String-level solar health, measured
Solar Performance Cloud, the independent solar-inspection platform Ali Ahmed built at BijliBachao.pk, grades every PV string against its neighbours on the same inverter. Across 94 sites, 118 inverters and ≈4.6 MW recorded, it monitors 906 strings and scored 582 on the latest settled day — flagging 86 as critically behind their peers and 50 as measurably behind. Every figure below is measured, portfolio-level and anonymised.
Source: the Solar Performance Cloud production database at BijliBachao.pk. Scale read on 25 August 2026; scored figures are the settled day, 24 August 2026. Licensed CC BY 4.0.
The dataset
How much solar is this?
| Metric | Value |
|---|---|
| Sites monitored | 94 |
| Inverters | 118 |
| Cloud platforms unified | 7 (6 manufacturers) |
| Capacity recorded | ≈4.6 MW recorded |
| Generation tracked | 1.93 GWh |
| Days live | 207 (since 30 January 2026) |
| Readings held | 2.69 million (30-day rolling window) |
SPC monitorsthese sites; it did not install them all. Capacity is stated as “recorded” because some sites are under-recorded — the real portfolio is above the figure shown.
Three numbers, one noun
How many strings, exactly?
A commercial rooftop is dozens of independent strings, and 'how many' has three honest answers. Here are all three.
| Definition | Count |
|---|---|
| Total DC string inputs present on the hardware | 1,091 |
| Strings monitored (reported live in the 30-day window) | 906 (860 in 14 days) |
| Strings scored on the latest settled day | 582 |
The scored count excludes one platform (20 sites) onboarded the day before the read — its first settled score follows.
Peer-relative, not a ratio
How is a string graded?
Every string is compared against its neighbours on the same inverter, under the same sky — so cloud, season and temperature cancel out. 100% means a string matches its peers.
| Percentile | Performance vs peers |
|---|---|
| Weakest (5th percentile) | 0% |
| Median | 100% |
| Best (95th percentile) | 119% |
| Strings scored | 582 |
This is a peer-relative comparison, not a performance ratio — SPC has no irradiance sensors and computes no PR. A string at 0% is making no useful output relative to its neighbours; one above 100% is running ahead of them.
Named, not just alarmed
What did it flag?
Faults are classified in the IEC 62446-1 test vocabulary, so a finding points at the verification test an engineer would run.
| Category | Strings |
|---|---|
| Performance verification (Warning) — behind peers but still producing | 50 |
| Module & string power (Critical) — severely behind peers | 86 |
Open-circuit (continuity & polarity) and offline (communication & data availability) are event and connectivity states rather than settled-day scores, so they are not counted here — a point-in-time read is available on request.
Daylight-gated availability
Average string daylight availability on the settled day — computed against each site’s own sun position, not a fixed clock.
| Metric | Value |
|---|---|
| String daylight availability | 84.6% |
Data honesty
Why some numbers here are 'not available'
The hard part of monitoring third-party device data isn't finding faults — it's not crying wolf. SPC assumes the data is wrong until it earns trust.
Before a reading is allowed to influence any verdict it passes a write-gate: is the sun actually up at this site right now, computed from its own coordinates rather than a clock? Is this genuinely new data, or the same vendor snapshot replayed? Is this current physically possible for the hardware — or a failed sensor reporting something like 998 amps? Is enough of the plant reporting for an average to mean anything?
Because the gate refuses those readings before they reach storage, there is no stored count of what it rejected — so this dataset marks the replay, out-of-range and dawn-suppression figures “not available” rather than publishing a number the database does not hold. The mechanism is real and dated in the case study; the honest thing on a dataset is to not invent the count.
Methodology
How this was measured
Every figure is read directly from the Solar Performance Cloud production database, not estimated or projected. SPC unifies 7 inverter cloud platforms into one data model and grades each string against its peers. Scale figures were read on 25 August 2026; the performance, fault and availability figures are the latest settled day, 24 August 2026, so they are dated rather than presented as timeless facts.
The platform is aligned to IEC 61724-1 (photovoltaic system performance monitoring) and IEC 62446-1 (PV test and inspection vocabulary) — it does not claim to implement, comply with or be certified against them. It computes no performance ratio and no carbon figure: it is the measured-generation layer a Scope-2 or certificate claim is built on, never the calculator.
Everything here is portfolio-level and anonymised: no client, site name or per-site figure appears, by design. The platform was built by Ali Ahmed at BijliBachao.pk.
Released under Creative Commons Attribution 4.0. Use it freely with attribution to Ali Ahmed / BijliBachao.pk. Download the CSV. See also the WattEY production dataset and the full data index.