Summary
- Arcep dated the eighth observatory 21 July; an independent report at 18:44 UTC corroborates publication inside the fixed window.
- Intervention-quality indicators now cover 46 networks representing 30% of French premises, described by Arcep as 13 million subscribers. That is the measured perimeter, not every French fibre premise or subscriber.
- In February 2026, 88% of measured critical defects in the cohort were corrected within 30 days. The rate does not cover every fibre fault in France.
- Over six months, average monthly outage reports stabilised around 0.12% and the average connection-failure rate around 6%. Networks below a 5% connection-failure rate fell from 82 to 49.
- Arcep's network-level results place Iliad and parts of portfolios acquired from Altice and Altitude below selected thresholds. Those cohort findings are not a blanket judgement on every line or customer of the named groups.
An outage-report rate uses an installed, active-service population and counts reports in a period. A connection-failure rate uses attempted customer activations and counts those that did not succeed. A household can never enter the outage denominator if the first appointment fails to connect it. Comparing the percentages without their populations would therefore turn two quality stages into a false contradiction.
The low outage-report average suggests that most active fibre lines do not generate a report in a typical month. It does not say that every report reflects a confirmed network fault or that every customer is equally represented. The 6% connection-failure average says roughly a different thing: a material minority of new activation attempts still does not complete successfully under the measure.
The remaining connection can cost more than the average one
France has already connected many straightforward premises. The addresses left in the migration can contain longer drops, blocked ducts, inaccessible buildings, complex ownership or poor legacy documentation. Arcep presents a growing share of complex remaining connections as a possible explanation for the weaker distribution. It is not proved as the sole cause.
This selection effect changes the economics. A first visit may discover civil work or permissions that were not priced into a standard appointment. A second crew or the infrastructure operator may need to intervene. Customer-support contacts multiply while revenue has not begun. The connection may be counted as failed even though the underlying area is passed by fibre.
The number of networks below a 5% failure rate fell from 82 in the previous edition to 49. That is a distribution warning, not evidence that the national average suddenly jumped by the same proportion. Network perimeters, volumes and local conditions differ, and a simple count gives each network one unit regardless of size.
The party taking the order is not always the party repairing the infrastructure
French fibre involves infrastructure operators that build and maintain passive networks, and commercial operators that sell service and send or commission installers. Subcontractors may carry out the physical connection. When a cabinet is damaged or a route is blocked, responsibility can cross those boundaries.
The customer experiences one failed appointment; the economic cost is split among repeat labour, support, wholesale processes and delayed billing. If incentives reward completed jobs without adequate quality verification, faults can be passed to the next intervention. If penalties are too blunt, operators may avoid the hardest addresses or dispute responsibility rather than solve the physical constraint.
Arcep's intervention indicators seek visibility into this handoff. The expanded 46-network perimeter covers 30% of premises and is described as 13 million subscribers. That is broader evidence than before, but it is not the whole national population and cannot be attached to a named operator's customer base as a national score.
Repair speed measures governance after a defect exists
The 88% 30-day correction rate shows that most critical defects observed in the measured cohort were repaired within the window in February. It leaves 12% outside that timing and says nothing by itself about defect severity, recurrence or the customers affected.
It also differs from both outage and connection failure. A workmanship defect can be found during an intervention-quality inspection, an outage can be reported by an active customer, and a connection attempt can fail before service starts. Each requires its own denominator, owner and clock.
Improving correction speed matters because defective cabinets and poor fibre routing can produce later faults and costly repeat visits. Prevention may require better training, audit and access control; repair requires clear assignment and capacity. The observatory provides indicators, not a complete cost ledger.
Copper withdrawal raises the continuity cost of failed activations
As copper closes in more areas, a failed fibre connection is not merely a delayed upgrade. It can become a barrier to fixed service continuity for households and businesses. That raises the value of pre-surveying complex premises, coordinating civil work and resolving ownership before the copper deadline.
The next evidence should keep the three series separate: connection-failure distribution as the remaining build changes, outage reports among active lines, and correction timing within the expanding intervention cohort. A useful improvement would be more networks returning below 5% without a deterioration in outage or repair measures.
France's fibre system can be reliable after activation and still struggle at the last connection. Arcep's figures do not demand a single league table. They demand stage-specific accountability: connect correctly, keep the line working and correct defective work before the same problem becomes the next customer's repeat visit.