# Method note — Energy decency through the building (OneDpe)

**Document for the press.** Edition of 15 October 2026. Counts established on the full dump of the
public energy performance certificate (EPC, French DPE) database of ADEME, the French environment
agency, **dump of 25 August 2026**, processed locally.
Primary sources consulted on **12 October 2026** — each source carries its date in § 11. Contact:
Steven Annonziata — presse@onedpe.fr

> This note is the reference for every number published by the study: press page, press releases,
> articles and figures. Every number in those texts is found here, or in our data files, available on
> request. Three kinds of figure coexist and do not mix: **counts measured** in the
> ADEME database (§ 2 to § 6), **measured minimums** for social housing (§ 7) and **estimates**
> scaled to the housing stock (§ 8). Our headlines and standfirsts never carry an estimate,
> and the study's questions are never added together.

---

## 1. What the study measures

### 1.1 The simulated rule

The study measures a provision of the French housing bill, the **projet de loi relance et
décentralisation du logement**, in the version **adopted by the Senate on 8 July 2026** (adopted
text no. 157, accelerated procedure). Its article 6 adds to article 6 of Law no. 89-462 of 6 July
1989 a section III that opens: "By way of derogation from II of this article, the home is considered
decent in terms of energy performance where at least one of the following conditions is met". Its
point 6°, from amendment no. 380 rect. ter, covers the home "located in a collective residential
building whose energy performance certificate referred to in article L. 126-31 of the Construction
and Housing Code establishes that the building reaches the level of energy performance required for
the home under II of this article" (our translation).

Our reading: **the home's own EPC still exists, and the home is considered decent in terms of energy
performance if its building's collective EPC reaches the required level**. The building's EPC is a
second route; it does not replace the home's class, which remains that of its own certificate.

### 1.2 This is not the law

It is a bill going back and forth between the two chambers. The economic affairs committee of the
National Assembly **rewrote** this provision on 22 September 2026 (text no. 3188); corrections have
been announced for the floor debate, which is not on the agenda as of 12 October 2026. Our study
measures the rule "if the building's EPC reaches the level, all its homes are considered decent". It
says what that rule would do, if the bill is adopted in that version; it gives no opinion on the
reform.

Two vocabulary rules follow. We write "considered decent in terms of energy performance", never
"deemed decent". And decency only concerns let homes: for an owner-occupied home we write "whose
building reaches the required level".

### 1.3 The timetable applied

In mainland France: a G-rated home has not been decent since 1 January 2025; F from 1 January 2028; E
from 1 January 2034; D or better, decent with no date. Hence three horizons:

| Horizon | Homes concerned (class of their own EPC) | Building that "reaches the required level" | Condition on the building's EPC |
|---|---|---|---|
| 1 January 2027 | G | class F or better | EPC in the database, as is |
| 1 January 2028 | F or G | class E or better | still valid on 1 January 2028 |
| 1 January 2034 | E, F or G | class D or better | **unchanged EPC** (§ 5.3) |

All classes for these three horizons are **recalculated under the 1.7 factor (applicable from 1
January 2027)**, for **all** existing EPCs (§ 4). That last point is an **assumption of the study**,
not a rule of law: the ministry announced, in a press release of 31 August 2026, that the update of
existing EPCs would be free and immediate, and we assume it will have been made for every EPC at all
three horizons.

---

## 2. Data and population

**Source.** ADEME database of EPCs for existing homes (since July 2021), full dump of 25 August 2026,
Licence Ouverte 2.0, loaded into a local database. The calculation chain is versioned and available on request.

**Population.** EPCs that are neither of the old method (2012), deactivated nor being deleted,
issued since 1 July 2021, still valid on 25 August 2026, excluding commercial premises: **16,889,683
valid EPCs**. We keep flats and buildings: 11,460,130 flat EPCs and **430,120 building EPCs**. Both
sets still cover the whole of France and every construction period: the scope of published counts,
below, restricts them.

**Flat de-duplication.** The same flat can carry several valid EPCs. We keep one per key "same
address × same address complement × same floor area (to 0.1 m²) × same typology", the most recent
(issue date, then number). **10,354,488 flats** remain, of which 1,398,873 rated E, 436,352 rated F
and 213,069 rated G on their published class, i.e. 2,048,294 flats rated E, F or G: the set from
which classes are recalculated (§ 4).

**Scope of published counts.** Mainland France; flats whose construction period is not after 2012;
building subject to the collective EPC obligation of article L. 126-31 (§ 3.3). Counts cover **assessed
flats** in the database, not the housing stock, and the EPC does not say whether the home is let.

---

## 3. Finding each flat's building

### 3.1 Buildings (partition and rule D2)

The 430,120 building EPCs are grouped into **362,420 buildings** by a strict partition: two EPCs
belong to the same building if they are linked by a replacement chain, or if they carry the same
address and the same signature "floor area × number of flats × number of storeys". A building's class
is that of its **most recent EPC** (rule D2).

### 3.2 The three blocks of flats

| Block | Flats | How the building is found |
|---|---|---|
| **(a)** generated units (flat EPCs derived from the building's collective EPC) carrying a building number | 438,408 | the building EPC **declared** as parent (rule C1); if the building carries several generating parents, a unit is kept only if its declared parent is the building's most recent EPC (rule C2: 40,424 units removed) |
| **(a')** units "generated from building EPC data" with no building number | 3,678 | by address, counted as generated |
| **(b)** flats assessed individually | 1,606,208 | by address; an address carrying several buildings is **not assessable** |

What happens to the E, F or G flats of each block:

| Status | (a) | (a') | (b) |
|---|---:|---:|---:|
| no building EPC found | 22,688 | 1,135 | 1,336,485 |
| address carrying several buildings (not assessable) | — | 1,483 | 56,557 |
| building outside the scope of article L. 126-31 | 1,837 | 22 | 20,212 |

Also set aside: 370 flats in overseas territories, 2,861 flats built after 2012 and the 40,424
duplicates of rule C2. **Of the 1,606,208 E, F or G flats assessed individually, 1,336,485 have no
building EPC found at their address in the database.**

**Consequence of rule C1.** A generated unit takes the class of the building EPC declared as its
parent, which is not always the building's most recent EPC, whereas a flat assessed individually
takes the class of the most recent EPC (D2). The same building can therefore appear in two groups at
the same horizon, and **building counts do not add up** (§ 5.2).

### 3.3 Eligible buildings (rule D3)

A building enters the study only if it falls under the collective EPC obligation of article L.
126-31: building permit before 1 January 2013 (read from the construction period), more than two
units (a building whose number of units is unknown is kept), mainland France. That obligation has
applied since 2024 to buildings of more than 200 units, since 2025 to those of 50 to 200 units, and
since 2026 to those of 50 units or fewer.

**283,825 eligible buildings** carry a building EPC in the database. On their most recent EPC and under
the 1.7 factor, **16.4% are class E, 5.6% class F and 2.0% class G** (46,681, 15,910 and 5,779
buildings); under the 1.9 factor, 6.6% were class F and 2.8% class G (18,810 and 7,981).

A building EPC is **usable** for a flat when the building is found, unique at the address and eligible.

---

## 4. Classes: harmonisation to the 1.9 and 1.7 factors

**Why.** The database mixes EPCs issued under the 2.3 factor with EPCs issued or recalculated under
the 1.9 factor (order of 13 August 2025). The study's horizons are read under the 1.7 factor, for
all existing EPCs (assumption, § 1.3): every EPC, for a flat or a building, is therefore
reclassified.

**Method.** Moving from one factor to another uses an identity on the published consumption figures,
checked on the EPCs for which ADEME kept both states (table below):

1. bring to the 1.9 factor: for EPCs still at 2.3, EP_1.9 = EP − (4/13) × (EP − EF), where EP is primary
   energy and EF final energy consumption (5 uses, per m²); EPCs already recalculated by ADEME or issued
   at 1.9 are taken as they are;
2. move to the 1.7 factor by the same identity: EP_1.7 = EP_1.9 − (2/9) × (EP_1.9 − EF);
3. reclassify with the statutory scale (floor area, altitude, climate zone), **leaving greenhouse
   gas emissions unchanged**: the class remains the worse of the two scores. For 414,881 EPCs, the
   class under the 1.7 factor stays that of the 1.9 factor because it is set by emissions
   (consumption alone would give a better class).

**Regimes found** across 2,478,414 EPCs (430,120 for buildings and 2,048,294 E, F or G flats):
2,151,832 recalculated by ADEME at 1.9; 311,783 issued at 1.9; 14,762 rebuilt from 2.3; 23 that cannot
be rebuilt and 4 without electricity, kept as they are; 10 with no primary energy, kept at their
published class.

**The two checks that could have blocked publication** (thresholds written before measuring):

| Check | Result | Threshold |
|---|---|---|
| The scale returns the published class from the published consumption | 2,469,279 out of 2,478,404, i.e. 99.632% | 99.5% |
| On EPCs for which ADEME kept both states, the 2.3 → 1.9 rebuild returns the consumption to within 1 kWh/m² | 2,150,146 out of 2,151,755, i.e. 99.925% | — |
| … and the class recalculated by ADEME | 2,140,653 out of 2,151,755, i.e. 99.484% | 99% |

Both checks passed. Classes under the 1.7 factor are **harmonised** classes: not the ones re-issued
EPCs will display, but the ones the same method gives when applied to all.

---

## 5. The three horizons

### 5.1 On 1 January 2027: back on the market

Of **161,537 flats rated G** under the 1.7 factor, **27,007** have a usable building EPC (16.7%). Of
these, **19,057** are in a building of class F or better: they would be considered decent in terms
of energy performance thanks to their building's EPC, and could return to the rental market. They
are in **10,724 buildings**. The other 7,950 are in a building itself rated G. For the remaining
134,530 (83.3%), no usable collective EPC was found in the database (§ 3.2: building not found at
the address, address carrying several buildings, building outside the scope of article L. 126-31).

**Reading under the 1.9 factor (control figure, not used in the press release).** Had the rule applied on the date of the dump,
under the 1.9 factor then in force, the count would be **22,949** G-rated flats (12,676 buildings), out
of 33,673 G-rated flats with a usable building EPC and 208,175 G-rated flats. The public figure is the
1.7 one, because the rule could not come into force before 2027.

### 5.2 On 1 January 2028: F stops being decent

Of **530,212 flats rated F or G** under the 1.7 factor, **119,213** have a usable building EPC. They
split as follows:

- building class E or better, EPC still valid on 1 January 2028: **54,783** flats (46.0%, **23,621**
  buildings), considered decent in 2028 thanks to the building;
- building class F: **52,087** flats (43.7%) in **7,476** buildings, which would need works before 1
  January 2028 for these homes to stay considered decent through the building. Of these, **6,815**
  G-rated flats are considered decent from 2027 thanks to the building (§ 5.1), and **45,272**
  F-rated flats are decent in their own right until 2028;
- building class G: **12,343** flats (10.4%, 2,536 buildings): the rule changes nothing for them.

**Building counts do not add up.** A building can appear in two of these groups: a generated unit is
attached to the building EPC declared as its parent (C1), a flat assessed individually to the
building's most recent EPC (D2). Hence 23,621 + 7,476 + 2,536 = 33,633 buildings, for 33,613
assessable buildings.

### 5.3 On 1 January 2034: E stops being decent, with an unchanged EPC

Of **1,708,572 flats rated E, F or G** under the 1.7 factor, **510,862** have a usable building EPC:

- building class D or better: **111,633** flats (21.9%, **35,291** buildings), considered decent in 2034
  thanks to the building;
- building class E: **307,685** flats (60.2%) in **28,336** buildings, which would need works before
  1 January 2034 for the same reason. Of these, **34,689** F or G-rated flats are considered decent
  from 2028 to 2033 thanks to the building;
- building class F or G: 91,544 flats (17.9%): the rule changes nothing for them.

Nearly three times as many flats rated E, F or G are in a class E building (307,685) as in a
building thanks to which they would be considered decent (111,633).

**The "unchanged EPC" assumption.** The collective EPC must be renewed or updated every ten years
(article L. 126-31): some of the building EPCs in the database will therefore have been redone by
2034. We assume the building redoes its collective EPC at the same class. A building that carries
out works leaves this count. **Variant**: keeping only building EPCs still valid on 1 January 2034,
the count of flats considered decent thanks to the building is 86,326 (28,569 buildings).

### 5.4 The overall table, by building class

Assessed flats with a usable building EPC, classes under the 1.7 factor:

| Class of the building's EPC | 2027: G-rated flats | 2028: F or G-rated flats | 2034: E, F or G-rated flats |
|---|---:|---:|---:|
| A to D | 4,845 | 20,094 | **111,633** |
| E | 7,397 | 34,689 | **307,685** |
| F | 6,815 | **52,087** | 74,557 |
| G | 7,950 | 12,343 | 16,987 |
| **Total** | **27,007** | **119,213** | **510,862** |

Reading: in 2027, G-rated flats in buildings of class A to F (19,057) are considered decent thanks
to the building; in 2028, F or G-rated flats in buildings of class A to E (54,783); in 2034, E, F or
G-rated flats in buildings of class A to D (111,633). Class F buildings in 2028 and class E
buildings in 2034 are those that would need works for their homes to stay considered decent through
the building. Columns do not add up: the same flat is counted at every horizon where its class is
concerned.

Detail by building class and by flat class:

| Flat \ building | A | B | C | D | E | F | G |
|---|---:|---:|---:|---:|---:|---:|---:|
| G | 4 | 68 | 982 | 3,791 | 7,397 | 6,815 | 7,950 |
| F | 10 | 135 | 2,495 | 12,609 | 27,292 | 45,272 | 4,393 |
| E | 42 | 532 | 12,000 | 78,965 | 272,996 | 22,470 | 4,644 |

### 5.5 The composition of each result, by block

Each count brings together the three blocks of § 3.2:

| Result | Total | (a) generated units, building number | (a') generated units, no number | (b) assessed individually | Share of generated units (a) |
|---|---:|---:|---:|---:|---:|
| Back on the market on 1 January 2027 | 19,057 | 5,451 | 11 | 13,595 | 28.6% |
| Considered decent in 2028 thanks to the building | 54,783 | 20,415 | 76 | 34,292 | 37.3% |
| Building class F: works before 2028 (F or G-rated flats) | 52,087 | 38,413 | 41 | 13,633 | 73.7% |
| Considered decent in 2034 thanks to the building | 111,633 | 48,835 | 343 | 62,455 | 43.7% |
| Building class E: works before 2034 (E, F or G-rated flats) | 307,685 | 247,148 | 505 | 60,032 | 80.3% |
| Building class G in 2028 (the rule changes nothing) | 12,343 | 5,942 | 0 | 6,401 | 48.1% |

Reading: among those who gain from the rule, flats assessed individually dominate (71.3% of the
19,057); on the works side, units generated from the building's EPC do (73.7% of the 52,087, 80.3%
of the 307,685). A count of homes on the works side therefore weighs first the number of flats in
the buildings concerned, which explains the gap between social housing shares by homes and by
buildings (§ 7.2).

---

## 6. Coverage: what limits everything else

The rule can only work if the building has a collective EPC. In the database, that is rarely the
case: **of 161,537 G-rated flats, only 27,007 have a usable building EPC found in the database
(17%)**, and 1,336,485 of the 1,606,208 E, F or G flats assessed individually have no building EPC
found at their address. The others mix addresses carrying several buildings, buildings outside the
scope of article L. 126-31, unmatched addresses and collective EPCs carried out but missing from the
database. Most buildings do not yet have a collective EPC in the database, despite the obligation (§
3.3). The real effect of the rule therefore depends first on how that obligation is applied: this is
what the estimates in § 8 address.

---

## 7. Social housing: measured minimums

### 7.1 Method

Each building concerned is matched against the **register of rental homes of social landlords (RPLS)**
of the French statistical service for the environment and housing (SDES), stock on 1 January 2025,
released on 20 January 2026 (5,416,826 homes):

1. by address, with three keys tried in order (strict, without repetition index, by municipality);
2. for an unmatched building, by an RPLS point 30 m away or less **on the same street** and, if the
   EPC carries a number, at the same number. Proximity alone is rejected: a placebo (points shifted by
   300 m) gives 9.3% false matches at 30 m, against 0.03% for the rule used;
3. through the **national address base** (BAN): both addresses, geocoded to the house number with a
   score of at least 0.6, must carry the same address identifier. On 970 control buildings already
   matched by address, the BAN returns the same address in 952 cases (98.1%).

A building is called **fully social** when its collective homes in RPLS amount to at least 80% of its
flats.

### 7.2 Results (measured minimums, with the BAN)

**By number of assessed homes:**

| Group | Flats | In a building with some social housing | In a fully social building |
|---|---:|---:|---:|
| Back on the market on 1 January 2027 | 19,057 | at least 23.0% | at least 12.6% |
| Considered decent in 2028 thanks to the building | 54,783 | at least 30.7% | at least 14.6% |
| Building class F: works before 2028 | 52,087 | at least 54.9% | at least 26.9% |
| Considered decent in 2034 thanks to the building | 111,633 | at least 38.2% | at least 17.2% |
| Building class E: works before 2034 | 307,685 | at least 57.0% | at least 22.6% |

Without the BAN step, the same shares are 21.1%, 28.5%, 50.8%, 35.3% and 51.8% (building with some
social housing) and 11.6%, 13.5%, 24.5%, 15.9% and 20.3% (fully social).

**By number of buildings:**

| Group | Buildings | With some social housing | Fully social |
|---|---:|---:|---:|
| Back on the market on 1 January 2027 | 10,724 | at least 18.9% (2,028) | at least 11.3% (1,212) |
| Considered decent in 2028 thanks to the building | 23,621 | at least 23.9% (5,635) | at least 13.5% (3,193) |
| Building class F: works before 2028 | 7,476 | at least 23.5% (1,760) | at least 15.3% (1,143) |
| Considered decent in 2034 thanks to the building | 35,291 | at least 31.7% (11,180) | at least 17.7% (6,237) |
| Building class E: works before 2034 | 28,336 | at least 29.8% (8,448) | at least 16.3% (4,616) |

Building counts do not add up from one row to another (rule C1, § 3.2).

**Why minimums.** Absence from the register does not prove a building is not social: matching misses
addresses, residential hostels and non-subsidised homes of semi-public companies are not in it, and
dates differ (RPLS on 1 January 2025, EPCs from 2021 to 2026). The recall of the matching, measured on
863 certainly-social addresses (EPCs carrying an RPLS number), is **78.9%** with the BAN (72.0%
without); 91.2% for the 692 addresses carrying a number (82.5% without).

**Estimated range (control figure, not used in the press release).** Dividing the measured share by the recall (78.9% for the upper
bound, 91.2% for the lower bound), the share of flats in a building with some social housing would
be 25.2% to 29.1% for the return to the market, 33.6% to 38.9% for homes considered decent in 2028,
60.2% to 69.6% for works before 2028, 41.9% to 48.4% for homes considered decent in 2034 and 62.5%
to 72.2% for works before 2034. This is an **estimate**: it assumes our buildings escape matching as
often as the control sample, 383 of whose 863 addresses are in the Morbihan département.

**Reading.** By number of buildings, the share with some social housing is comparable between those
who gain from the rule and those that would need works (from 18.9% to 31.7%). By number of assessed
homes, it exceeds half on the works side, because the social buildings concerned are large and their
units are most often generated from the building's EPC (§ 5.5); at least one home in five is in a
fully social building there. These are minimums, and how decency applies to social housing is not
covered by the study.

**SDES release rules**: no statistic by landlord; no cell of fewer than 11 RPLS homes; no list of
buildings; building-level detail never leaves our machines.

---

## 8. Estimates scaled to the private rental stock

> This whole section is an order-of-magnitude **estimate**, presented in second rank: our headlines
> and standfirsts do not carry it. The five questions do not add up (they overlap on the home side).

### 8.1 Scope and anchors

**Scope**: flats in the **private rental stock** built before 2013, mainland France, **including homes
provided free of charge**, which belong to the official SDES "private rental" scope (the variant excluding them is given in § 8.5).

**Anchors**: the housing stock by energy performance class on 1 January 2025 published by SDES and the
national energy renovation observatory (ONRE), under the 2.3 factor: private rental flats by class,
heating energy, building status (co-ownership or single ownership), co-ownership size and construction
period. These anchors are reclassified to the 1.9 and then 1.7 factors with transition matrices
**measured** on the flats assessed individually in the database, energy by energy. Anchors reclassified
under the 1.7 factor: 136,726 G-rated flats, 279,829 F, 754,677 E.

### 8.2 The three methods, and the published range

- **A, "at current collective EPC coverage"**: the anchor is multiplied by the share of assessed flats
  whose building meets the condition, **with the collective EPCs already filed on 25 August 2026**
  (share computed by construction period).
- **B and C, "if every co-ownership had its collective EPC"**: the share measured only among flats whose
  building has a usable EPC is applied, as if undiagnosed buildings resembled diagnosed ones. C corrects
  B for co-ownership size; the gap between B and C stays below 5% on every question (size does not drive
  the result). It is a **potential**, not an expected outcome.

The published range runs from A to C. **Fully social** buildings are removed from the central value; the
bounds in brackets cover the social-housing variants and, for 2034, two assumptions on E-rated flats
missing from the database.

### 8.3 Results (estimates; private rental flats built before 2013)

| Question (1.7 factor) | A, current coverage | B | C, if every co-ownership had its collective EPC | Published rounding |
|---|---|---:|---|---|
| Back on the market on 1 January 2027 (G) | 11,327 [10,483 – 12,120] | 110,035 | 111,803 [110,901 – 112,806] | ≈ 11,000 → ≈ 110,000 |
| Considered decent in 2028 thanks to the building (F or G) | 27,360 [24,763 – 29,600] | 257,474 | 265,550 [260,234 – 269,400] | ≈ 27,000 → ≈ 270,000 |
| Building class F: works before 2028 | 12,006 [11,491 – 12,543] | 106,910 | 102,058 [100,368 – 105,547] | ≈ 12,000 → ≈ 100,000 |
| Considered decent in 2034 thanks to the building, unchanged EPC (E, F or G) | 46,534 [40,010 – 53,142] | 431,591 | 453,967 [436,514 – 477,002] | ≈ 47,000 → ≈ 450,000 |
| Building class E: works before 2034 | 48,388 [44,938 – 51,846] | 440,031 | 434,643 [428,134 – 447,921] | ≈ 48,000 → ≈ 430,000 |

If every co-ownership had its collective EPC, the effect would be eight to ten times larger than at
current coverage. 2034 variant with a building EPC still valid on 1 January 2034: A 40,323, C 390,584.
Return to the market under the 1.9 factor (control figure, not used in the press release): A 13,499, C 133,113.

**Co-ownerships.** Scaled to the national register of co-ownerships (RNIC, 648,481 co-ownerships in
mainland France, of which 448,202 certainly built before 2013), the number of class F co-ownerships,
if all had their collective EPC, would be **25,000 to 35,000** (25,389 to 34,791); class E, **74,000
to 100,000** (74,063 to 101,757). This is an estimate: a registered co-ownership is not a building,
and single-owner buildings have no anchor. For reference, the measured counts: 7,476 class F
buildings and 28,336 class E buildings house at least one flat concerned. They do not compare
directly with the 15,910 and 46,681 eligible buildings rated F and E in § 3.3: those are classed by
their most recent EPC (D2), whereas a generated unit takes the class of the EPC declared as its
parent (C1), which may be older.

### 8.4 The benchmark, and the criterion revised after measurement

Before any calculation, criteria were written to decide whether the reclassified counts could be
published. The benchmark applies our exit rates from classes F and G, measured by energy, to the
official private rental stock under the 2.3 factor, and compares the result with the official figure
under the 1.9 factor (871,000 energy sieves in the private rental stock, SDES/ONRE publication).

| Criterion | Statement | Result |
|---|---|---|
| **K1** (blocking) | total within ±10% of 871,000 (783,900 to 958,100) | 856,053, i.e. −1.72%: **met** |
| **K2** (blocking, written before measuring) | exit-rate gaps by energy of the same sign for all four energies | **not met** |
| **K2 revised** (after measuring) | gap on electricity below 3 points | 1.53 points: met |
| K3 | no cell (class × stratum) below 1,000 homes, otherwise merged | merges made: G by period (1975 to 2012 grouped), G by co-ownership size (20-79 and 80 units or more grouped), F by period (1989 to 2012 grouped) |
| K4 | gap between B and C below 10% | from −4.9% to +4.8%: met |
| K5 (blocking) | the projection chain reproduces the counts of § 5 exactly | met |
| K6 (non-blocking) | estimate of E-rated flats missing from the database within ±5% by energy, checked on the class F range | not met for gas (+5.2%) and heating oil (+8.2%) |

Exit rates from classes F and G between 2.3 and 1.9, by energy: electricity 46.93% against 45.40% in the
official figure (+1.53 points); gas 4.99% against 4.33% (+0.66); heating oil 1.01% against 1.37%
(−0.36); wood and heat networks 9.11% against 10.13% (−1.02).

**What we did with the failure of K2, said plainly.** Criterion K2, written before measuring, required
our gaps to have the same sign for all four energies. They do not: two are positive, two negative. The
rule written in advance then required publishing only upper bounds under the 2.3 factor, or no
projection at all. **On 12 October 2026, after seeing this result, the publisher replaced K2 with the
criterion already used in our study on the electricity conversion factor (OneDpe, 25 August 2026): a gap on
electricity below 3 points.** The measured gap is 1.53 points. This is an **after-the-fact** change to a
publication criterion; we flag it as such. Readers who hold to the original rule should keep only the
measured counts of § 5 to § 7.

Consistency check, non-blocking: our rates give 280,000 exits from classes F and G between the 1.9 and
1.7 factors, against 275,000 (248,000 to 294,000) in our study and "about 300,000"
in the official figure.

### 8.5 Assumptions and direction of bias

- **Database coverage.** A applies the coverage of 25 August 2026, and two biases pull in opposite
  directions: individual EPCs follow sales and lettings, which may make A overstate; collective EPCs
  accumulate every week, which makes A understate for any date after the dump. The net direction is
  not measured.
- **B and C** assume undiagnosed buildings resemble diagnosed ones, at equal individual class and size.
  This cannot be tested; no penalty is known for a missing collective EPC. It is a potential.
- **Anchors**: stock on 1 January 2025 under the 2.3 factor, occupancy status on 1 January 2022; the stock
  has improved somewhat since, hence a slight overestimate.
- **Social housing**: only fully social buildings are removed; the remaining social housing makes the
  private figure slightly overstated.
- **Homes provided free of charge**: included (official scope). Without them, the return to the market
  would be 10,620 (A) to 104,828 (C).
- **E-rated flats missing from the database** (an E under the 2.3 factor that became D at 1.9 is no
  longer in the export): estimated by inverse weighting; only 2034 is affected, and the range includes it.

### 8.6 What makes the projection possible, and what limits it

What makes it possible: an official anchor by type of home and by co-ownership size, and a benchmark on
the conversion factor. What still limits it: the EPC does not say which home is let, and collective EPC
coverage is not a random draw.

### 8.7 Secondary figure: the whole multi-unit stock

All occupancy statuses together (owner-occupiers, private and social tenants), flats used as main
residences built before 2013, mainland France, the order of magnitude is about twice that of the
private rental stock: a little less for the return to the market if every co-ownership had its
collective EPC, nearly three times for 2034 at current coverage. SDES/ONRE anchors (supplementary
data on 1 January 2025), reclassified under the 1.7 factor: 261,306 G-rated flats, 607,329 F,
1,766,315 E. The benchmark specific to this scope (±10% of 1,029,000 energy sieves among flats) is
met at −1.20%. No social building is removed.

| Question (estimates, all statuses) | A, current coverage | C, if every co-ownership had its collective EPC |
|---|---:|---:|
| G-rated flats whose building is class F or better, on 1 January 2027 | 23,591 | 215,694 |
| F or G-rated flats whose building is class E or better, in 2028 | 65,191 | 565,358 |
| Building class F: works before 2028 | 25,421 | 208,025 |
| E, F or G-rated flats whose building is class D or better, in 2034, unchanged EPC | 130,942 | 1,067,699 (≈ 1.07 million) |
| Building class E: works before 2034 | 116,526 | 966,076 |

For an owner-occupied home we speak of neither decency nor market: we speak of a home whose building
reaches the required level.

---

## 9. The limits, as we declare them

1. **This is not the law.** A bill still going between the chambers, whose provision has already been
   rewritten in committee at the National Assembly. If the rule changes, our counts change.
2. **Assessed, not the stock.** Counts cover the assessed flats of the ADEME database. The EPC does not
   say whether the home is let: "could return to the rental market" describes homes that would become
   lettable, not let homes.
3. **Harmonised classes.** Classes under the 1.7 factor are recalculated with a checked identity (§
   4), not read from re-issued EPCs. Applying them to all existing EPCs is an assumption of the
   study (§ 1.3).
4. **2034, unchanged EPC.** The building's class is assumed unchanged; a building that carries out works
   leaves the count.
5. **Matching.** An address carrying several buildings is not assessable (56,557 flats of block b);
   a building whose address differs from one EPC to another is not found; a generated unit takes the
   class of its declared parent, so building counts do not add up.
6. **Social housing.** Minimums; how decency applies to social housing is not covered.
7. **Estimates.** Orders of magnitude, assuming undiagnosed buildings resemble diagnosed ones, with a
   publication criterion revised after measuring (§ 8.4).
8. **What the study does not produce.** No figure by landlord, no list of buildings, no opinion on the
   reform.

---

## 10. Département annex

The file `decence-immeuble-departements.en.csv` (and its French version) gives, for each mainland
département, the measured counts for the three horizons: flats concerned, of which with a usable
building EPC, considered decent thanks to the building, and flats in a building that would need
works (columns `flats_in_class_f_buildings_works_before_2028` and
`flats_in_class_e_buildings_works_before_2034`). A cell of fewer than 11 homes is masked; a
publishable cell is masked in turn when it alone would allow the first to be recovered by
difference. No ranking, no municipality, no building. Département sums reconcile with the national
counts before masking.

---

## 11. Primary sources, and their dates

| Source | What we take from it | Reference | Date |
|---|---|---|---|
| ADEME, EPC database for existing homes | All measured counts | full dump, Licence Ouverte 2.0 | dump of 25 August 2026 |
| French Senate, housing bill, adopted text no. 157 | Article 6 of the bill, amending article 6 of Law no. 89-462 of 6 July 1989 (III, 6°): simulated rule | https://www.senat.fr/leg/tas25-157.html | adopted 8 July 2026, consulted 12 October 2026 |
| National Assembly, committee text no. 3188 and report | Rewriting of point 6° | https://www.assemblee-nationale.fr/dyn/17/textes/l17b3188_texte-adopte-commission | committee of 22 September 2026, consulted 12 October 2026 |
| Ministry, press release of 31 August 2026 | Assumption of the study: all existing EPCs updated under the 1.7 factor (free and immediate update) | press release | 31 August 2026, consulted 12 October 2026 |
| SDES/ONRE, housing stock by energy performance class on 1 January 2025 | Private rental anchors (data, tables 2, 4, 8, 9); benchmark (publication, table 2) | https://www.statistiques.developpement-durable.gouv.fr/le-parc-de-logements-par-classe-de-performance-energetique-au-1er-janvier-2025 ; https://www.statistiques.developpement-durable.gouv.fr/media/8809/download?inline ; https://www.statistiques.developpement-durable.gouv.fr/media/8813/download?inline | published November 2025, consulted 12 October 2026 |
| SDES/ONRE, supplementary data | Whole multi-unit stock anchors | https://www.statistiques.developpement-durable.gouv.fr/media/8810/download?inline | published 10 November 2025, consulted 12 October 2026 |
| SDES, register of rental homes of social landlords (RPLS) | Social housing minimums | detailed home-level data, stock on 1 January 2025 | released 20 January 2026 |
| Anah, national register of co-ownerships (RNIC) | Co-ownerships by size bracket and period | https://www.data.gouv.fr/datasets/registre-national-dimmatriculation-des-coproprietes | updated and downloaded 12 October 2026 |
| National address base (batch geocoding) | RPLS matching, step 3 | https://data.geopf.fr/geocodage/ | 12 October 2026 |

---

## 12. Where to read the study, and what to check it against

| Item | Where |
|---|---|
| Press page of the study — release, figures, département annex, this note | `onedpe.fr/en/presse/decence-energetique-par-l-immeuble-2026` |
| Calculation files | on request from presse@onedpe.fr |

**Disclosure.** Study produced by OneDpe (RIANN SAS), publisher of EPC analysis tools: the commercial interest is
declared. Right of reply: droit-de-reponse@onedpe.fr, under article 6 IV of French Law 2004-575 of
21 June 2004.

---

*OneDpe method note — Energy decency through the building, edition of 15 October 2026. Every number comes from the ADEME database dump of 25 August 2026 and
from the primary sources cited above.*