Since 2023 I've been assessing deeptech startup files for Team for the Planet, a citizen-owned fund that finances innovations with high emission-reduction potential. I look at three things: technical feasibility, real impact, business model. I go deeper on that role in my Team for the Planet case study.

And there's one sentence that always puts me on alert when I read it in a file: "we're almost industrial-ready".

"Almost" is not a level. TRL exists precisely to replace that word with a number everyone reads the same way. If you're building a deeptech company and you can't state your TRL in one sentence, you're losing money and time — not because your technology is weak, but because nobody on the other side of the table knows what to do with your file.

1. What TRL actually is

TRL — Technology Readiness Level — is a scale from 1 to 9. It was born at NASA in the 1970s, formalised in the 1990s, then adopted by the European Commission with Horizon 2020. Today it is the default language of innovation funding in Europe.

It answers one single question: how much of the remaining risk is technical?

That matters, because most misunderstandings start there. TRL is not a quality score. It's not a rating of ambition, nor a ranking of serious companies against the rest. A TRL 3 isn't "worse" than a TRL 7: it's earlier. So it's cheaper to fund, and riskier. Those are two useful pieces of information, not a judgement.

What TRL really measures is the amount of evidence accumulated. Each level corresponds to a specific type of proof. Moving up a notch isn't about being more confident: it's about having produced a new kind of evidence.

2. The nine levels, in founder language

Here's the scale as I read it, with the proof each level actually demands:

  1. TRL 1 — basic principle observed. You have a publication, or an observation. You don't have a product, and that's fine.
  2. TRL 2 — technology concept formulated. You know what to build and why it should work. Nothing is built yet.
  3. TRL 3 — experimental proof of concept. It worked once, in a lab, with the person who designed it standing right there. The first real milestone.
  4. TRL 4 — components validated in the lab. The subsystem holds up, in an environment you fully control: your temperature, your power supply, your sample.
  5. TRL 5 — validated in a relevant environment. The most misunderstood step of all. This is the first time reality is allowed to break your technology: dust, vibration, temperature swings, an operator who isn't you.
  6. TRL 6 — demonstrator in a relevant environment. This time it's the complete system, not an isolated component. The difference between 5 and 6 is almost always right there.
  7. TRL 7 — demonstrator in an operational environment. On site, at a real user's premises, under real operating conditions, with real constraints.
  8. TRL 8 — system qualified. It works and it is certified, approved, compliant. The part everyone underestimates in the schedule.
  9. TRL 9 — proven in operation. It runs, it's sold, it surprises nobody any more.

The zone that matters is 4 → 5 → 6. That's where most of the deeptech risk plays out — the zone commonly known as innovation funding's "valley of death": too mature for academic research, too immature for an investor who wants revenue within 24 months. A lot of very good technologies die there, without the science ever having been proven wrong once.

3. Why Europe forces you to know your TRL

This isn't a theoretical debate, and 2026 makes it very concrete. The Commission adopted a 2026 European Innovation Council work programme worth roughly €1.4 billion for deeptech. And that funding is carved up by TRL band:

  • EIC Pathfinder — upstream research, the lowest levels of the scale.
  • EIC Transition — precisely the crossing of the valley: starting from TRL 3-4 to reach 5-6, with tickets up to €2.5M.
  • EIC Accelerator — from the middle of the scale upwards, towards qualification and deployment.

The direct consequence: your TRL isn't branding, it's an eligibility criterion. Overstate it and you apply to an instrument whose boxes you don't tick — you lose six months. Understate it and you leave money on the table, plus a slice of dilution you could have avoided.

The same logic applies on the private side, just less written down. A fund with a ten-year horizon can carry a TRL 4. A fund that needs revenue within two years cannot, however good your science is. When an investor tells you "it's too early for us", they're almost always talking about TRL, not about your team.

4. The three most common TRL mistakes

Three mistakes I run into regularly in the files, and that cost dearly:

Mistake 1 — presenting the component's TRL as the system's. Your membrane may well be at an advanced stage on a test bench. But the machine around it — pumps, sensors, control software, maintenance — is far more upstream. A system's TRL is that of its weakest link, never that of its most brilliant part. It's also the one that's fastest to spot.

Mistake 2 — dressing up the lab as a relevant environment. "Relevant" has a precise meaning: you no longer control every parameter. If the test happens at your place, on your bench, with your prepared sample and your engineer at the controls, it isn't relevant. It's TRL 4 done well — which is already something.

Mistake 3 — giving your TRL in the future tense. "We'll be TRL 7 by year-end" answers a question nobody asked. The question is: where are you today, and with what proof? A trajectory without a verifiable starting point isn't a trajectory.

5. Assessing yourself honestly: three questions

You don't need a consultant to place yourself. Three questions are enough, and they're deliberately hard:

  1. What is the last thing that physically worked, and where? Not the last simulation, not the last calculation. The last thing that ran in the real world.
  2. Who saw it work outside your own team? A pilot customer, an industrial partner, a certification body, a third-party lab. An external witness moves you up a level; internal conviction doesn't.
  3. What broke last time? That one is the most revealing. A founder who can't name a recent failure hasn't run a real test yet — or isn't telling you about it.

This extends my usual trick question, the one I detail in my checklist for evaluating a climate startup: show me the data from your latest run in relevant conditions, not in the lab. If the answer is fuzzy, you are not where the pitch claims you are.

6. Why being honest about your TRL earns you points

This is the point I most want to land, because it's counter-intuitive: a low TRL doesn't disqualify you. A misstated TRL does.

Look at the spread of levels across the innovations Team for the Planet has funded — public files, public figures:

  • Cool Roof France — a reflective paint that cuts a building's cooling needs by around 40%. Mature technology, top of the scale, deployable immediately: hundreds of thousands of m² of roofing already repainted across France. No ten-year scientific bet.
  • Beyond the Sea (~€2.5M) — kite traction sails for cargo ships targeting −20 to −40% fuel. A 50 m² sail in dynamic flight tests in the Arcachon Basin: that is exactly what crossing from the middle of the scale towards an operational demonstrator looks like. Concrete, filmable, verifiable.
  • Leviathan Dynamics (~€2.34M) — industrial cooling produced with water, replacing high-GWP refrigerants. First machines shipped, ~2,000 t CO₂ avoided over the past year, entering industrialisation. The top of the scale being earned in real time.
  • Seaturns (~€0.8M, wave energy) and Monomeris (near-infinite plastic recycling) — two markedly more upstream bets. Technically riskier, with enormous tipping potential if the industrial step gets cleared.

Across more than 30,000 innovations assessed and 14 deployed to date, TRL is never a binary filter. It's a way of knowing what kind of risk you're taking, and therefore what kind of money and what horizon fit. One important precision: I score, I don't decide. Team for the Planet has around 133,000 shareholders and funding decisions there are collective — my score is only one opinion among the other assessors', and that is exactly what makes the system robust.

So what really sinks a file isn't being at TRL 4. It's claiming 7 when you're at 4. Because the moment an assessor finds a gap between what you assert and what you prove, they re-read everything else with that doubt in mind. You don't lose one point on feasibility: you lose your credibility on impact, on the model, on the team. Saying "TRL 4, aiming for 5 within eighteen months, here's the test that will validate it" is a signal of command. "We're almost industrial-ready" is a signal of imprecision.

If you're building a deeptech company and want to stress-test your self-assessment before it goes into a file — or if you're an assessor and want to compare our grids — email me at contact@thanaelfontaine.eu or book a slot from the contact page. On TRL, being wrong in good faith is expensive; placing yourself accurately costs nothing.