The Lighthouse
Scout

A corporation that wants to innovate, and a startup that wants to sell into one, have two halves of the same bottleneck: the sides cannot find each other fast enough. The landscape moves daily; the scouting happens quarterly, if at all.

So the corporate's report is stale the moment it lands and generic rather than tied to its real problems, and the startup chases corporates that were never going to buy while missing the one that would. The cost is not just wasted effort. It is the partnership that never happened because the two sides found each other a year too late, or never at all.

The Lighthouse Scout closes that gap by treating innovation as what it actually is, a match between a real problem and a real answer, and by working that match continuously from either side. For a corporation, it surfaces the startups that fit a specific problem. For a startup, it surfaces the corporates that have a genuine, evidenced need for its product. In both directions it shows its evidence for every match, and hands you the few that are real rather than the hundreds that merely sound relevant.

This is not a concept paper. The Lighthouse Scout is the first Scout Fleet capability we have taken into real innovation work. The rest of this paper is what it does, how it should behave, and what must be true before we ask people to trust its recommendations.

A Lighthouse Scout does not run once. Like every Scout, it owns a standing mission, and that mission holds two things in view at the same time.

• What is on offer. It sweeps where innovation shows up: company websites, investor and accelerator portfolios, university spin-outs, grants and awards, patents and research, product launches, and the funding that follows. Continuously, not once a quarter.
• What is needed. It holds the real demand: the specific problems corporations are trying to solve, the capability gaps they are trying to fill, the partners they are actually looking for, and the current evidence that proves the demand is more than a generic innovation page.

It keeps both in view because the value is in connecting them, and it serves whichever side is asking. A corporation points it at a problem it owns and asks what is out there to meet it. A startup points it at what it offers and asks who genuinely needs it. Same watch, two directions.

Because it never stops, timing becomes a feature rather than an accident. It can flag a fit as it emerges, while there is still room to start a conversation, instead of after the round has closed or the budget is spent. And as the landscape shifts, the picture updates on its own. The watch is the product.

The heart of a Lighthouse Scout is the match: connecting a real problem on one side to a real answer on the other. Most tools give you one side, a database of startups, or a list of corporate needs. The value is entirely in the overlap, and the overlap is the hard part.

Because the match has two sides, so does the Scout:

Two examples make the two directions concrete.

A corporation finds innovation through startup signals. A food manufacturer needs to extend the shelf life of fresh produce without adding preservatives. The Scout has been watching the landscape, and it picks up the early signals of an answer: a university spin-out with a new edible coating, a freshly granted patent, a first small round of funding, a quiet pilot with a regional grocer. None of it was on the manufacturer's radar, and the spin-out is far too small to appear on any vendor list. The Scout puts the match in front of the manufacturer while the spin-out is still small enough to partner with on good terms, and it shows the evidence for why the two fit.

A startup finds product-market fit through a corporate problem. A small team has built a way to spot hairline cracks in metal parts from an ordinary phone camera. They have a real product and no clear idea who urgently needs it. Elsewhere, a heavy-equipment maker has published an innovation problem-statement of its own: it wants faster, cheaper inspection of welds out in the field. The Scout reads both sides and sees that the startup's product is a direct answer to the maker's stated problem. It surfaces the maker to the founders, with that problem-statement as the evidence, so they walk into the conversation already knowing the need is real and the budget exists. That is product-market fit found from the demand side, rather than guessed at.

A real match is not "this startup is in your industry," and it is not "this corporate runs an accelerator." It is a specific need met by a specific answer, with evidence on both ends, and a reason to move now. The Scout makes that judgment by meaning, not by matching keywords, so the two sides are paired because they genuinely fit, not because the same buzzword appears on both pages.

The fastest way to lose trust in a matching tool is to drown you in maybes. Loosen the definition of a fit far enough and almost anything can be made to look like an opportunity: a corporate that once mentioned your sector, a startup whose website happens to share a buzzword. So the Lighthouse Scout is strict about the difference between real evidence and wishful matching, and it says which is which. The test cuts both ways: a generic "we love startups" page is as weak a signal as a startup that merely name-drops a big-company logo.

Three things separate what it surfaces from what it sets aside. Underneath them, every piece of demand evidence is first graded on a six-level scale, from an explicit stated need, through a need with a deadline attached, a corroborated proxy, a weak proxy, and watchlist-only, down to not-demand-at-all - and only the top of that scale can be published as an opportunity:

• Strong, evidenced fit. The need is real, the answer clearly meets it, and there is a reason to act. This is what gets put in front of you.
• An indirect hint. Something suggests a fit but the proof is thin. It goes on a watchlist to keep an eye on, not into your inbox as a recommendation.
• Words without substance. A page that mentions the right topics but shows no real fit, on either side. Set aside, on purpose.

Every match it does surface carries the sources behind it, so you can see why it believes what it believes, and check it yourself. A Lighthouse Scout would rather hand you five real leads than five hundred maybes.

It also has a useful answer when no fit exists. Some startups are too early for the current corporate market, some are really a feature rather than a company, and some have not yet made their product legible to a buyer's budget. In those cases the Scout publishes an honest no-fit advisory: what is missing, what would make the company more aligned to corporate demand, and what it should watch for next.

None of this is a one-off tool. A Lighthouse Scout is one of a wider fleet of Scouts, so the same handful of qualities carry over to innovation:

• It does not forget. It remembers what it has already seen and which matches landed, and it gets sharper each time, so it wastes less of your attention as it goes.
• It works as a team, not a soloist. A Lighthouse Scout is not one program but an agentic mesh: many specialist agents working the same mission, dividing the landscape between them, checking each other's findings - a finding that cannot survive refutation against a sibling agent's evidence is demoted to a watchlist instead of counting toward the published result - and pooling what each one learns. So a good find belongs to the whole mesh, not to a lone agent that can discover it once and lose it.
• It shows its work. Every match comes with its sources, so a recommendation is something you can audit, not something you have to take on faith. And the record itself is sealed: each published recommendation is written to a tamper-evident log, tied to the exact scout that made it, and checked again after it ships, so the trail cannot be quietly rewritten later. Withdrawing a recommendation works the same way. A retraction is filed as its own visible entry rather than an erasure, and everything downstream that leaned on it - directly, or through a chain of later findings that leaned on those - is flagged to be refreshed or pulled, each with its own receipt.
• It uses what you already have. Beyond the open world, it can work from what you tell it you are looking for and the tools your team already uses, so what it surfaces reflects your real priorities.
• Your information stays yours. One organisation's problems and data never cross into another's, and you decide what is ever shared outside.

And it keeps improving on its own, within the limits you set, learning from which matches actually turned into conversations. That is the difference between a database you query and a colleague who gets better at knowing what you are looking for. It also knows when to stop. A Scout that keeps spending your budget and returning nothing of value retires its own mission and comes off the schedule, instead of running quietly forever.

The section above calls it an agentic mesh. Here is the shape of that mesh. A Lighthouse Scout is not one program but a parent scout that runs in waves. Each wave, the parent takes its current watch-list, divides it across a standing pool of about ten specialist worker scouts, and sets each one on its own slice. The pool cycles through the list on a rolling weekly rotation, so the whole watch is covered without ever spinning up a separate program for every name.

Every worker runs the same two loops that every Scout runs. Within a single job it plans, acts, and checks, repairing and retrying when a step slips and escalating only when it is genuinely stuck. Across jobs it keeps a fixed seed instruction it never edits and stacks verified lessons on top, so its next pass starts a little smarter. The parent runs that second loop one level up: it weighs which sources, which workers, and which plays actually surfaced a real fit, and re-tunes how it dispatches the next wave. The watch sharpens at two levels at once, the workers in their corners and the orchestrator over the whole field.

Two things keep this honest. First, the learning is in the open: both the parent and each worker publish an append-only record of how they re-tuned their own instructions after each run, so you can read the watch getting sharper rather than take it on faith. Second, the loops are pointed at the right target. Today the watch is deliberately cautious, and far more often than not a worker returns a clean "no credible demand yet" rather than a forced match. That restraint is the point while it earns its quality. So what the loops sharpen first is not a tally of matches. It is the discipline of the search itself, which is the subject of the next section: knowing when to say no.

The hard part of innovation scouting is not finding more names. It is knowing when a name should not be shown as an opportunity. A Lighthouse Scout is useful only if it can resist the temptation to force a story where the evidence does not support one.

• The corporate need has to be real. A broad innovation page is not enough. The Scout looks for a concrete problem, priority, program, buying signal, partnership need, or market pressure.
• The startup answer has to be real. The product, market, team, and traction all have to support the claim being made. Each is graded separately against the startup's cited evidence, and an answer that cannot evidence all four is demoted to the watchlist instead of being published as a completed opportunity - as is one where that check cannot be run at all. If founder or product details are available from the startup source, those should be carried forward and refreshed when newer public evidence appears.
• The fit has to make sense in plain English. A semiconductor startup should not be paired with an agritech company unless there is a clear reason the agritech company needs that exact capability. Same words are not enough; the business need has to line up with the product.
• The next step has to be actionable. A good recommendation should explain why the conversation matters now, who would care, what the first move should be, and what evidence you can check.
• Not every source counts the same. Provenance is a six-rung ladder with fixed weights: a rumour scores 0.10, a press release 0.30, an LLM-normalised record 0.45, a database entry 0.50, an official portal 0.80, and a direct relationship 0.95 - anything the ladder does not recognise defaults to 0.30. That weight sets the confidence axis before any routing decision, and confidence is never folded into priority, so a rumour's confidence tops out at 0.10 and can never clear the routing floor: a weak-sourced claim is pushed toward the watchlist on its provenance alone. And while the Scout can make its own bar stricter as it learns, it can never lower it below a fixed floor.
• It argues against its own case. Before a match is shown, the Scout makes a deliberate attempt to knock it down using only the evidence on file. A match is surfaced only if that challenge fails, and if the challenge cannot run, it holds the match back rather than wave it through.
• No fit is a valid answer. If no credible corporate demand signal exists yet, the Scout should say so. It can still help by advising the startup on how to reposition, narrow, package, or evolve the product so future demand becomes easier to find.

That discipline is what separates the Lighthouse Scout from a database or a newsletter. It does not win by producing the most cards. It wins by being right enough that readers learn to trust the few cards it does publish.

The Lighthouse Scout is the first Scout Fleet capability we have taken out of the lab. It is in sandbox deployment with a government-led innovation ecosystem in Korea that is tracking 4,000 startups across all sectors for global opportunity signals, working the startup side of the match for the programs and founders it backs, with engagements underway with corporations and innovation programs in Japan on the demand side. The work is deliberately careful while it proves quality: a recommendation only becomes a real opportunity when the evidence is complete.

We are deliberate about that word, sandbox. It runs continuously and publishes for real readers, but it earns wider use by proving that its recommendations are honest. The final claim is not simply that it looked at many startups. It is that the demand evidence was diverse, the published recommendations were useful, and each one told the truth: either a complete opportunity, or a clear explanation of why no valid corporate demand signal exists yet.

Innovation is not lost for want of good companies or good problems. It is lost because the two sides cannot find each other in time, against the right need, with enough confidence to act. A periodic, one-sided report cannot do that. A standing, two-way match can.

The Lighthouse Scout turns scouting from a thing you commission into a thing that is always running: it watches the whole landscape, it knows what each side is looking for, it tells a real fit from a logo, and it surfaces the match while there is still time to do something about it. When the fit is not real, it says so and keeps watching. It does not decide who should partner with whom. It makes sure the right counterpart, or the right strategic next step, is in front of you before the moment passes.

For any corporation whose advantage depends on partnering with the right startups before its rivals do, and any startup whose future depends on finding the buyer who truly needs it, that match is the whole game.