RedlineZero

1. One-liner

Catches the AHJ corrections in your solar plan set before you submit it, not three weeks after.

2. Trend signal — why now?

Three things moved in the last twelve months, and they moved in opposite directions — which is exactly where the money sits.

Plan-set drafting collapsed in price. GreenLancer residential permit packages start around $250–$300, with PE stamp add-ons at $150–$200. Indian engineering BPOs undercut that. In-house drafting runs $530–$810 per project all-in once you count salaries, licensing, software, and idle time. The drawing itself is a commodity now.

But rejection stayed expensive. A single rejection stalls a job one to two weeks; at the top end a rejected project costs $2,000–$5,000 once you count admin time, crew rescheduling, and client delays. Worse, a first-submission rejection can push a project back 30 to 90 days once you lose your queue slot. And SEIA data says a one-week delay raises client cancellation rates 5–10%. So the $300 drawing routinely triggers a $3,000 loss.

The code churned underneath everyone. NEC 2026 is landing on a fragmented adoption schedule — jurisdictions adopt on their own cycles and interpret differently, so two neighbouring towns can demand different sheet sets, labels, or roof pathways. NEC 690.8 violations alone cause 30–40% of solar permit rejections nationwide, most often from incomplete application of current multipliers and temperature correction factors. Leading EPCs hit 90%+ first-submission approval; the long tail does not.

The tell that this is unbuilt: a July 2026 industry piece on AI-assisted plan review says design teams “will soon be able to use AI to flag designs that are likely to be rejected.” Soon. Future tense. The category is being discussed, not sold.

Provenance:

Provenance:
  - Signal 1 (Demand): NEC 690.8 violations cause 30–40% of solar permit rejections nationwide; a rejection costs $2,000–$5,000 and 30–90 days of queue position, and a 1-week delay lifts cancellation 5–10% (SEIA) — https://energyscaperenewables.com/post/solar-plan-set-cost-in-house-vs-outsourced/ + https://energyscaperenewables.com/post/solar-permit-plan-set-checklist-2026-ahj-ready-guide/ — 2026-07-30
  - Signal 2 (Feasibility): Plan-set drafting is commoditized ($250–$480 outsourced) while AI-assisted pre-submission plan review is described in industry press as still arriving — "design teams will soon be able to use AI to flag designs that are likely to be rejected" — https://energyscaperenewables.com/post/ai-assisted-plan-review-clean-solar-plan-sets-2026/ — 2026-07-30
  - Signal 3 (Economic): SolarAPP+ now runs in 240+ communities across 17+ states (~1/3 of national market) with strict eligibility — roof-mount, existing homes, generally under 38.4 kW — proving AHJs accept automated compliance checking while leaving the other ~2/3 of the market, plus batteries, ground-mounts and commercial, on manual review across 20,000+ jurisdictions — https://www.energy.gov/cmei/systems/articles/160-communities-now-automating-solar-permitting-solarapp + https://energyscaperenewables.com/post/virtual-solar-permitting-2026/ — 2026-07-30
  Category: Tech-unlock

3. The opportunity

The solar permitting market has been solved from both ends and left hollow in the middle.

On one end, plan-set production is a race to the bottom — GreenLancer, Exactus, Wattmonk, Energyscape, plus a wall of Indian BPOs. On the other end, SolarAPP+ made permitting instant for the narrow slice of jobs that fit its box: roof-mounted, existing home, under 38.4 kW, in one of 240-odd adopting communities. Florida has zero adoption statewide.

Nothing sits in between. If your job is a battery retrofit, a ground-mount, a commercial rooftop, or simply in a town that hasn’t adopted SolarAPP+, your plan set goes into a human review queue in one of 20,000+ jurisdictions, each with its own local amendments and its own NEC adoption year. You find out what’s wrong 5–20 business days later, in a correction letter.

The incumbent weakness is structural, not lazy. Design services are paid per plan set, so a redline is a second billable event for them — they have no economic reason to build a tool that catches errors upfront. Design software (Aurora, OpenSolar) optimizes for sales proposals and layout, not for whether Springfield’s building department will accept sheet E-2. And SolarAPP+ is a government-funded instant-approval rail, not a diagnostic that tells you why your non-conforming job would fail.

So the gap: a pre-submission diagnostic that reads the plan set you already have and tells you, per jurisdiction, what will come back redlined. Not a drafting service. A gate you run before the drawing leaves the building.

The 10× is arithmetic, not magic. Catching one redline pays for a year of subscription. An installer doing 200 jobs a year with a 25% redline rate is eating 50 correction cycles; halving that is 25 jobs × $2,000+ in recovered margin and queue position.

4. Target market

Primary customer: Owner-operator or permitting/ops manager at a residential-and-light-commercial solar installer doing 50–600 installs a year in the US — typically 8–60 employees, $3M–$40M revenue. The buyer is whoever owns the permit calendar: often titled Permitting Coordinator, Operations Manager, or the owner in smaller shops. Best beachhead states are those with weak SolarAPP+ coverage and high volume — Florida (zero SolarAPP+ jurisdictions), Texas, Georgia, the Carolinas, Arizona.

Why they buy: Because the redline is the single most expensive recurring surprise in their operation and it lands after they’ve already scheduled a crew. Industry reporting puts the leading rejection causes as incomplete or inconsistent documentation, code non-compliance, structural deficiencies, and electrical diagram errors — and notes a specific failure pattern: a plan set showing panels in one roof location while the structural calculations reference a different section gets returned for correction. The recurring frustration in the trade press is jurisdictional whiplash — “what worked in a neighbouring city may not fly in the next jurisdiction over,” with requirements varying not just by county but by project type within the same county. An installer expanding into a new metro is flying blind on that metro’s amendments.

Rough TAM reasoning: Being deliberately conservative. The US has a few thousand active residential solar installers. Call it 4,000–6,000 firms in the 50–600 install band nationally. At $400/mo average that’s a $19M–$29M theoretical ceiling. I don’t need the ceiling — I need 400 of them. That’s the right shape for this portfolio: too small to interest a VC-backed entrant, plenty for a two-person company.

Why now for them: NEC 2026 is being adopted on a staggered, jurisdiction-by-jurisdiction schedule right now. Every installer’s institutional knowledge of “what this town wants” is going stale simultaneously, and the historical fix — a senior person who just knows — is exactly the knowledge that’s expiring. Meanwhile the tax-credit-driven demand environment means crew time is the scarce resource, so a 30–90 day queue loss hurts more than it did in a slack year.

5. Product sketch (MVP)

• Drop in a plan set, get a redline forecast. Upload the PDF (any origin — in-house, GreenLancer, BPO) plus the jurisdiction. Back comes a ranked list of what a plan checker will flag, with the specific sheet, the specific code citation, and the fix.
• Cross-sheet consistency check. The unglamorous killer: panel counts, inverter model, conductor sizes, breaker ratings, and roof plane references compared across the site plan, single-line diagram, structural calcs, and equipment datasheets. Catches the “panels in one location, calcs reference another section” class of error automatically.
• NEC 690.8 calculation audit. Re-derives current multipliers and temperature correction factors from the equipment and location, and flags where the drawing’s numbers don’t survive the math. This is the single highest-yield check in the product.
• Jurisdiction profile. For each AHJ: adopted code edition, known local amendments, required sheets, label standards, PE-stamp requirement, submittal format. Starts thin and thickens with every submission outcome customers report.
• Code-edition mismatch flag. Tells you when the plan set was drawn to a code year the target jurisdiction hasn’t adopted (or has already moved past) — a whole rejection category that’s pure clerical waste.
• Correction-letter parser. Paste or upload the redline you did receive; it maps each comment to the sheet and drafts the resubmittal response. This is the trojan horse — it’s useful on day one before the predictive model has any track record.
• Outcome ledger. Per jurisdiction and per design source: first-pass approval rate, average days to approval, most frequent comments. Turns “I think Broward is strict” into a number, and tells the owner which design vendor is actually costing them money.
• Pre-submittal packet. One-click checklist confirming every required sheet and document is present in the jurisdiction’s expected order and format.

6. AI angle — what’s load-bearing

Remove the AI and this product cannot exist. That’s the test, and it passes cleanly.

The core task is reading an unstructured, non-standardized engineering PDF — plan sets come from dozens of drafting shops with no common layout — and extracting a structured understanding of the electrical design: what equipment, what conductors, what roof planes, what labels, on which sheet. Vision-capable models made that tractable in the last two years. Before that you’d have needed either a rigid CAD-file integration (which fails the moment a customer’s plan set arrives as a flattened PDF from a BPO) or a human reviewer (which is the incumbent, at incumbent cost and speed).

Three AI jobs, in order of load:

1. Extraction — parse a heterogeneous multi-sheet PDF into a structured design model. This is the hard part and the moat’s foundation.
2. Cross-sheet reasoning — reconcile the same fact asserted in four places with different notation, and decide whether a discrepancy is a real inconsistency or a formatting artifact. False positives destroy trust faster than misses, so this needs to be conservative.
3. Language generation — draft the resubmittal response letter mapping each AHJ comment to the corrected sheet.

Note what is not AI: the NEC 690.8 arithmetic. That’s deterministic and must stay deterministic — a hallucinated ampacity calculation is worse than no product. The model extracts the inputs; hard-coded engineering logic does the math. Getting that boundary right is most of the engineering judgment in the build.

7. Localization angle

N/A — this is a US-only play, and the "localization" is domestic. The wedge is jurisdictional fragmentation, but it’s fragmentation inside one country: 20,000+ AHJs, each with a code edition and a local amendment set. The equivalent product for another market would be a different product — I already have a separate India rooftop-solar submission idea in this portfolio (DiscomPass), and it addresses DISCOM net-metering format rejection, not NEC calculation error. Different buyer, different failure mode, no overlap. Deliberately not forcing a localization angle here.

8. Business model — path to $1M–$5M ARR

Pricing: Tiered by install volume, because that’s what redline exposure scales with.

• Solo/small — $199/mo, up to 15 plan sets/mo
• Growth — $499/mo, up to 60 plan sets/mo, multi-jurisdiction profiles, outcome ledger
• Multi-market — $999/mo, unlimited, multi-branch reporting, API into their project tracker

ACV: ~$4,800 blended (weighted toward Growth).

Math to $1M ARR: 210 customers at $4,800. Or concretely: 120 Growth + 60 Multi-market + 80 Solo ≈ $1.06M. Two hundred-ish installers out of 4,000–6,000 is roughly 4% penetration of the addressable band. That’s a believable number for a focused product, and it’s the number I’d hold the business to.

Math to $5M ARR: Needs more than seat growth — it needs the adjacent buyer. Three levers: (a) sell the same diagnostic to the design services and BPOs themselves as a white-label QA gate, since their redline rate is their reputation and they’d pay per-plan-set to lower it; (b) expand into battery storage and small-commercial plan review, where jobs are bigger and SolarAPP+ doesn’t reach at all; (c) sell the outcome dataset back as jurisdiction benchmarking. Lever (a) is the most interesting because it flips the incumbent from competitor to channel. I’d want (a) working before claiming $5M is real.

Expansion path: Per-plan-set volume growth as customers grow, then seats for multi-branch operations, then the design-vendor scorecard as a paid module. Gross margin should sit at 80%+ — inference cost per plan set is cents-to-low-dollars against a $199–999 monthly subscription, and the deterministic checks are free.

The unit economics have one honest risk: heavy-PDF inference on a $199 tier with a customer who dumps 15 large multi-sheet sets a month. Volume caps per tier exist for exactly that reason, and they need to be enforced from day one, not retrofitted.

9. Go-to-market wedge — first 100 customers

The correction-letter parser is the door-opener, because it demonstrates value on a redline the installer already has in their inbox — no trust required, no prediction to believe.

• Run the free teardown play on public permit data. Many jurisdictions publish permit application status and correction records. Pull a target metro — Broward and Miami-Dade first, since Florida has no SolarAPP+ coverage — identify installers with visible resubmittal cycles, and send the owner a two-page teardown: here are your last N corrections, here’s the pattern, here’s what it cost you in queue days. That’s not a cold pitch, it’s their own operating data reflected back. Expect a high reply rate because it’s specific and slightly alarming. Target 300 letters → 15–25 conversations → 5–8 paying.
• Ambush the code-transition anxiety. Every jurisdiction adopting NEC 2026 creates a cohort of installers who just lost their institutional knowledge. Publish a free, genuinely maintained “what changed in your AHJ” tracker per metro, gated by email. This is the one content play I’ll accept, because the artifact itself is the product’s jurisdiction database — building it for marketing builds the asset. Feeds the top of funnel continuously.
• Go through the permitting coordinators, not the owners. There’s a real community of permitting/ops staff in solar — regional SEIA chapter events, Facebook and Slack groups for solar operations people. This role is measured on approval rate and gets blamed for redlines. Give 20 of them free accounts, ask for outcome data in exchange. They become both the reference list and the jurisdiction dataset.
• Partner with the design shops, cautiously. The mid-tier plan-set vendors compete on rejection rate. Offer a revenue-share white-label QA gate. This is the highest-leverage channel and also the most dangerous — they’re the most likely acquirer and the most likely copier, so I’d approach it after 50 direct customers, not before.
• Equipment distributors as a wedge. Racking and inverter distributors already bundle plan-set services (IronRidge resells GreenLancer plan sets, for instance). A pre-submission check is a natural attachment to that bundle and puts the product in front of hundreds of installers through one relationship.

10. Build complexity — justification

Medium. Roughly 14–18 weeks to a credible v1 for two people.

Off-the-shelf: PDF ingestion, vision-model extraction, web app, auth, billing. The correction-letter parser is genuinely easy — it’s text in, structured response out — which is why it ships first and earns revenue while the harder half is built.

Custom and non-trivial: the deterministic NEC 690.8 calculation engine (must be exactly right, needs an electrical engineer’s review, not a weekend of guessing), the cross-sheet reconciliation logic, and the jurisdiction profile database. That last one is the real cost — it’s a manual research grind at the start, seeded from published AHJ checklists and thickened by customer-reported outcomes. It’s also the moat, so the grind is the point.

What keeps this out of High: no custom model training, no hardware, no regulatory approval, no marketplace to bootstrap. It’s document understanding plus a rules engine plus a database somebody has to be willing to build by hand.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Advisory diagnostic on documents the customer owns. Must not represent itself as engineering certification or a PE stamp — that’s a licensing boundary, and the product sits clearly on the advisory side of it.
Ethical — no harm / dark patterns	✅	Improves code compliance and electrical safety. Aligned with the AHJ’s own goal. The one discipline required: never let it become a way to game a reviewer rather than fix a design.
Market exists (evidence above)	✅	30–40% of rejections from one code section; $2,000–$5,000 per rejection; installers already pay $250–$810 per plan set and outsource QA today.
1–5 person team can build this	✅	Two people, 14–18 weeks, plus an EE reviewer on contract for the calculation engine.
Launchable with <$50K / ₹40L	✅	Two founders’ time, inference costs, and a few thousand dollars of engineering review. Well under.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real money, felt on a recurring cycle, with a specific painful moment (correction letter arrives after crew is scheduled). Not a 19 because installers have functioning workarounds — senior staff who know the jurisdictions, and outsourced designers who absorb some of the blame. Pain is expensive but survivable, which means switching effort matters.
Demand evidence	15	12/15	Strong indirect evidence: quantified rejection causes, quantified rejection cost, a paid-services market already absorbing this work, and 240+ SolarAPP+ jurisdictions proving automated checking is acceptable to AHJs. Held at 12 because I could not source verbatim installer quotes about redline pain — the trade press covers it, but I found no first-person practitioner voice. That’s a real gap, not a rounding error.
Build feasibility	15	10/15	Pair ships in 14–18 weeks. Document extraction from heterogeneous plan sets is the risk; the NEC calculation engine needs outside engineering review; the jurisdiction database is manual grind. Nothing research-grade, but not a 6-week build either.
Distribution clarity	15	12/15	The public-permit-data teardown is a named channel with a named list and specific metros. The permitting-coordinator community is real and reachable. Not higher because conversion on cold outreach to installers is unproven and the distributor channel is a hypothesis.
Revenue mechanics	15	11/15	Pricing is benchmarked against what installers already pay per plan set, and the ROI argument is one-redline-pays-for-a-year. 210 customers to $1M is credible. Docked because the $5M path depends on the design-vendor channel, which is exactly where the competitive risk lives, and because inference cost on heavy PDFs needs disciplined tier caps.
Time to first revenue	10	7/10	The correction-letter parser can be sold before the predictive engine is finished, so revenue in ~8 weeks of launch is plausible. Not 9–10 because installers buy on a considered cycle and will want to see it work on their own jurisdictions first.
Defensibility	10	5/10	Honest 5. The jurisdiction database plus accumulated submission outcomes compound into a genuine data moat by month 12 — but at month 3 there is nothing here a funded design-services incumbent couldn’t build, and they own the customer relationship already. Execution-and-head-start moat, and I’m not going to dress it up as more.
Total	100	73/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

This needs someone who can build serious document-understanding infrastructure and either has solar permitting scars or hires them immediately. Getting NEC 690.8 wrong once, publicly, kills the product’s credibility permanently — this is not a domain where you can learn in production. If you don’t have the electrical engineering access, don’t start.

Key assumptions to validate (3–5)

1. Assumption: Installers in the 50–600 install band have a first-pass approval rate materially below the 90% that leading EPCs hit — call it 70–80% — and they can quantify it. How to test: 25 structured calls with permitting coordinators. Ask directly for their last 20 submissions and count. If they don’t know their own number, that’s informative too: the outcome ledger becomes the lead feature rather than a supporting one.
2. Assumption: A vision model can extract equipment, conductor, and roof-plane data from arbitrary third-party plan sets accurately enough that flagged issues are trustworthy. How to test: Collect 40 real plan sets from 5 different drafting sources plus their actual correction letters. Measure recall against the real redlines and — more importantly — measure the false-positive rate. Target: catch ≥60% of real comments with <15% false positives.
3. Assumption: Installers will pay $199–$999/mo for a diagnostic rather than just paying their design vendor to redo it. How to test: Sell the correction-letter parser at $199/mo to 10 installers before building the predictive engine. Ten paying customers proves the wallet; three means the design vendor absorbs this pain and the wedge is wrong.
4. Assumption: The jurisdiction profile database can be built to useful depth for a beachhead metro in under two weeks of one person’s effort. How to test: Actually do it for Broward and Miami-Dade. If two counties take a month, the model doesn’t scale to 20,000 and the product has to narrow to a few states permanently.
5. Assumption: Design-services vendors will white-label rather than clone. How to test: Direct conversations with three mid-tier vendors — but only after 50 direct customers exist, because this conversation is also a competitive disclosure.

Risk flags

1. Incumbent adjacency (highest risk): GreenLancer, Wattmonk, Exactus and the rest already hold the customer relationship, already see thousands of plan sets, and already know the jurisdiction requirements informally. If one of them decides pre-submission QA is a feature rather than lost revenue, they ship it with a distribution advantage this product will never have. The counter-argument — that a redline is a second billable event for them — is real but it’s a business-model argument, and business models change under competitive pressure.
2. Platform/policy risk from SolarAPP+ expansion: SolarAPP+ is federally supported and growing. Every jurisdiction that adopts it, and every eligibility expansion (higher kW caps, new construction, storage), shrinks the manual-review market this product serves. Watch the adoption curve quarterly. Mitigation is to lean toward the segments SolarAPP+ structurally can’t serve — commercial, ground-mount, complex retrofits.
3. Trust cliff / false-positive risk: The product’s value is that its warnings are worth acting on. A tool that cries wolf gets ignored within two weeks and churns at month two. This argues for shipping deliberately narrow and precise — fewer checks, higher confidence — which is the opposite of the instinct to look comprehensive at launch.
4. Liability boundary: If an installer relies on a clean report and gets rejected anyway — or worse, builds something unsafe — the positioning must have been unambiguously advisory from day one. Terms of service and product language matter here more than in most SaaS, and this needs a lawyer’s eye before launch, not after.
5. Demand evidence gap: I’m publishing this without first-person installer quotes. The economics are well-sourced; the voice is not. That’s the specific thing the validation sprint has to close.

14. Structured verdict

Score:                  73/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder strong in document AI, with a solar
                        permitting operator as co-founder or first hire. Do not
                        attempt without electrical engineering review access.
Time to revenue:        8–10 weeks post-launch (correction-letter parser first)
Capital to launch:      $12–20K (₹10–17L) — mostly inference and EE review
Top 3 assumptions to validate first:
  1. Installers at 50-600 installs/yr have first-pass approval well below 90%
     — 25 permitting-coordinator calls, ask for their last 20 submissions
  2. Vision extraction hits >=60% recall on real redlines with <15% false
     positives — 40 real plan sets from 5 drafting sources vs. actual
     correction letters
  3. Ten installers will pay $199/mo for the correction-letter parser alone,
     before any predictive capability exists
Kill criteria:
  - Abandon if fewer than 3 of 25 interviewed installers can name a redline
    that cost them a crew day in the last quarter
  - Abandon if false-positive rate stays above 25% on the 40-plan-set
    benchmark after two iterations — the trust cliff makes it unsellable
  - Abandon if a major design-services vendor ships pre-submission QA before
    your v1 — they have the distribution and you do not
  - Abandon if SolarAPP+ eligibility expands to cover storage and
    new-construction across your beachhead states

15. Next step — 1-week validation sprint

The goal is a falsifiable answer on the one thing I couldn’t source from the desk: does the pain exist in the installer’s own words, and can the extraction actually work.

• Day 1–2: Build the benchmark set. Get 40 real plan sets paired with their actual AHJ correction letters. Source them from installer contacts, permitting coordinators, and published jurisdiction records. Pay for them if necessary — $2,000 here de-risks the whole build. Without paired plan-set-plus-redline data there is no way to measure anything, so this day is non-negotiable.
• Day 3–4: Run extraction against the benchmark, by hand if needed. No product, no UI. Feed each plan set to a vision model, produce a flagged-issue list, and score it against the real correction letter. Record recall and false-positive rate per rejection category. Simultaneously run 25 calls with permitting coordinators asking one question first: what was your last redline and what did it cost you? Capture verbatim.
• Day 5: Decide on two numbers. Go if (a) extraction catches ≥60% of real correction comments at <15% false positives, and (b) ≥15 of 25 interviewed coordinators name a specific redline from the last quarter that cost them crew time or a queue slot. No-go if either fails.

Failing (a) means the technical premise is wrong and no amount of go-to-market fixes it. Failing (b) means the pain is real in trade-press economics but not in operator priority, which is the more dangerous failure — it’s the one that produces a product people admire and don’t buy.