Is US public EV charging getting more reliable?

Draft v1 — 2026-08-13

The question arrived wearing one word, "reliable," and the first thing the evidence did was take that word apart. What a driver means by reliability — I plugged in and it worked — turns out to be measured by nobody in particular and claimed by everybody: satisfaction surveys measure how charging felt, port counts measure whether a charger exists, uptime dashboards measure whether a machine reported itself alive. The question journey on [Q1] is the record of forcing those three strands apart and asking which of them can actually answer the question as posed.

What each evidence class could and couldn't say. The J.D. Power EVX Public Charging series [A4] is the closest thing to a longitudinal national instrument, and it earned a narrower scope verdict: its headline is a satisfaction index (654 for DC fast charging in 2025, on a 1,000-point scale, falling), which is an experience measure, not a machine measure. But buried inside the same instrument is the series that matters — the share of owners who visited a public charger and could not charge: 20% in 2023 [A31], 19% in 2024 [A30], 14% in 2025 [A4][C1], with 60% of the remaining failures attributed to chargers that were simply out of service. The DOE AFDC data [A14][A15] earned adjacent: public ports grew from 136,307 in 2022 to 168,388 in 2023, with 4–7% quarterly growth continuing into 2024 [C4] — availability, not reliability, and the count's own base shifted definition in 2022 to include temporarily-unavailable ports. The NEVI rule [A16] is adjacent for a different reason: 23 CFR 680.116(b) requires >97% annual uptime per federally funded port [C3], which explains why networks now instrument uptime at all, but a standard is not a measurement, and its exclusions let a compliant port still strand a driver.

The as-worded strand — measured functionality — has a floor and no line. NREL's synthesis [A21] fixes a grim 2022–2023 baseline: networks self-reported 95–98% uptime while a Berkeley field study — held here as primary [P1] — found only 72.5% of Bay Area CCS connectors would deliver a two-minute charge, and UC Davis telemetry measured 7% failed sessions [C2]. ChargerHelp's 2025 report [A22] supplies one recent measured point — 71% first-attempt success, with 35% of failures on chargers that looked operational [C5] — but its full methodology sits behind a form gate, so it enters at grade C. And the probe for a public, national, longitudinal measured-uptime series came back empty for a reason that is structural, not a search failure: the federally mandated per-port uptime data flows into EV-ChART, a submission-only hub for funding recipients [A24][C6]. The zero-yield is recorded on [Q1] as a corpus gap.

That fracture forced a re-pose. "Is charging getting more reliable?" became: did the driver-facing failure rate — the share of attempts that don't end in a successful charge — measurably fall between 2023 and 2026? On that sharpened question the answer bands as follows.

Direct: yes, on the one repeated instrument that exists. The J.D. Power non-charge-visit rate fell 20% → 19% → 14% across 2023–2025 on a consistent survey methodology, each wave now held from its own primary page [A31][A30][A4][C1], and the just-released 2026 wave reportedly extends the fall to 12% [A27] — though that number rests on trade-press coverage until the bot-walled primary is captured. Adjacent: everything else points the same way — ports grew [C4], the 97% uptime rule created the instrumentation pressure [C3], satisfaction with reliability factors improved even as cost satisfaction dragged the index down [A4]. Unresolved: whether the measured failure rate — sessions, not survey recall — actually improved nationally. The 2022 baseline (72.5% functional) and the 2025 ChargerHelp point (71% first-attempt success) are not the same denominator, and the data that could settle it exists but is not public [C6].

Assessment, kept honest by keeping three things apart. Confidence: moderately high that the driver-facing failure rate genuinely fell — two independent instrument families (owner survey, field telemetry toplines) move the same direction, and the counter-narrative (composition effects from Tesla's network opening) is acknowledged in the source itself [A4] but cannot account for a six-to-eight-point fall. Coverage: partial by construction — the survey strand is covered 2023–2026, the measured strand has two incommensurable points, and the uptime strand has zero public data. Usefulness: a reader can retire the stock "one in five chargers fails" line, which is now three vintages stale; they cannot yet say "chargers are 97% reliable" — nobody outside EV-ChART can.

What reopens it. Two armed triggers: capture of the J.D. Power 2026 primary (released 2026-08-12, currently unreachable through the bot wall and un-archived), which would move the 12% figure from lead to provenance; and the ChargerHelp 2026 annual report (expected September–October 2026), the only public measured series with a repeated methodology — tracked as dormant [Q2]. A commission [K1] stands proposed to refresh the picture when the next reporting cycle lands, over this notebook's captured universe and explicitly barred from re-discovering it.