Solar
Performance
Cloud

Most often a single string or DC input has quietly failed while the inverter runs on with the rest — correct behaviour that an inverter-level dashboard reports as a healthy machine. String-level faults are 26.89% of all observed solar power loss (Raptor Maps, 2025), the largest single category, and the one you cannot see at the inverter. Solar Performance Cloud grades every string against its neighbours on the same inverter and names the fault the week it happens, not on next quarter’s electricity bill — the question the platform was built to answer.

Compare each string against the others on the same inverter, at the same moment, under the same sky — not against a fixed threshold, which changes with the hour, season, weather and panel age. The string sitting consistently below its neighbours is the one to inspect. Solar Performance Cloud does this automatically across every string on every inverter and names the likely fault: a DC-side break shows voltage but no current, shading tracks the sun, and soiling persists despite good irradiance — so the bad string surfaces the week it fails, not on the next quarter’s electricity bill.

Solar Performance Cloud — an independent solar inspection platform, built as Head of Product & Platforms. It reads seven inverter cloud platforms from six manufacturers, normalises them into one vocabulary, and grades every PV string against its neighbours on the same inverter. It has run continuously since January 2026 across 92 sites, 115 inverters and 874 monitored strings, holding 2.6 million raw readings live and computing a health score for every string once a day.

An inverter is fed by many independent strings, and it keeps running when one of them fails — that is correct behaviour, not a fault. Inverter-level monitoring therefore reports a healthy machine while a tenth of its input is dead. The densest deployment on this fleet carries 36 strings on a single inverter: 36 independent points of failure that the inverter’s own dashboard reports as one number. Reading them separately is the whole design.

A manufacturer’s own portal gives deep, string-level readings only for its own inverters; the moment a fleet spans two brands, you are logging into two portals that grade performance differently and cannot be compared side by side. Platform-agnostic monitoring reads every manufacturer’s cloud, normalises them into one vocabulary, and grades every string on one screen. Solar Performance Cloud reads seven inverter-cloud platforms from six manufacturers this way — which matters because any fleet assembled over time, or across multiple EPCs, inevitably spans several brands: Huawei and Sungrow alone are about 55% of the global inverter market, with no other vendor above 5% (Wood Mackenzie, 2024).

Each manufacturer exposes a different cloud API — different authentication, different polling limits, different data shapes, and different ideas of what a “string” even is. The platform reads all of them on a schedule, normalises the readings into one vocabulary, and, critically, distrusts the incoming data until it earns trust: vendor clouds fail by continuing to answer (replaying the last reading after a logger goes quiet), sensors report physically impossible values, and strings on one channel wake minutes apart at dawn. Solar Performance Cloud reads seven such platforms from six manufacturers, holds millions of raw readings live, and grades every string only after the data clears those gates. The hard part was never the integration — it was making the merged data trustworthy.

You need a measurement layer independent of the manufacturers — reading every site’s real generation at the string level and holding it as a write-once record you control, not a vendor dashboard you cannot audit. Solar Performance Cloud grades every string across a multi-brand portfolio against its neighbours, names underperformance the week it happens rather than at annual review, and keeps per-site, per-day, per-string generation with an auditable internal lineage. That is the raw material for benchmarking actual output against a P50 model or a performance guarantee — the analysis is the operator’s; the platform supplies the measured, independent data it rests on.

Every monitored string is graded into one of five states, each named in the vocabulary of IEC 62446-1 so a finding points at the verification test an engineer would run. Normal — performing in line with its peers. Warning — measurably behind its peers but still producing, usually shading or soiling. Critical — severely behind its peers, a panel-level fault that needs a site visit. Open circuit — making voltage but no current, a physical break on the DC side (a connector, fuse, isolator or cable), explicitly not a panel fault. Offline — no signal at all, either lost communications or a switched-off input; it is deliberately not called “disconnected”, because all the platform knows is that the signal stopped.

Not detecting faults — not raising false ones. Flagging a string that produces less than its neighbours is easy; doing it without becoming unusable is not, because the data arrives broken in ways nobody warns you about. One morning produced 283 false critical alerts in thirty-five minutes, because strings on the same channel wake minutes apart. A failed sensor reported roughly 998 amps, a physically impossible value that poisons the average of every healthy string beside it. And vendor clouds do not fail cleanly — several keep serving the last reading they received, so at midnight you are handed a sunny afternoon. Roughly 192,000 replayed readings were refused in a single week. The platform now assumes the data is wrong until it earns trust.

The first release compared every string against the best performer on its inverter. That pins the bar to a single overperformer, so healthy strings read as though they were behind and any inverter with one strong subgroup turned amber across the board. A client saw it and correctly called it wrong data. It shipped on 19 June 2026 and was reverted on the 23rd, replaced with a comparison against the typical peer rather than the extreme one. A separate fix, in July, was deliberately never shipped: it worked, but its safety guard used a clock window rather than a real sun-position check, and in a Pakistani winter that window opens before sunrise.

Most of it. The solar-specific part is knowing that voltage without current means a broken wire rather than a bad panel. Everything underneath is domain-agnostic: refusing readings that are replays of an earlier snapshot, gating on whether the sun is actually up at that site rather than trusting a clock, rejecting values that are physically impossible for the hardware, benchmarking each device against its local peers instead of a fixed threshold, and withholding a verdict when too little of the system is reporting to justify one. Any platform ingesting device data from vendors it does not control meets the same failures — and they are the reason such projects take a year longer than expected.

It is aligned to IEC 61724-1, which defines how photovoltaic performance is measured and reported, for its daylight gating, availability and yield definitions; and to IEC 62446-1, which defines commissioning tests and inspection for grid-connected PV, for its fault vocabulary — so each finding names the verification test an engineer would run. Both alignments are partial and stated as such. It has no irradiance sensors, meets no IEC monitoring class, computes no performance ratio, runs no commissioning tests, holds no certification, and has had no third-party audit. It also applies no emission factor anywhere, so it reports energy and never carbon.

As the source, not the calculator. SPC produces per-site, per-day, per-string generation with an auditable, write-once lineage — the metered evidence a defensible Scope-2 figure, an avoided-emissions estimate or a renewable-energy certificate is built on. It applies no emission factor and issues no certificate; a carbon accountant or registry turns the kilowatt-hours into the carbon number. SPC never computes carbon itself.

Increasingly, yes. Under the EU’s CSRD, reported climate figures move to limited assurance in 2026 and reasonable assurance by 2028, and the auditor tests traceability — every number must trace back to source data. A solar-offset Scope-2 claim therefore needs auditable, timestamped megawatt-hour-by-site generation records to survive assurance. Solar Performance Cloud produces exactly that source layer; it does not compute the carbon figure itself.

A REC or I-REC certifies one megawatt-hour of renewable generation, and it cannot be issued without proof of that generation: at registration the registrant nominates metered volume evidence, and issuance cross-references metering data with operational logs before certificates are created. No trustworthy generation meter means no certificate. SPC provides the per-site, per-string metered record that evidence rests on — it issues no certificate itself.

On-site solar and procured renewables act on Scope 2 — a company’s purchased-electricity emissions. Generating your own clean power, or contracting for it, lowers the market-based Scope-2 number. But that reduction is only as credible as the evidence of how much you actually generated. Solar Performance Cloud is that evidence: auditable, per-site, per-string generation data.

Companies report Scope 2 two ways. Location-based uses the grid-average emission factor for where the electricity is consumed. Market-based reflects the specific electricity a company contracts for — via RECs, Guarantees of Origin or PPAs — and it is the market-based method that depends on auditable per-megawatt-hour generation evidence. SPC produces that generation record; a carbon accountant applies the emission factor.

It supplies the measurement those claims rest on. Avoided-emissions estimates and MRV (measurement, reporting, verification) both hinge on measured generation against a baseline, and a verifier will test that measurement. SPC is the “M” in MRV — per-site, per-string metered generation with a write-once lineage. It provides the measured energy; it does not make the avoided-emissions claim or run the verification itself.

Those platforms calculate the Scope-2 or avoided-emissions number — but from utility bills and generic grid emission factors, with no device-level measured generation underneath; they are reporting engines. On the other side, inverter portals measure real energy but stop at consumer-grade “CO₂ avoided” tiles. Neither owns the measured, per-string, multi-vendor generation record a defensible claim actually depends on. That is SPC’s lane: the auditable measurement layer between the two — the one input a carbon-accounting tool structurally cannot produce itself. SPC supplies that record; it does not compute the carbon number.

A green bond has to report its results, and the ICMA Harmonised Framework for Impact Reporting names annual renewable energy generation — in MWh or GWh — as a core impact metric, alongside the estimated emissions avoided. That measured generation figure is exactly what Solar Performance Cloud produces: per-site, per-string MWh with an auditable internal lineage, across a mixed-brand fleet. The platform supplies the measured generation an impact report and its avoided-emissions estimate are built on — it does not issue the bond’s certification or compute the carbon figure itself.

Climate finance — green bonds, sustainability-linked loans, results-based finance — increasingly moves only against measured, verifiable generation data, and weak measurement is a known bottleneck. Solar Performance Cloud is that measurement layer: independent, per-string generation across a multi-brand portfolio, named the week a fault happens and kept as a write-once record. That is the raw material for green-bond impact reporting, sustainability-linked KPIs, and lender or investor verification of real output. It supplies the measurement, never the credit, certificate, assurance opinion or carbon figure — and for grid-connected solar it is deliberately positioned for impact reporting and verification, not for minting carbon credits, where additionality rather than measurement is the constraint.

Yes — that is the whole design. Solar Performance Cloud reads seven inverter cloud platforms (Huawei, Sungrow, Growatt, Solis, Canadian Solar, GoodWe and the Solarman white-label), normalises them into one data model and one health taxonomy, and grades every string the same way regardless of brand — so a mixed fleet becomes one view instead of seven vendor portals.

Ali Ahmed built one end to end, running in production. The hard part is rarely solar — it is making third-party device data safe to act on: normalising vendor APIs that disagree on everything, refusing replayed and physically-impossible readings, gating on real sun position rather than a clock, and benchmarking each device against its local peers. That engineering ports to any IoT fleet ingesting data from vendors it does not control.

That he can take a messy, multi-vendor, real-world data problem and turn it into a system people rely on. Solar Performance Cloud reads seven inverter-cloud platforms from six manufacturers that agree on almost nothing, normalises them into one vocabulary, and grades 874 strings across 92 sites every day — while solving the unglamorous problems that break these systems in week two: false alarms at dawn, instruments that report impossible values, and vendor clouds that fail by continuing to answer. He built it as Head of Product & Platforms at BijliBachao, it has run continuously since January 2026, and none of the hard parts are specific to solar — they are what happens to anyone ingesting device data from vendors they do not control.

It runs as a private platform inside BijliBachao — the Pakistani solar-engineering firm founded by Engr Reyyan Niaz Khan, whose 92-site fleet it monitors. Adoption for another operator’s fleet is handled through BijliBachao (spc.bijlibachao.pk), and the same engineering can be built into a different stack. Ali Ahmed built the platform and has done both — deployed it on a live mixed-brand fleet, and can stand an equivalent system up elsewhere.