PicoRastro

1. One-liner

Tells a Mexican plant which machine set its CFE demand peak, on a bill CFE never explains.

2. Trend signal — why now?

Three things moved in Mexico inside twelve months, and they all land on the same piece of paper: the industrial electricity bill.

The billing mechanic got sharper. CFE’s medium-voltage tariffs (GDMTO under 100 kW of demand, GDMTH at 100 kW and above) charge on demanda facturable — a charge driven by the single highest power draw your plant pulled in any 15-minute interval during the month. One interval sets the charge for the entire billing period. As one Mexican energy consultancy puts it, “a single poorly coordinated startup at peak time can set a peak that gets billed for the whole month.” The capacity rate is not trivial money: the Valle de México Centro division’s GDMTH capacity charge ran $386.37/kW in July 2026. Mexican industrial bills carry between 12 and 18 different charge lines, and the formulas setting capacity and billable demand for commercial and medium-voltage industrial users were adjusted upward effective 2026.

A regulator put a hard date on power factor. Effective 8 April 2026, Código de Red raised the minimum power factor from 0.95 to 0.97 lagging for load centres in high tension and medium tension with contracted demand at or above 1 MW. The bigger change is not the number — it’s the measurement rule. The monthly average is dead: compliance now means staying in range 95% of the time, measured in five-minute intervals. Transient dips that a monthly average used to absorb are now visible and countable. Below that 1 MW line — where most PYMEs sit — there’s no Código de Red obligation, but CFE still applies its own power-factor penalty below 0.90 via the formula (3/5) × ((0.90 / FP) − 1) × billable amount, and the gap between running at 0.85 and 0.95 is worth 3–5% of the total invoice, month after month.

And the data that would explain all of it is locked. This is the part that makes the idea. CFE has deployed 32–35 million AMI meters since 2015 that already record consumption in 15-minute intervals, capturing active energy, reactive energy and demand. Medium-tension customers have had advanced metering for over a decade. But in the words of the same analysis, “no existe regulación que obligue a CFE a liberar los datos a los usuarios que los generan” — there is no regulation obliging CFE to release the data to the users who generate it. Mexico has no Green Button equivalent. Load curves exist but must be requested manually, plant by plant, with no automated portal.

That’s the whole opportunity in one sentence: the meter already knows which quarter-hour cost you the money, and you are not allowed to see it.

Meanwhile the load itself keeps rising. Nearshoring drove $23.591 billion of FDI into Mexico in Q1 2026, a record quarter, with nearshoring accounting for 58% of total FDI, while national electricity consumption grows near 4% annually and CFE’s grid investment lags. More machines, tighter grid, meaner tariff, no visibility.

Provenance:
  - Signal 1 (demand): CFE demanda facturable is set by a single 15-minute interval; capacity charge $386.37/kW (Valle de México Centro, July 2026); bills carry 12–18 charge lines and PF below 0.90 is penalized by formula, worth 3–5% of invoice — https://www.energiareal.mx/blog/cargo-por-demanda-cfe-factura-industrial + https://enerlogix.org/en/blog/factor-potencia-penalizacion-cfe — 2026-08-23
  - Signal 2 (feasibility): CFE has 32–35M AMI meters recording 15-min intervals, but "no existe regulación que obligue a CFE a liberar los datos a los usuarios que los generan" — no Green Button in Mexico, load curves are manual-request only; meanwhile 3-phase CT-clamp meters with Modbus/MQTT and PF measurement are $60–$90 plus $80–$200 of CTs — https://www.odondebuenr.com.mx/eficiencia-energetica/green-button-mexico-acceso-datos-energeticos/ + https://www.iammeter.com/blog/3phase-reactive-power-modbus-mqtt — 2026-08-23
  - Signal 3 (economic/regulatory): Código de Red raised minimum PF from 0.95 to 0.97 effective 8 April 2026, replacing monthly average with 95%-of-time compliance in 5-minute intervals for ≥1 MW load centres; nearshoring FDI hit a record $23.591B in Q1 2026 while consumption grows ~4%/yr — https://ap-automatizacion.com/articulo/codigo-de-red-2-0-en-el-2026-fp-de-0-97/ + https://www.mundohvacr.com/2026/04/codigo-de-red-2026-que-cambia-y-por-que-el-promedio-mensual-ya-no-alcanza/ — 2026-08-23
  Category: Geographic arbitrage (a data-access right that exists in the US and does not exist in Mexico) + Tech-unlock

3. The opportunity

There is a live energy-services market in Mexico and every player in it sells capital equipment. Energía Real sells BESS with predictive software and 24/7 monitoring under leasing or shared-savings schemes. Capacitor banks for power-factor correction run 80,000–200,000 MXN installed for 100–300 kVAR, with 6–18 month payback. The Código de Red compliance industry sells studies and UVIE certification. All of it is rational — hardware and certification carry margin.

The nearest thing to a software play is Batu Energy, and it’s aimed somewhere else. Batu automates CFE bill downloads, monitors panels and batteries, and does multi-site management, but its stated customers are “Instaladores, EPCs, integradores” — solar installers managing client portfolios — and its Pro plan runs $22,000 MXN/month. It reads the bill. It does not tell you which machine caused the number on the bill.

So the hole is specific. A metalworking shop or plastics moulder in Querétaro or Guanajuato with a 180 kW peak is on GDMTH, is paying real money on demanda facturable, and gets exactly one artifact: a PDF with 12–18 lines, arriving weeks late, containing a kW figure and no explanation of which Tuesday afternoon produced it. Nobody will sell that shop a $40K battery, and it can’t buy a $22,000 MXN/month platform built for solar integrators.

The 10× is attribution and timing, not generation. Two questions no incumbent answers for this customer:

1. Which 15-minute interval set this month’s demand charge, and what was running during it?
2. Is this month’s peak already set — or can I still beat it if I stagger the compressor and the injection moulder for the next nine days?

That is a knowledge product, not an equipment product. Which is precisely why the equipment vendors have not built it.

4. Target market

• Primary customer: The plant manager or gerente de operaciones at a Mexican manufacturing PYME — 20–200 employees, single site, on CFE tariff GDMTO or GDMTH (roughly 40–400 kW of demand), concentrated in the industrial corridors: Querétaro, Guanajuato/Bajío, Estado de México, Nuevo León, Jalisco, Puebla. Metal fabrication, plastics/injection moulding, food processing, cold storage, textiles, small foundries. Buyer is typically the owner or a single operations lead — not an energy manager, because they don’t have one.
• Why they buy: Because the demand charge is a real, recurring line they cannot explain or predict, and because the fix is usually free. The consultancy framing is direct: a sudden consumption spike from simultaneously starting heavy machinery or centralized A/C can blow up the demanda facturable for the entire billing period. Rescheduling a startup costs nothing. Not knowing which startup to reschedule costs every month. Industry estimates put 5–12% of savings available “without moving a single piece of equipment” for medium operations.
• Rough TAM reasoning: I’ll be honest that CFE does not publish a public count of GDMTO/GDMTH service accounts, so this is bounded reasoning, not a sourced number. What is sourced: SMEs are 80% of Mexico’s economy and 20.4% of employment in this category is in manufacturing. Any plant drawing over ~40 kW is on a medium-voltage demand tariff. A defensible working estimate is tens of thousands of medium-voltage industrial and commercial accounts nationally; I need only ~700 of them to hit $1M ARR. This is a niche-capture problem, not a market-size problem, and I’d rather state that plainly than invent a TAM figure.
• Why now for them: Tariff formulas for medium-voltage industrial and commercial users adjusted upward in 2026. Código de Red’s 8 April 2026 change made power factor a continuously-measured obligation for the ≥1 MW tier — and it is dragging awareness downmarket, because the consultants and equipment vendors chasing the 1 MW crowd are now cold-calling everyone. The smaller plants are being told they have a problem and are being quoted capacitor banks. Many of them don’t need a capacitor bank; they need to know what’s happening. Nobody is selling them that.

5. Product sketch (MVP)

• A clamp-on meter you install in an afternoon. Three CT clamps in the main panel plus a small logger. No electrician-week, no CFE paperwork, no interruption to the CFE meter itself — this measures alongside it. Ships pre-configured; the customer scans a QR to bind it to their account.
• The peak autopsy. For every billing period: the exact 15-minute interval that set demanda facturable, what the plant was drawing before and during it, and a ranked list of the load signatures that stacked up to make it.
• “Your month is still winnable.” A running view of the current billing period’s highest interval so far, days remaining, and whether the peak is already locked or still beatable. Alerts in Spanish over WhatsApp when the plant is climbing toward a new monthly maximum — while it is happening, not thirty days later.
• Machine fingerprinting. The operator tags a handful of loads once (“compresor”, “inyectora 2”, “horno”). After that, the system attributes interval peaks to named equipment from their electrical signature rather than making the owner guess.
• Power-factor watch. Continuous PF tracking against both the CFE 0.90 penalty threshold and the Código de Red 0.97 / 95%-of-time rule, so a plant approaching the 1 MW line knows where it stands before a consultant tells them.
• A staggering plan. Concrete, plant-specific sequencing advice: which two loads are colliding, and what a 20-minute offset is worth in pesos on next month’s bill.
• A bill-reconciliation view. Upload or link the CFE PDF; the system checks the charged demanda facturable and PF penalty against what it independently measured, and flags mismatches.
• Spanish-first, WhatsApp-native. Reports, alerts and the monthly summary land in WhatsApp, because that is where a Mexican plant manager actually reads things.

6. AI angle — what’s load-bearing

Strip out the AI and this collapses into a chart, which is a thing several vendors already sell badly. The load-bearing work is disaggregation and attribution: taking one aggregate three-phase signal at the main panel and inferring which machines were running during the interval that set the charge.

That’s non-intrusive load monitoring, and it is genuinely a model problem — inrush signatures, duty cycles, coincident starts, reactive-power behaviour that distinguishes a motor start from a resistive heater. The customer will not submeter twelve machines; that’s the expensive answer they already declined. So the product’s entire reason to exist is getting machine-level answers out of panel-level data.

Second load-bearing piece: forecasting whether the month’s peak is still beatable. Predicting the remaining-days risk of exceeding the current maximum, given the plant’s own observed patterns, is what converts this from a post-mortem into something that changes behaviour inside the billing cycle. A post-mortem is a report. A forecast is a product.

Third, smaller but real: turning a 12–18 line CFE bill and a raw interval series into a two-paragraph Spanish explanation an owner reads on a phone. That’s ordinary LLM work, but it’s the difference between a dashboard nobody opens and a WhatsApp message that gets acted on.

7. Localization angle

This is not a localized version of a global product — the localization is the product, and it’s why a US player won’t come here.

• The data-access inversion. US demand-charge tools (Sanalife, Vutility, and the whole Green Button ecosystem) are built on utilities being required to expose interval data via ESPI APIs and OAuth. Mexico has no Green Button and no obligation on CFE to release the data. A US-shaped product literally cannot boot here — there is no API to call. That forces a hardware-first shape, which is more work and is exactly why the space is empty. It’s also a moat with a shelf life: if Mexico ever mandates data access, the hardware advantage erodes (see risk flags).
• Tariff structure is CFE-specific. GDMTO vs GDMTH, the 100 kW boundary, demanda facturable, the base/intermedia/punta hourly blocks, the (3/5) × ((0.90/FP) − 1) penalty formula, division-specific capacity rates. Encoding this correctly is unglamorous domain work that compounds.
• Código de Red is a Mexican regulator with a Mexican date. The 0.97 / 95%-of-time rule has no analogue elsewhere. Knowing where a plant sits relative to the 1 MW threshold — and what happens if nearshoring growth pushes it over — is local knowledge with real value.
• Price point and rails. A ₱2,500–6,000 MXN/month tier works where a $22,000 MXN/month integrator platform doesn’t. Billing in pesos, invoicing with CFDI (which every Mexican business demands), and selling in Spanish over WhatsApp are table stakes, not features.
• Distribution is regional and physical. Querétaro, the Bajío and Monterrey industrial parks, CANACINTRA chapters, and the existing network of electrical contractors. This is a market you win by showing up at an industrial park, not by ranking on Google.

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

• Pricing: Hardware + SaaS. $4,500 MXN one-time for the meter kit (CT clamps + logger, roughly $150–290 USD of parts at $60–90 for a 3-phase meter plus $80–200 for CTs, leaving healthy margin to fund install support), then $2,900 MXN/month (~$155 USD) per site for the software. A larger tier at $5,900 MXN/month adds Código de Red power-factor compliance tracking and multi-meter coverage for plants near or over the 1 MW line.
• ACV: $34,800 MXN/year ($1,860 USD) on the base tier; $70,800 MXN ($3,780 USD) on the compliance tier. Blended realistic ACV $42,000 MXN ($2,240 USD).
• Rough math to $1M ARR: ~840 sites at blended $2,900–3,500 MXN/mo. Call it 700–900 plants. At a 4-industrial-park-per-quarter ground game that’s a 2.5–3 year build, or faster with contractor channel partners.
• Rough math to $5M ARR: Needs three things true: (a) ~3,500–4,000 sites, which means going beyond manufacturing into cold storage, hotels, hospitals and retail chains on the same tariffs; (b) multi-site accounts — a chain with 15 locations at one contract; (c) plausibly a second country on the same shape (Chile, Colombia and Peru have analogous demand-charge tariffs and similar data-access gaps). $5M is a stretch on Mexican manufacturing PYMEs alone; I’d rather say that than pretend.
• Expansion path: More meters per site (submetering a specific production line once the panel-level view proves out), the Código de Red compliance tier as a plant grows toward 1 MW, multi-site rollups, and a referral cut on the capacitor-bank or BESS install when the answer genuinely is equipment. That last one matters: this product will sometimes correctly conclude the customer needs hardware, and there’s a clean commission there rather than a conflict.

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

• Walk the industrial parks, with a free autopsy as the door-opener. Querétaro, Celaya/Irapuato, Apodaca and Toluca have dense, addressable industrial parks with published tenant directories. The offer is concrete: let me put a logger on your panel for two weeks, free, and I’ll tell you which machine set your demand charge. Two weeks of data on a plant with a real spike problem produces a number the owner has never seen. Target 40 free installs to close 15–20. This is a physical sales motion and I’m not going to dress it up as growth hacking.
• Ride the Código de Red panic downmarket. Consultants and UVIE inspection units are actively selling compliance studies into the ≥1 MW tier right now because of the 8 April 2026 change. Every one of them is generating rejected leads — plants that called, got quoted 80,000–200,000 MXN for a capacitor bank, and walked away. Partner with 5–8 of those consultancies and electrical contractors for referral of the sub-threshold plants they can’t monetize. They keep the equipment jobs; I take the ones that just need visibility, and refer back the ones that turn out to need iron.
• CANACINTRA and CAINTRA chapter talks. Mexican industrial chambers run regional chapters with regular member sessions and are actively worried about electricity costs. A 30-minute talk titled “¿Qué máquina te está costando el cargo por demanda?” with a live bill teardown is a credible way in front of 40–80 plant owners at once. Book six chapters in the first two quarters.
• The bill teardown as content, in Spanish. Publish anonymized autopsies: a real CFE bill, the interval that set the charge, the machine responsible, and the peso value of a 20-minute stagger. This is the one content play I believe in here, because the artifact is inherently specific and shareable in WhatsApp groups where Mexican plant managers actually talk. It supports the ground game; it does not replace it.
• Land one anchor per park, then farm it. Industrial parks are gossip networks. One plant that cuts its demand charge and says so at a tenant meeting is worth more than fifty cold emails. Explicitly price a referral incentive into the first 100.

10. Build complexity — justification

Medium. The hardware is genuinely off-the-shelf: three-phase CT-clamp meters with Modbus/MQTT and reactive-power and power-factor measurement are commodity items at $60–90 plus CTs, with cellular backhaul for plants with unreliable wifi. No custom silicon, no certification gauntlet, no manufacturing capex — source, configure, and ship. The SaaS layer (ingest, intervals, alerts, WhatsApp via Business API, Spanish reporting) is standard work.

The real work is in three places: CFE tariff logic encoded correctly across GDMTO/GDMTH, divisions, hourly blocks and the PF penalty formula; load disaggregation good enough that machine attribution is credible rather than embarrassing; and field installation — this is a physical product going into live electrical panels, which means install guides, a partner electrician network, and support calls. That last one is the part software people underestimate.

Realistic estimate: 4–5 months to a credible v1 for a pair (one embedded/data, one full-stack) plus a Mexican electrical-domain advisor. The disaggregation model can start crude — panel-level peak timing and coarse signature clustering are already valuable — and sharpen with accumulated data. That staging is what keeps this Medium rather than High.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Customer-owned monitoring on the customer’s own side of the meter. No CFE authorization needed, no tampering with CFE’s meter, no regulated activity. Selling advice, not electricity.
Ethical — no harm / dark patterns	✅	Reduces energy waste and customer bills. Honest failure mode is telling a customer their peak is structural and the software can’t help — which must be said plainly rather than churned.
Market exists (evidence above)	✅	Funded equipment vendors, a live consulting industry, documented tariff mechanics, and a dated regulatory change all confirm spend in this category.
1–5 person team can build this	✅	Two builders plus a domain advisor and a contract electrician network.
Launchable with <$50K / ₹40L	✅	Initial hardware run of 100 units at ~$250 landed = $25K, plus development. Fits, though it’s more capital-forward than a pure-SaaS idea.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Recurring monthly money on a line the customer genuinely cannot explain, with a documented 3–5% PF swing and 5–12% of savings available without buying equipment. Felt every billing cycle. Not 19 because it’s a margin leak, not an existential threat — plants have absorbed it for years.
Demand evidence	15	11/15	Strong indirect evidence: funded BESS vendors, a capacitor-bank industry with 6–18 month paybacks, a live compliance-consulting market, upward tariff adjustments. Docked because I could not source verbatim complaints from Mexican plant managers specifically about demand-charge attribution — the pain is inferred from vendor behaviour, not heard in the customer’s own words. That’s the weakest axis and the first thing to validate.
Build feasibility	15	11/15	Off-the-shelf hardware and standard SaaS, but 4–5 months with a physical install motion and a disaggregation model that has to be non-embarrassing. Not a 6-week solo build.
Distribution clarity	15	11/15	Named parks, named chambers, a named referral source (Código de Red consultants with rejected sub-1MW leads), and a free-autopsy opener with real conversion logic. Docked because it’s a boots-on-ground motion that doesn’t compress — 40 installs to close 20 is slow.
Revenue mechanics	15	12/15	Pricing sits credibly between free-and-useless and the $22,000 MXN/mo integrator platforms, hardware margin funds install support, and ~840 sites to $1M is reachable. $5M requires vertical and geographic expansion, which I’ve flagged rather than assumed.
Time to first revenue	10	7/10	Hardware lead time plus a two-week free pilot plus a physical install means first paid site realistically 8–12 weeks after launch. Slower than pure SaaS, faster than enterprise.
Defensibility	10	5/10	Encoded CFE tariff logic, accumulated load-signature data, and installed hardware create real switching cost by month 12. But the concept is copyable, and the deepest risk is structural: if Mexico mandates Green Button-style data access, the hardware moat becomes a hardware liability.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · sales-heavy · domain-expertise-required

This needs someone who can do embedded data work and walk into a Querétaro industrial park and talk to a plant manager in Spanish. The domain requirement is real — CFE tariff structure is not learnable from a weekend of reading, and getting it wrong in front of a customer ends the sale. A Mexican electrical-engineering co-founder or a very committed advisor is close to mandatory.

Key assumptions to validate

1. Assumption: Mexican plant managers experience demand-charge attribution as a felt pain, not just a theoretical inefficiency. How to test: 25 in-person interviews across two industrial parks. Ask them to explain last month’s demanda facturable line and name the machine responsible. If most can, or most don’t care, the premise is wrong.
2. Assumption: Panel-level disaggregation is accurate enough that a plant manager believes the machine attribution. How to test: Instrument one friendly plant at both panel and machine level for 30 days. Measure whether panel-only inference identifies the true peak-driving load. Below ~70% and the core claim doesn’t hold.
3. Assumption: A plant will pay ~$2,900 MXN/month for knowledge with no equipment attached. How to test: Pre-sell. Take the free-autopsy result to 15 plants and ask for a signed 6-month commitment before any software exists. Under 3 signings is a red flag.
4. Assumption: Peaks are actually actionable — that a meaningful share are caused by schedulable coincident starts rather than an irreducible process load. How to test: In the 30-day pilot data, classify peaks as schedulable vs structural. If most are structural, this becomes a hardware-referral business and the SaaS thesis weakens.
5. Assumption: Código de Red consultants will refer their sub-1MW rejects rather than sit on them. How to test: Approach eight consultancies with a concrete referral split. Fewer than three interested kills the cheapest channel.

Risk flags

1. Regulatory risk (moat-erosion, not compliance): The entire feasibility wedge rests on CFE not exposing interval data. There is an active, published proposal to bring Green Button to Mexico, citing California and Ontario precedent. If that lands, the hardware requirement flips from moat to cost disadvantage overnight, and US-shaped competitors can enter cheaply. Mitigate by making the disaggregation and tariff intelligence the durable asset, and by treating the meter as a means, not the product.
2. Capital and ops intensity: Hardware means inventory, RMAs, install support, and working capital that pure SaaS doesn’t carry. It also means a failed unit is a truck roll. This is the main reason the idea is Medium not Low, and it caps how fast a two-person team can scale sites.
3. Attribution credibility: If machine-level attribution is wrong in front of a plant manager who knows his floor, trust is gone and it doesn’t come back. Better to ship coarse-but-correct (“this 15-minute window, these two loads coincided”) than precise-but-wrong.
4. Incumbent encroachment: Batu Energy already has CFE bill automation and an installer channel. Adding attribution is a plausible roadmap item for them. The defence is speed and a different customer — they sell to integrators, this sells to plants — but it’s not a durable structural barrier.
5. Market-timing risk on Código de Red: The 8 April 2026 change drives awareness now. That urgency decays as the ≥1 MW tier finishes complying. The downmarket referral channel is best in the next 12–18 months and gets colder after.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical pair (embedded/data + full-stack) with a Mexican
                        electrical-engineering co-founder or domain advisor; must be
                        willing to run a physical, in-person industrial sales motion
                        in Spanish.
Time to revenue:        8–12 weeks from launch (hardware lead time + 2-week pilot + install)
Capital to launch:      $25–40K USD (~₱470–750K MXN), hardware-inventory-forward
Top 3 assumptions to validate first:
  1. Plant managers can't explain their own demanda facturable and care that they can't
     — 25 in-person interviews across two industrial parks.
  2. Panel-level disaggregation identifies the true peak-driving machine ≥70% of the time
     — 30-day dual-instrumented pilot at one friendly plant.
  3. A plant will pay ~$2,900 MXN/mo for knowledge with no equipment attached
     — pre-sell 6-month commitments to 15 plants off a free autopsy, before building.
Kill criteria:
  - Abandon if fewer than 3 of 15 free-autopsy plants will sign a paid commitment.
  - Abandon if panel-level attribution accuracy stays below 70% after the 30-day pilot.
  - Abandon if the 30-day pilot shows most peaks are structural process loads rather than
    schedulable coincident starts — the advice has no lever and this becomes hardware resale.
  - Abandon if CFE or CRE mandates customer interval-data access before v1 ships.

15. Next step — 1-week validation sprint

• Day 1–2 — Get the tariff math indisputably right, and find one friendly plant. Pull actual CFE bills for GDMTO and GDMTH from any contact with a plant, and reconstruct demanda facturable and the PF penalty from first principles until the computed number matches the printed number to the peso. If I can’t reproduce a CFE bill, I have no business selling attribution against it. In parallel, secure one plant willing to host a logger.
• Day 3–4 — Instrument it and walk a park. Install a clamp meter on the friendly plant’s main panel (parts are commodity and available same-week). Simultaneously, do 15–20 unscheduled walk-in conversations at a single Querétaro or Bajío industrial park. One question drives it: “Show me last month’s bill — can you tell me which machine set this charge?” Record how many can, how many care, and how many ask what it would take to find out.
• Day 5 — Decide on a falsifiable threshold. Go/no-go on this specific bar: of 15+ plant managers interviewed, at least 8 cannot explain their demand charge AND at least 5 volunteer that they’d want it explained; plus at least 3 agree in principle to a paid pilot at ~$2,900 MXN/month. Below that, the pain is theoretical and I walk. Above it, the week’s logger data becomes the first free autopsy and the first sales artifact.

The falsifiable part is deliberate: this idea dies if plant managers can already answer the question. One afternoon in an industrial park settles it.