** of those heated with gas and **98.3%** of those heated with oil are held there by
their **carbon** — against **0.5%** of the electric ones. The lock is not spread evenly: it is
fossil. This is the fact behind the 604,669 cohort, seen from the side of the national stock.

### 9 bis.4 The limits of the transposition, declared

1. **Occupancy status is not recorded on the EPC.** Our rates are measured **across all occupancy
   statuses**; the official denominator is in **principal residences**. This is not repairable, and
   no weighting gets round it (limit 6).
2. **The transposition of step ② cannot be better than what it transposes.** Rate ② is ◨
   **reclassified against the statutory threshold table**, and ⬜ **simulated** for 7.8% of rows:
   the transposition **inherits** those markers. It does not become ⬛ because it is multiplied by
   an official headcount.
3. **The four gaps run in the same direction — a good sign, and also a limit.** Nothing cancels
   (§ 9 bis.1), but a bias that runs the same way **four times** is **systematic**, not random. It
   points to a difference of **field** — occupancy status, vintage of the stock, geography — and
   not to measurement scatter. The range on step ② **bounds** it; it does not remove it.
4. **The denominator is mainland.** The 97x codes are outside the official source's field: the
   rates used for the transposition are therefore measured **on the mainland field**, whereas our
   counts include the overseas *départements*. The two series do not recompose into one another.
5. **No interval.** The transposition is a weighted mean with **official weights**, redoable on a
   calculator: it is not a model, and the "no ±" rule of § 11 (limit 8) applies to it unchanged.
   The range on step ② is a **bias bound**, not a confidence interval.

⛔ **Guard-rail.** No transposition figure in a **headline**, a **standfirst** or an **e-mail
subject line**. The headline figure remains a **count**: 392,948.

---

## 10. The value block: where the €/m² gradient comes from, and what it forbids

**Exact provenance of the gradient:**

| Element | Value |
|---|---|
| Source | **BDNB, published by CSTB, France's building science centre — 2025-07-a vintage** — DVF property-transaction records and EPCs **matched**, single-lot transactions, fictitious geometries excluded. Data already published on our EPC discount page |
| Perimeter of the gradient | median price per m² by class, across **825,719 HOUSE sales** (total matching: 956,923 single-lot transactions) |
| Metric | median price per m² by class, expressed relative to class D — the Notaires convention |

The seven values: **A €2,670 · B €2,495 · C €2,328 · D €2,119 · E €1,972 · F €1,782 · G €1,649 per
m²**, that is gaps to D of +26.0% · +17.7% · +9.9% · ref. · −6.9% · −15.9% · **−22.2%**.

⚠️ **This is not a controlled hedonic regression.** Residual confounders acknowledged:
within-zone location, age, general condition. The **Notaires de France 2024** study is one, and
serves as our oracle: **−25%** for a G-rated house against a D-rated house, where we measure
**−22.2%**. Our house gradient is therefore **slightly more cautious** than the oracle.

⛔ **What our own source forbids, verbatim**: "*Flats: never publish the raw national figure
(artefact: the Paris effect flips the sign). Handle by geography.*" The fact behind the ban:
**outside Paris and the inner suburbs, a G-rated flat sells per m² 6.2% ABOVE a D-rated one**;
within Paris proper the discount exists but stays around 3%. Applying the house gradient to a flat
is not an approximation, it is a **sign error**. Three calculated and logged amounts were withdrawn
from the file on that ground alone — flats, blocks, and the total of the three segments. **No euro
on a flat, now or later**, short of going through a controlled hedonic regression, which is not our
method.

**What survives, and its population — declared under each figure, never as a total:**

| Value | Population |
|---:|---|
| **€1.95 billion** | The **114,738 houses** that leave the bracket at the 2.3 → 1.9 step, **counted subset**, average floor area 89.4 m² |
| **€190.00/m²** | The weighted average gap carried by a **house** exit, across both steps |

Three points we make ourselves.

1. **The block values the counted subset only.** Simulated exits carry **no euro**. That is cautious:
   the simulated part over-counts exits by 9.87%. It is an argument, not a silence.
2. **The €190.00/m² is not independent proof, it is a consistency check.** €190 is, to the cent, the
   difference between the median price of an E-rated house and that of an F-rated house in the
   gradient used. That the weighted average lands there is an **arithmetical consequence** of every
   exit being an F→E. **The proof of the F→E lies elsewhere**: the direct count on two **published**
   ratings gives **114,737 F→E and a single G→E** out of 114,738 house exits. What the €190.00 checks
   is the **homogeneity of the cohort**.
3. **A home rated F at 1.9 cannot reach class D at 1.7.** The step removes at most
   `1 − 1.7 / 1.9 = 10.53%` of primary energy, a bound reached at all-electric; to fall into D it
   would have to cross a gap of at least **18.8%** — and **24.2%** in the lowland reference regime.
   Arithmetic bounds the world; measurement closes our reconstruction: the class at 1.7 of the
   147,117 counted exits fits in **a single cell, E**.

⛔ **Imposed wording wherever a euro appears**: "**the price gap associated with the rating**", with
its population — counted subset, house segment — under the figure, never as a total. **The gradient
is not causal**: it describes an observed price difference between two classes, it does not say what
a change of rating would produce.

---

## 11. The limits, as we declare them

We give them before they are put to us. They are ordered by severity.

**1. Certificates ≠ homes, and deduplication is only partial.** We remove 1,196,586 superseded
certificates, of which 1,153,887 are proved replacements. But two independent EPCs on the same home
— a sale, then a re-letting — are linked by no replacement chain and **are both still counted**. That
residue is not measured. This is why we write "**EPC-assessed** homes" throughout, and why our figure
is a **floor** through four distinct channels: a restricted perimeter, the exclusion of
non-reconciling rows, partial deduplication, and over-counting in the simulated part. What we know
about **coverage** is measured and declared in § 3: **16,264,575** certificates in force against
about **30.9 million** principal residences, that is **at best** one dwelling in two — and probably
fewer, since several certificates can bear on the same dwelling.

**2. The denominator excludes 11.8% of category A, and the bias runs against us.** Those rows are set
aside because the per-vector sum does not reconcile with the published primary energy. Their
energy-sieve share is almost identical (11.50% against 10.72%) — so no selection bias on status — but
they are **mixed-fuel in 83.7% of cases against 63.5%**, therefore more often held by carbon:
excluding them removes mainly homes that **would not have left**. **Our exit rate therefore
overstates slightly.** As the bias is geographical (mixed fuel is), any gap between territories we
publish is a floor.

**3. The simulated part over-counts, and we do not claim to have corrected it.** The quantity used to
rebuild 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.

**4. Uncovered certificates are not neutral — and no rule of three is permitted.** The 3,265,051 rows
outside S1 carry 15.02% published energy sieves against 8.88% among covered ones, and 57.46% are
superseded. **No scaling by division by the coverage rate is defensible**, and we perform none. On the
target perimeter, **94.5%** of uncovered certificates were issued from 1 January 2026, born at 1.9,
for which ADEME never produced a state at 2.3: their lack of coverage is a property of the
**calendar**, not a data defect. None of their dimensions would justify a correction (energy sieves
12.31% against 12.80%, blocked by carbon 51.48% against 48.94%, median floor area 68.1 against
69.1 m²).

**5. The reconstruction margin, quantified.** On the check for 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.
On the second step, the strict definition gives 150,814 and the wide definition 151,776, that is
**+0.64%**. Threshold fragility: **4,600 exits (3.05%)** sit less than 1 kWh/m²/year below the
boundary, and 4,457 energy sieves miss it by as little — a perturbation of ±1 kWh would move the net
balance by about **143 rows (0.09%)**, against a median margin below the boundary of 20.53
kWh/m²/year. Finally, **2,031** rows (0.022%) are not reproduced in F/G by our threshold table,
**all** at ≤ 40 m².

**6. No comparison of headcounts between our COUNTS and an official figure is defensible —
including when it would suit us.** Four simultaneous gaps: population (14.9 million certificates
against about 30.9 million modelled principal residences), date (2 June 2026 against 1 January 2023
or 2025), status (an ex post measurement against an ex ante estimate), geography (overseas included
on our side, mainland only on theirs). Between counts, the honest comparison is therefore **in
rates**: the official simulation takes **17.7%** of energy sieves out among principal residences, we
count **22.86%**.

⚠️ **We have corrected our own reading of that gap, and the correction matters.** We put it down to
a greater "responsiveness" of our panel — *29% more responsive in relative terms*. **That is
false**: at equal main heating energy our rates are **slightly lower** than the official ones
(43.42% against 45.40% on electricity; same sign across all four items, § 9 bis). The overall gap
comes **entirely from composition** — our energy sieves are more electric than those of the national
stock. We therefore remain **forbidden** to present ourselves as "smaller, therefore more cautious",
and **forbidden** to apply our overall rate to the national stock by a rule of three: that number would measure nothing but our selection bias, and the back-test in § 9 bis proves it by
making that method miss a known result by +29%. The **only** comparison with the national stock we
publish is the **transposition** of § 9 bis, which borrows the official denominator one stratum at a
time. ⚠️ Adjacent limit, blocking and not circumventable: **the EPC does not record occupancy
status**, so we cannot restrict to principal residences — and that is the first likely cause of the
−6.5% measured at the back-test.

**7. The flat / house split differs from the ministry's statistical service.** On the first step it
gives **46% / 54%** (flats from 1,349,000 to 1,029,000, houses from 2,566,000 to 2,193,000) where we
give **65.1% / 34.8%**. The populations differ — a **modelled** stock of principal residences on their
side, **certificates** on ours — and that base declares itself biased towards transactions and new
lettings, two markets in which flats are over-represented. ⛔ We never write "close to two-thirds in
flats" without its population.

**8. No confidence interval, no test, no "±".** Our figures are exact counts, for which these are
beside the point. An inference would require clustering on the building: up to **51.9%** of the
certificates in a *département* derive from a building-level EPC, and that non-independence rules out
any unclustered interval.

**9. No mean and no extreme value — neither of emissions, nor of the ratio between the two
published consumptions, whose tail is pathological too.** The tail of the distribution is pathological (302
homes above 300 kg, 66 above 500, 26 above 1,000) and missing values exist. We publish only
**medians, percentiles and counts**.

**10. The dataset's internal inconsistencies, published because we looked at them.** **3,356** rows
are classed F or G on the carbon axis while their emissions are below their own threshold (§ 8).
**3,749** rows (0.0251%) have a former primary energy **below** the current one, which is physically
impossible: **none is among the exits**, the effect on the 392,948 is nil. And **338,081** covered rows
(2.26%) have no floor area recorded and were handled at the standard thresholds; they are not excluded
from the headline figure, which is a count of classes.

**11. The first official ex post assessment does not yet exist** — the "stock at 1 January 2026"
vintage has not been published — and it may not say the same thing we do. Our object is a **reckoning
at constant thresholds, reform by reform**, which that vintage does not produce: it states itself that
"interpreting the downward change remains delicate" and that "the reform of the EPC calculation
explains close to 40% of that fall".

**12. What we have not done.** No reconciliation with the database served by our online tools (§ 2).
No municipal figure — we work only from the INSEE municipality code, never from a postcode. No euro
outside the house segment. No cross-tabulation with any climate data.

**13. A contradiction with our own June 2026 study, which we are not correcting.** It put at
**722,000** the F-rated homes of the private rental stock affected by the 2028 deadline. It is
calibrated on an official private-rental-stock figure predating the factor's entry into force, and
**we do not know how to update it**: the certificate does not carry occupancy status (limit 6). We
leave it as it stands — it is a dated artefact, and correcting it without being able to recompute it
would be worse than declaring it.

---

## 12. Publication guard-rails

- **No *département* rate below 1,000 energy sieves under the 2.3 factor.** **96 *départements*** are
  retained, all mainland; the eight overseas codes are set aside (138 exits in all, 0.02% of the
  span). A missing *département* is not "barely affected": it is below the publication threshold.
- **The "not geocoded" row is not a territory.** It weighs **156,262** certificates, **8,023** exits
  (1.45% of the span) and 5,521 homes of the carbon cohort. It is **necessary to the checks** —
  without it the territorial sum no longer reproduces the national figure — and **barred from every
  ranking**. Consequence, to be written: the *département* map covers **98.53%** of the span, not all
  of it.
- **No correlation between *départements***, no `r` coefficient, no scatter plot. The share of
  electricity per *département* is **measured** and is published as a **side-by-side description**.
  Reason: on this corpus, a territorial bias has already flipped an `r` of 0.82 to −0.32.
- **No municipal figure derived from a postcode.**
- **No transposition figure in a headline, a standfirst or an e-mail subject line** (§ 9 bis). The
  transposition is published **in second place**, always accompanied by its back-test; the headline
  figure remains a count. **The verb is "transpose", never "estimate".**
- **No national-stock headcount obtained by a rule of three**, neither on our overall rate nor on a
  deduplicated one (§ 9 bis.1).
- **No euro on a flat, a block or an all-segments total** (§ 10).
- Rates are rounded to the hundredth of a point; headcounts are given to the unit and their press
  rounding is explicit ("nearly 400,000", "more than 600,000").

---

## 13. Primary sources, and their consultation dates

**Data**

- **Public database of energy performance certificates**, ADEME — Licence Ouverte, the French open
  licence. Extract frozen at **2 June 2026**. Certificates issued since July 2021, 3CL-2021 method,
  category A.
- **BDNB, published by CSTB — 2025-07-a vintage** — DVF × EPC matching, single-lot transactions, for
  the €/m² house gradient only (§ 10).
- **Notaires de France, 2024 study** — external oracle for the gradient (−25% for a G-rated house
  against a D-rated one).

**Texts**

- **Order of 31 March 2021** on the EPC (3CL-2021 method; annex 3, the conversion factor for
  electricity). **Version consolidated at 1 January 2027 as served by Légifrance**
  (`JORFTEXT000043353335`), consulted on **17 August 2026**: it carries **"+ 1.9"**.
- **Order of 13 August 2025**, NOR **ATDL2519132A**, *JORF* no. 0197 of 26 August 2025, text no. 9 —
  the most recent text amending the conversion factor (2.3 → 1.9). Consulted on 17 August 2026.
- **Order of 25 March 2024**, annex 1 (`JORFARTI000049446339`), which replaces annex 5 of the order of
  31 March 2021 — the small-floor-area threshold table. **Verified cell by cell** on 17 August 2026:
  528 cells, 0 divergence.
- **Draft 1.7 order and its explanatory memorandum**, NOR **VLOL2617030A**, public consultation page
  open from **9 July to 6 August 2026**. Consulted on 17 August 2026 on the documents themselves:
  **neither carries a number of homes, a perimeter or a method** (explanatory memorandum: 2,174
  characters; draft: 7,902).
- **Post-consultation summary**, put online on **13 August 2026**. French original: «*Cette évolution
  conduira effectivement à la sortie d'un certain nombre de logements du statut de passoires
  énergétiques (estimé à 300 000 logements), sans modification physique réelle des bâtiments ni
  réduction immédiate de leurs consommations.*» OneDpe translation: "This change will indeed lead to
  a certain number of homes leaving energy-sieve status (estimated at 300,000 homes), with no real
  physical modification of the buildings and no immediate reduction in their consumption." ⚠️ The word
  *environ* ("around") **is not in this sentence**; it appears elsewhere in the file, attached to the
  factor. 110 contributions, 108 unique, 78 unfavourable, 29 favourable. Consulted on 17 August 2026.
- **Conseil supérieur de l'énergie**, France's higher council for energy, agenda of the sitting of
  **23 July 2026**, item 18. French original: «*Projet d'arrêté modifiant le facteur de conversion de
  l'énergie finale en énergie primaire de l'électricité relatif au diagnostic de performance
  énergétique*». OneDpe translation: "Draft order amending the final-to-primary energy conversion
  factor for electricity relating to the energy performance certificate." That document carries **no
  figure** and **no vote**: only the examination is established. Consulted on **18 August 2026**.

**Official estimates for the 1.9 reform**

- **Ministerial FAQ**, published on **17 July 2025**, updated 22 July 2025, and the corresponding
  consultation summary — **initial** estimate of 850,000 homes.
- **Opinion of the Conseil supérieur de la construction et de l'efficacité énergétique of 24 July
  2025**, France's higher council for construction and energy efficiency — **unfavourable** (15
  against, 4 in favour, 10 abstentions). ⚠️ The text states, in its **observations** and not in the
  grounds of the opinion, that «*certains membres regrettent l'absence d'une étude d'impact
  socio-économique*» — "some members regret the absence of a socio-economic impact assessment";
  the Fédération française du bâtiment says so by name in the annex. **Do not attribute it to the
  Council as a finding**: two separate occurrences, one of them named, say more and check out.
  Consulted on **18 August 2026**.
- **Service de la donnée et des études statistiques**, the ministry's statistical service, "stock at
  1 January 2025" vintage, published **12 November 2025**: energy sieves among principal residences
  would fall from **3.9 to 3.2 million** (12.7% → 10.4%). Its **table 2** — energy sieves by **main
  heating energy**, principal residences, before (cep = 2.3) and after (cep = 1.9) — is the
  **denominator borrowed** by the transposition of § 9 bis: we recompute none of its values, we
  take them as they stand. Its **footnote 1** records the revision
  itself: «*l'évaluation ex-ante de la réforme avait été estimée sur la base du parc au 1ᵉʳ janvier
  2023, ce qui conduisait à environ 850 000 passoires en moins*» — OneDpe translation: "the ex ante
  evaluation of the reform had been estimated on the basis of the stock at 1 January 2023, which gave
  around 850,000 fewer energy sieves." Confirmed by the answer to written question no. 7184, published
  3 February 2026 (current estimate: around 700,000).
- **Same service, publication of 20 November 2023**: 4.8 million energy sieves (15.7%) among about 30
  million mainland principal residences at 1 January 2023 — the basis for the 850,000 — and 6.6
  million (17.8%) across the total stock.

**Two citation traps, flagged in advance**

- ⚠️ **The CSE is not the CSCEE.** The Conseil supérieur de l'**énergie** examined the draft 1.7 on 23
  July **2026** (opinion not published). The Conseil supérieur de la **construction et de l'efficacité
  énergétique** gave its opinion on 24 July **2025**, on the **1.9**. Two bodies, two texts, two
  years: conflating them would attribute to the 1.7 an opinion given on the 1.9.
- ⚠️ **A false positive to rule out.** The only 2026 order amending *an* EPC order is that of **30 March
  2026**, NOR ECOR2605189A, *JORF* no. 0098 of 25 April 2026. It updates the CO₂ content of **heating
  and cooling networks** and **does not touch** the conversion factor for electricity. Anyone
  presenting it as "the 1.7 order" has the wrong text.

⏳ **This section is perishable.** The watch covers Légifrance alone; the last *Journal officiel*
swept is no. 0190 of 15 August 2026. **The "not published" status of the 1.7 order replays in one
query** — the version of the order of 31 March 2021 consolidated at 1 January 2027 — and must be
replayed on the morning of any reissue.

---

*OneDpe method note — the reckoning on the electricity conversion factor, edition of 25 August 2026.
Prepared by OneDpe (published by RIANN SAS, Paris). No query was made against our production
databases; every figure comes from the public ADEME database and from the primary sources cited
above. Our figures are free to reuse, including commercially, with the credit "Source: OneDpe.fr".*