Solar planset requirements: the 20 ingredients that decide approval

Every grid-tied system joins the utility somewhere, and the NEC permits only a handful of ways to do it. Picking the wrong one, or picking the right one and failing to prove it on the page, is the highest-stakes decision on the set.

The three families. Name the one you used and cite it. Citations here are 2023 NEC, and our interconnection method selector will show you the edition-aware citation for your own code cycle.

• Load-side connection under NEC 705.12(B)(3). The busbar here is the metal bar inside the panel that every breaker lands on, and its ampacity is what the whole rule turns on. 705.12(B)(3) offers six methods, not one. The familiar one is 705.12(B)(3)(2), the 120% rule: 125% of the source output current, plus the rating of the device protecting the busbar, may reach 1.2 times the busbar ampacity. When 120% will not close, the alternatives are (B)(3)(1), the plain 100% sum; (B)(3)(3), the sum of all overcurrent devices excluding the one protecting the busbar; or (B)(3)(4) for a center-fed dwelling panelboard.
• Supply-side connection under NEC 705.11, ahead of the service disconnect. 705.12 does not apply, but 705.11 and 230.82(6) do, and the conductors are sized at 125% of the source output.
• Power Control Systems under Power Control Systems under NEC 705.13, where a listed controller limits output so the busbar math closes.

Opposite-end placement is not a fourth option. It is a condition of the 120% rule. 705.12(B)(3)(2) applies where the primary source and the power source sit at opposite ends of a busbar that carries loads, with the required permanent warning label installed. Backfeed mid-busbar and claim 120% and the set fails review. And in a 2020 or 2023 set the quantity you add is 125% of the source output current, not the backfed breaker rating, which is the 2017 arithmetic a lot of templates still carry.

What good looks like: the calculation is printed, not implied. A reviewer should read the busbar ampacity, the rating of the device protecting it, 125% of the source output current and the resulting allowance without doing arithmetic. On an MLO or feed-through panel, say so: the device protecting that busbar is upstream, not in the panel.

Where experts still get burned

• The note contradicts the drawing. Standing text citing a 705.12(B)(3)(2) backfed breaker on a set that shows a supply-side connection. Templates also cite 705.12(B)(2) for backfeed constantly, which is the tap rule, not the busbar rule. If you keep a library of standing notes, read the one you placed against the drawing before you issue.
• The 120% rule applied to the wrong busbar. With a gateway or smart panel in the design, backup panels, feed-through lugs and integrated smart ports each change which conductor is really the sum point.
• Supply-side conductor length, and which rule actually limits it. Unprotected supply-side conductors inside a building are limited, but the citation moved between cycles: the 2020 NEC carried the constraint in 705.11, and the 2023 NEC sends overcurrent protection to 705.11(F) and on to Article 230 Part VII, where 230.70(A)(1) puts the disconnect at the readily accessible point nearest where the conductors enter. NEC 230.6 is a different lever: it defines when conductors count as outside the building by construction method, such as encasement in two inches of concrete or burial under eighteen inches of earth. It is not a length rule. Either way, none of it is visible on a one-line unless someone put the run length and the method on it.

The single-line diagram, also called the one-line and the same sheet either way, is the one the reviewer reads first and trusts most. Everything else is either supporting evidence for it or a consequence of it.

A complete one-line answers, without ambiguity:

• What sources exist, and what path each takes to the grid
• What protects each segment
• Where the disconnects are
• Where the grounding electrode system attaches
• What the service is

Every device symbol carries a designator (P1, I2, PV-A) and every conductor run carries a circuit letter matching the conductor schedule.

The discipline that separates a good one-line from a decorative one: every symbol is connected to something, and every connection is electrically legal. A symbol floating with no conductor to it is a question the reviewer will ask. A conductor landed on a terminal that cannot legally take it is a design that will not build.

Where experts still get burned

The one-line drifts from the design. Someone swaps an inverter model and the one-line still shows the old breaker size. Someone adds a string and the combiner still shows three inputs.

The one-line was right when it was drawn, and an equipment change downstream left it behind. It reads as a design error to a reviewer even though the design is fine, which is why it costs a full cycle to clear.

Every wire run with its size, conductor material, overcurrent device, equipment grounding conductor, conduit, length and voltage drop. The most mechanical thing on the set, and the most frequently wrong.

The rules that trip people, in the order they trip them

• Size on current, not on OCPD. Size the conductor for the calculated continuous current, then check it against the overcurrent device. Sizing straight off the breaker produces conductors correct by accident and wrong the moment the load changes. On PV circuits watch the two separate 125% factors: 690.8(A) sets maximum circuit current at 125% of Isc, and 690.8(B) then asks for the greater of 125% of that current without correction, or that current with all adjustment and correction applied. Stacking them blindly gives 156%, which is a classic double count.
• Reset the size when the current changes. If the current changes and the size does not, you have a stale row, and it is the row nobody re-reads.
• Conduit fill and ambient correction apply to the whole raceway. Add a fourth current-carrying conductor and every wire in that conduit derates. A schedule that derates row by row with no conduit fill and grouping is not modeling the physical install.
• Terminal temperature limits the whole run. NEC 110.14(C) sizes the conductor to the terminal rating, usually the 75 degree C column, no matter what the 90 degree C column allows. Sizing off 90 degree C ampacity onto 75 degree C lugs is the most common conductor error on a plan set.
• EGC follows the OCPD (250.122). GEC follows the service (250.66). Two different tables, routinely confused.
• Supply-side conductors are a different animal. Between the utility and the first overcurrent device you are in service-conductor territory. Separately, NEC 310.12 allows an 83% ampacity, but read its scope: a one-family dwelling or an individual dwelling unit, single-phase 120/240V, rated 100A through 400A, on conductors carrying the entire load. 310.12(A) covers the service conductors and 310.12(B) covers a feeder carrying that same entire load, so it does reach the meter-main to main-panel feeder. It does not reach branch circuits or three-phase and commercial services, and it never reduces the numbers you use in the busbar calculation.

Where experts still get burned

Voltage rise on the AC interconnection, which is not the same problem as voltage drop on a load circuit. An inverter pushing current back up a long, undersized run raises the voltage at its own terminals. Under IEEE 1547-2018 the first thing that happens is not a trip. Volt-var and then volt-watt quietly curtail real power, well before the over-voltage trip point, which is commonly around 1.10 per unit for a couple of seconds. The exact thresholds come from the utility source requirements document, not from one universal number.

Curtailment is the expensive case precisely because it is silent. The system passes inspection and underperforms forever, and the customer calls it a production issue. It is a conductor sizing issue.

The structural sheets prove the array stays on the roof in a design wind event and that the roof can carry it. This is where a set stops being a drawing exercise and becomes engineering.

The chain a reviewer follows, where every step is a citation and not an assertion:

• Design wind speed and exposure category from ASCE 7 for that address
• Roof zone geometry and pressure coefficients per zone
• Attachment spacing that keeps every lag bolt under its allowable withdrawal
• Rail span within the manufacturer published table
• Dead load summary showing the added pounds per square foot

A rail is not a series of simple beams

Intuition says a support carries the load directly above it, so equal spans mean equal reactions. True for a simple beam, false for a continuous one.

Solve the actual continuous beam and the reactions redistribute. With equal spans, uniform load and no cantilever past the end attachments, the second-from-each-end support carries roughly three times the reaction of the very end support, and about 13% more than a deep interior support. Cantilever the rail, which is normal practice and what the manufacturer span table assumes, and the end reaction climbs steeply: at a third-of-a-span overhang the second support carries only about 1.3 times the end one. Either way the point holds. The reactions are not equal, which support governs depends on the overhang you actually detailed, and span tables that only check bending per span are blind to all of it.

What this means for the sheet: attachment spacing has to be justified against a per-support capacity, and the sheet should say which support governs rather than quoting one span number for the whole row.

Panels straddle wind zones

A residential edge zone is usually three to four feet wide, since ASCE 7 sets it at the lesser of 10% of the least plan dimension or 0.4 times the mean roof height, with a three foot floor. A modern module runs roughly 40 to 45 inches by 66 to 75 inches. A panel near the eave is genuinely in two zones at once.

Assigning the whole panel to whichever zone its centroid lands in is conservative on most panels and wrong on the boundary ones. Weighting the pressure across the zones a panel actually touches, biased toward the worst of them, is the defensible treatment.

What this means for the sheet: the zone layout has to be drawn, not described, so a reviewer can see which zone every panel sits in and check the pressure you used against it.

Know the edges of your method

• The edges are geometric, not height. ASCE 7-22 section 29.4.3 covers flat roofs and gable or hip roofs under 7 degrees. Section 29.4.4 covers panels parallel and close to the roof surface, with a minimum setback from every edge and ridge. Outside those conditions you fall back to Chapter 30 components and cladding.
• Both of those sections read "buildings of all heights," so there is no 60 ft cutoff on them. The 60 ft limit people remember belongs to section 29.4.1, which covers rooftop structures and equipment rather than solar. It is a common enough mix-up that it is worth checking which section your own span tables and software were written against.
• Licensed rail span tables commonly stop at 180 mph. In a high-velocity hurricane zone the binding constraint is usually product approval anyway, a Florida Product Approval or a Miami-Dade NOA for the racking assembly, not the mph ceiling on the table.
• Snow drift surcharge near parapets and rooftop units can multiply snow load by two or three, and is a separate analysis.

If your span table or wind method does not cover the case in front of you, that is a stop, not a pass. A blank is not a compliant number, and the sheet has to cite a basis that actually reaches your conditions.

An accurate overhead showing the array, every piece of equipment, the fire access pathways, and how a firefighter or meter reader physically reaches each one.

• Building footprint to scale with a north arrow
• Every piece of electrical equipment where it will actually be mounted, not schematically
• Fire pathways and setbacks as dimensioned geometry, not a note claiming they exist
• The point of utility service and the route from array to equipment

Where experts still get burned

Pathway and setback requirements are among the most locally amended parts of the code. IFC and IRC baselines get modified by state and county, and coverage-percentage exemptions vary. A set drawn to the national baseline in a county that amended it comes back every time.

Second: equipment on the site plan and equipment on the one-line drift apart. The site plan gets drawn early, the electrical design evolves, and nobody re-opens the site plan.

The permanent signage the code requires at disconnects, inverters, service equipment and rapid shutdown initiation points. Effort 2 against a real rejection risk, which makes it the cheapest whole category of comments on this list to retire.

The label set is determined by the design, not by preference:

• 690.13 for the PV disconnect
• 690.15 where an isolating device is required
• 690.56(C) for rapid shutdown
• 705.12(B)(3)(2) and 705.12(B)(3)(3) at the interconnection, both with wording the code prescribes
• 705.10 for the plaque directing to the location of each power source, plus service equipment marking

Show the actual label artwork with its text, not a note promising labels will be provided.

Where experts still get burned

Labels are the classic correct-but-unclear failure. The system complies, the reviewer cannot see that it complies, the comment says provide labels, and everyone is annoyed.

Labels are also the thing most often copied wholesale between jobs, which means the label text can end up describing something this design does not do. Read the label sheet against the one-line before you issue.

Three separate things that get collapsed into one word:

• Equipment grounding conductors (EGC) bond metal enclosures back to the service. Sized per 250.122 from the overcurrent device.
• The grounding electrode system connects the service to earth. The grounding electrode conductor (GEC) is sized per 250.66 from the service conductors. Different table, different basis.
• The neutral-to-ground bond belongs in the service disconnect enclosure and nowhere downstream of it. 250.24(A)(1) and 250.24(B) require the grounded conductor to reach each service disconnect and be bonded there with an unspliced main bonding jumper sized per 250.28. 250.24(A)(5) is the other half: it prohibits re-grounding the grounded conductor anywhere on the load side. Where a service has more than one disconnect, each enclosure carries its own bonding jumper. Downstream panels have separated bars.

Where experts still get burned

Double bonding. Add a gateway, smart panel or transfer switch and the question "which enclosure is now the service disconnect" has a new answer, and the panel that used to hold the bond must not hold it any more.

This survives a casual read, fails an inspection, and puts objectionable current on the grounding path in the meantime. Any time you add equipment ahead of the panel, re-ask where the bond belongs.

One case that looks like a violation and is not: a supply-side tap disconnect is itself service equipment, so 250.24(B) gives it its own main bonding jumper wherever the grounded conductor is run to it. The other legitimate additional bond is a separately derived system, which takes a system bonding jumper under 250.30(A)(1).

The local amendments, submittal preferences and documented quirks of the specific authority having jurisdiction (AHJ) reviewing this set. Nationally compliant is not the same as locally approvable.

We publish per-jurisdiction permit requirements for the counties we cover. Before you draw, know whether this jurisdiction:

• Has adopted which code cycle
• Amended fire pathway requirements
• Requires grounding shown explicitly on the plan
• Requires a production meter
• Requires structural blocking details
• Sits in a high-velocity hurricane zone
• Issued your last three comments, and what they were

Where experts still get burned

Assuming the code you know is the code in force. Adoption cycles lag, amendments stack on top, and two counties an hour apart can want visibly different sets for the same system.

The last bullet is the one people skip and the most accurate of the seven. What a jurisdiction asked for on your last set describes what it wants better than anything published on its website.

The serving utility has its own rules, separate from the AHJ. The one thing they always want to find on the page is exactly where and how you are touching their system.

• Label the point of interconnection explicitly, with the actual conductors identified
• Where your utility requires an external AC disconnect, and a growing number no longer do for IEEE 1547 compliant inverters, state its attributes as requirements rather than as claims: load-break rated, visible blade, lockable open with provision for a utility padlock, NEMA 3R, exterior, with the handle in the height range that utility specifies
• Include the per-utility boilerplate that utility asks for

Where experts still get burned

Assuming AHJ approval implies utility approval. They are separate reviews on separate timelines. A set that sails through the building department can stall for weeks at the utility over an unlabeled point of interconnection or a disconnect location their crew will not accept.

Proving every string stays inside the inverter voltage window on the coldest morning and the hottest afternoon, and inside the current limit of each maximum power point tracker (MPPT) input. The full math is in the NEC 690.7 string sizing guide.

Two opposing failure modes:

• Cold-morning open-circuit voltage. Voc rises as temperature falls. Apply the temperature coefficient across the gap between 25 degrees C and the site extreme minimum, multiply by string length, and compare against the lowest of three ceilings: the inverter maximum DC input, the module maximum system voltage on its nameplate, and, on a one- or two-family dwelling, the 600V limit in NEC 690.7(C). That last one governs most residential work and is the one a calculation checking only the inverter will sail past. Exceeding any of them is a violation, and exceeding the inverter rating can damage the input.
• Hot-afternoon maximum power voltage. Vmp falls as cell temperature rises, and cell temperature runs well above ambient. Add a cell temperature adder and check string Vmp is still above the MPPT minimum. Fall below and the inverter stops tracking.

Then the current side, which is two separate checks that get collapsed into one. For conductors and overcurrent devices, NEC 690.8(A) gives maximum circuit current as 125% of Isc times parallel strings. For the inverter, compare the actual Isc sum against the manufacturer maximum short-circuit current per MPPT and the Imp sum against its maximum input current. Those are manufacturer limits, not NEC ones. Read the datasheet closely: the per-MPPT figure is usually the total across that MPPT's parallel inputs, but there is often a separate per-input connector or fuse limit that binds first.

Where experts still get burned

• Using ambient for the hot case. Cell temperature is not air temperature. Skipping the adder gives a comfortable Vmp the array never actually produces.
• Using the wrong low temperature. The extreme annual mean minimum for the site, not a seasonal average and not the record low from a weather app.
• The full-load window. Some inverters have a narrower voltage band at full power than their advertised MPPT range. A string sized to the outer range can sit inside the MPPT window and still be unable to reach full rated output.

Hot-climate work outside the continental US inverts the priority. Where there is no meaningful cold, the hot Vmp case governs nearly every design, and US-centric temperature data sources will not cover the site at all.

The same fact appears on six sheets, and it has to be the same fact on all six.

• Panel count on the cover, the site plan, the equipment schedule, the string table and the one-line
• Inverter model in the schedule, on the one-line, on the label sheet and on the datasheet you attached
• System size on the cover and in the calculations
• Address in the title block on every sheet

Where experts still get burned

This is the ingredient whose effort scales worst. Two sheets are easy. Twelve sheets with a late equipment change is a long, careful read, and it is the one most often skipped under deadline.

It is also the failure that most damages credibility. A reviewer who finds one number disagreeing with itself starts checking everything.

The facts that have to agree

The list is shorter than it feels. Panel count, panel model, inverter model and quantity, system size, address, busbar rating and main breaker, and the designators themselves. Check those and you have checked cross-sheet consistency.

Address, customer, jurisdiction, utility, scope, sheet name, date, revision and the responsible parties. Effort 1. Still gets sets rejected.

Where experts still get burned

Not the typing. Anything inherited from a previous job or a template:

• A sheet duplicated from a previous job that keeps the previous job address
• A template carrying the wrong engineering identity or contractor
• A revision date that never advanced

None of these are design errors. All of them read to a reviewer as carelessness, which changes how hard they look at everything else.

Real manufacturer, real model, real rating, real quantity, each tied to the designator used on the drawings.

• Manufacturer and model as they appear on the listing, not as shorthand
• Ratings that match the datasheet you attached
• Designators matching the one-line exactly, so a reviewer can trace I1 from schedule to diagram to datasheet without guessing

Where experts still get burned

Placeholder equipment that survives to submittal. TBD, 400W panel, Inverter. Early design needs a stand-in, and the stand-in is easy to forget. Field weight is 4 here because the installer buys from this table.

Manufacturer documentation for every listed component. Include what is installed and nothing else.

Two failure directions

• Missing sheets for a component swapped late
• Bloat: a sixty-page appendix of every datasheet the team has ever attached, which buries the relevant page and signals nobody curated it

The check is the same in both directions. Read the appendix against the equipment schedule, one row at a time.

A reviewer needs four things from an elevation:

• Position: where along the wall, and how high above finished floor
• Size: real width by height
• Identity: which designator maps to which make and model
• Compliance: working space per 110.26, handle height per 404.8(A), ready access per 240.24(A), plus the manufacturer's listed installation instructions, which NEC 110.3(B) makes enforceable. Around batteries, add the separation and egress requirements of NFPA 855 where your jurisdiction has adopted it, usually by reference from IFC 1207 or IRC R328

Where experts still get burned

Treating the elevation as decorative. Equipment gets sized generically, clearances get asserted in a note, and the sheet proves nothing.

Drawn to real dimensions with the clearance zones actually shown, an elevation either proves compliance or exposes a genuine conflict at design time rather than at inspection. Battery work has moved elevations from nice-to-have to expected in many jurisdictions, because energy-storage spacing and egress cannot be answered from a plan view.

Notes should state requirements as requirements ("shall be installed"), not as as-built claims ("is installed"), because the set is issued before the install exists. Code basis notes should name the cycle that jurisdiction actually adopted, not a generic year.

Where experts still get burned

Standing notes carry authority. A reviewer reads them as statements about this design, so a note that contradicts your drawing is worse than no note at all. It is affirmative evidence on the page that the set does something it does not do.

The usual culprit is a general-notes block carried over from a previous job: interconnection language for a method you did not use, a code cycle this jurisdiction has not adopted, a reference to equipment that is not on this set. Read every standing note against the drawing before you issue.

A binary gate rather than a quality ingredient. Where a jurisdiction requires a professional engineer seal, an unsealed set is not reviewed, it is returned.

Effort is 1 if the engineering relationship is already in place and 5 if you are sourcing one under deadline, which is the entire reason to check the requirement before you design rather than the week you submit.