The technology that eats your roadmap is a preprint right now. We read those.
WILLOW Tech-Scouting Siphon is a standing weak-signal sweep of the open research literature, scored against your technology thesis. Every cycle it triages the new preprints in the fields you care about, deep-reads the candidates that matter, and hands you a short dossier: the technology, the team behind it, why it matters to your thesis, and the exact source passages that back every claim. By the time a capability is a funded startup with a booth, you have been watching it for months.
Put your technology thesis under watch See a sample candidate dossier
THE PROBLEM
Corporate tech scouting runs at human cadence against a literature that does not. The scouting team commissions a landscape report, the report is stale the week it lands, and the interval between reports is exactly where the surprises live. Meanwhile the actual early signal is public the whole time: it is sitting on preprint servers, posted by a lab group nobody in your industry has heard of yet, in a category adjacent to the one you watch. The volume is the killer. No scouting team reads tens of thousands of new papers a month, so they sample, and sampling is how a company with a standing interest in, say, solid-state batteries finds out about the relevant electrolyte work eighteen months late, from a competitor's press release. The alternative on offer (an AI summary tool) has its own failure mode: it reads everything and then confidently tells you things the papers do not say, with no way to check which sentences are load-bearing.
HOW IT WORKS
You give us a technology thesis, not a keyword list: the capability you care about, the adjacent fields it could come from, what would make a candidate matter to you, and what to ignore. The Siphon then runs the same sweep engine AYA uses on its own R&D. Every cycle, every new paper in the configured categories gets triaged against your thesis and scored. The top candidates get deep-read from the actual full text, not the abstract, and worked into a candidate dossier: what the technology is, who the team is, what stage the work is at, and why it intersects your thesis. Every claim in the dossier is labeled one of two ways. Verified means it is grounded in a named passage of the real source, and the dossier points at that passage, a receipt you can check. Assessed means it is our read (an inference about maturity, trajectory, or fit), labeled as exactly that. The two never blur. A human analyst reviews every dossier before it reaches you, and the count reconciliation is built in, so a quiet week is a real quiet week, never a silent outage dressed as one. Sourcing is OSINT only: public preprint servers and open literature. No paywall circumvention, nothing classified, no scraping where we are not welcome.
WHAT YOU GET
Per cycle (weekly by default), one scouting packet you can route straight into your pipeline: - the candidate dossiers, technology, team, stage, and thesis-fit for each surfaced candidate, - the receipts, every verified claim pointing at the exact source passage that grounds it, - the labeled judgment calls, our assessments marked as assessments, with the reasoning stated, - the sweep ledger, what was scanned, what was surfaced, what was dropped and why, reconciled by count, so you know the radar actually ran, - the monthly roll-up, movement across cycles: which candidates are accelerating, which went quiet, what entered the field.
WHO IT'S FOR
Corporate innovation and CTO-office teams whose next decade depends on catching a technology early, first. Then deep-tech investors sourcing pre-seed signal from the literature, and public sector innovation units restricted to open sources. If your field moves through preprints and you are watching it through quarterly reports, this is for you. If what you need is paywalled market databases, patent litigation support, or deal flow, it is not, and we will say so.
PRICING (the ladder)
We price the ladder, not a single number, and every number below is a hypothesis we validate with you, not a commitment. - Pilot sweep, a fixed-price 4-week proof on one thesis: intake session, weekly dossiers, one roll-up review. You judge the signal quality on your own thesis before committing to anything. - Standing thesis, per thesis per month, once the pilot proves the signal: continuous watch, weekly dossiers, monthly synthesis, thesis tuning as your interest sharpens. - Portfolio, for funds and CTO offices watching several theses, with cross-thesis collision alerts. Negotiated.
IP is licensed, never assigned. Any step that takes your money is gated and confirmed before it runs, nothing charges silently.
THE PROOF (dogfood)
The engine is not hypothetical: it is the same radar AYA points at its own development, sweeping the AI, security, and database literature against our own build priorities (multi-agent orchestration, verification, de-identification, grounded text-to-SQL). The Siphon is that exact loop with your thesis in the slot where ours sits. The build left its own receipt too: the deliverable workflow and all eight atoms it composes were verified on disk and scored by our pattern quality gate before this page was written. We did not write new machinery for this product, and that is the point, the engine compounds, the thesis is what changes.
HONEST NOTE
We would rather under-promise. What exists today: the sweep engine, structurally complete, quality-gated, and pointed at our own thesis. What has NOT happened yet: no external client has ever received a scouting report from this product, no client thesis has been configured, no pilot has run, and the analyst review gate, while designed in, has never been exercised against a real delivery. The standing self-run evidence stream is architected but not yet captured (it needs our mesh booted and keys funded, which is scheduled, not done). The first pilot thesis is the natural first test, and its buyer will know they are first. Where a claim can be grounded we ground it and attach the receipt; where it is our judgment we label it as judgment; and a gap in coverage is declared, never papered over. That refusal to fake certainty is the whole product.
Put your technology thesis under watch
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*