Technology Intelligence
The scientific edge in
emerging technology investment
Innovation Lens provides technology intelligence to VCs and institutional investors on the emerging technologies that will define the next decade, grounded in the latest research from leading labs worldwide.
1,171
Institutional managers’ real books tested
+1.31 pp
Added per year to the average book
85%
Of those books improved
0 of 1,000
Random orderings that matched it
Hypothetical backtest on public 13F filings, March 2020 to October 2026, gross of trading costs. The holdings are real; the re-weighting of them is a simulation, not a live account. Past performance does not guarantee future results.
What We Do
Intelligence before consensus
We translate the frontier of scientific research into technology intelligence years before it reaches mainstream market awareness.
Deep Scientific Monitoring
We continuously track published and unpublished research in computer science, physics, and biomedicine — identifying breakthroughs before they surface in earnings calls or analyst reports.
Quantified Predictions
Each research signal is assigned a confidence score that measures alignment with commercial applications. In backtested simulations, high-confidence predictions have outperformed market benchmarks over the period studied.
Research for Investors
We partner with venture capital firms, family offices, and institutional investors to identify which emerging technologies merit attention before the crowd arrives.
How It Works
We map the scientific landscape and predict what rises
Our system embeds millions of papers into a unified research landscape, grouping them by semantic proximity across disciplines. We then identify clusters of papers our model believes will drive future commercial breakthroughs and mark them as predictions. The maps below show what happens when we overlay predictions made in 2018-2021 with articles published in 2022-2026. On the left, all published articles. On the right, all articles that received at least 300 citations. The same clusters remain, showing our model reliably points to work that goes on to matter most.
Every topic our model predicted as significant during 2018-2021, overlaid on the published research landscape of 2022-2026. Orange dots show our predictions.
Filtering to show only those papers that received 300 or more citations, the clusters hold. Our predictions disproportionately capture the research that becomes most impactful.
Each hexagonal cell represents a distinct research topic cluster. Orange dots are topics and papers flagged by our prediction model prior to citation accumulation.
Scientific Rigor
Research-grounded conviction
Our predictions are not based on market sentiment or analyst reports. They are derived from systematic analysis of primary scientific literature, including papers from the world's leading research institutions that have not yet reached commercial awareness.
Literature Ingestion at Scale
We process tens of millions of preprints and peer-reviewed papers across disciplines, identifying technology-commercialization signals before they enter the mainstream narrative.
Multi-Domain Synthesis
Breakthroughs rarely emerge from a single discipline. We connect dots across computer science, physics, and biomedicine to identify convergence opportunities.
Verified Prediction History
Every prediction is dated, recorded, and scored, creating an auditable track record that speaks for itself.
Case Studies
Dated, recorded, and ahead of the news
When our predictions keep landing in the same region of the research landscape, the signal is one worth acting on. Two clusters that formed well before the market noticed.
2D Quantum Materials
We flagged 2D electronics before the NSF did
One prediction landed on low-dimensional electronics in Dec 2024. Over the next nine months seven more landed in the same region of the map — the same neighbours, the same cluster, tightening. In Sep 2026 the NSF opened its $150M ELEGANT program in exactly that area.
21 months
from first prediction to NSF funding
Each hexagon is a research topic; yellow marks a prediction, and the pale halo is the region it covers. The same frame, nine months apart — what changes is how much of it we had called.
AI Code Generation
8 months
ahead of Cursor's Series A
We flagged AI coding before the round that repriced it
Our predictions clustered around AI-augmented software development through 2024 — one in Dec 2023, five more by that July. Cursor raised at a $400M valuation the following month, and was later bought by SpaceX at $60B, roughly 150× that mark.
Working Paper · 1,171 Managers
We tested it on other people's portfolios
The obvious objection to any research signal is that its backtest was built to succeed. So we stopped picking stocks. Each quarter from March 2020 we re-weighted the reported US equity holdings of every 13F filer meeting fixed data rules — asset managers, pension funds, banks, insurers, hedge funds — keeping their holdings and their prices, and changing only the weights. The score uses no prices, no fundamentals and no analyst data, and nothing in it postdates the quarter it is used in.
+1.31 pp
Added a year (random: +0.33)
85%
Of 1,171 books improved
458
Beat random at p < 0.05
+0.59 pp
Alpha after style (t = 2.3)
Put the same holdings through the same procedure in a random order and the tilt earns +0.33 points a year. Not one draw of a thousand reached what the real ordering earned. Run against the real index funds rather than a manager's book, the same rule adds 2.01 points a year on the S&P 500 and 2.69 on the Nasdaq-100 — 1.16 and 1.95 after style.
Hypothetical backtest on public filings, 26 quarters to October 2026, gross of trading costs. The benchmark is each manager's reported quarter-end book, not their realised performance, and only long US equity positions are visible. Part of the gain is a growth exposure; the figure above is what survives market, value and size. Past performance is no guarantee of future results.
Registered Forward Test
Dec 2026
First rebalance on holdings that do not yet exist
The scores are sealed before the returns exist
A backtest asks you to trust the people who ran it. So we registered a one-year forward test before the first rebalance: a sealed Zenodo record holds the SHA-256 hash of the protocol and of every score file, each deposited before the holding period it covers. When the files are released, anyone can check them against a hash published in advance and recompute the result from public filings and prices. Results are due by May 2028.
Case Study · One Book
What it looks like on a single manager
The census measures 1,171 books at once, which is the right way to test a claim and a poor way to picture one. Here is the same overlay on one of them. Gary Anderson — author of The Janus Factor (Bloomberg Press) and winner of the Charles H. Dow Award — picks stocks on market signals: momentum and relative strength. He challenged us to improve on his selection using scientific signals alone, scored independently of anything his own method looks at.
Same stocks, same days, same prices — only the weights change
We tilt each holding toward its Innovation Lens score, with the threshold chosen walk-forward rather than in hindsight
29.3%
Re-weighted, a year
24.2%
His portfolio, untouched
+5.1 pts
Added per year, Jun 2018 – Aug 2026
An illustration, not evidence, and deliberately not part of the paper above. One book is one draw: once the threshold search is paid for, this result's permutation p is 0.104, so on its own it would not clear the bar the census clears. Hypothetical walk-forward backtest, before tax, not a live account — his trades are real, the re-weighting of them is a simulation, and the edge arrives in three of the eight years rather than evenly. Past performance is no guarantee of future results.
Private Markets
Our edge multiplies in private markets
If our predictions generate alpha in liquid public markets — where prices reflect millions of participants — imagine the advantage in pre-IPO companies where information asymmetry is far greater and pricing is far less efficient.
Earlier Signal Capture
We identify technology themes 3–7 years before they reach public market consensus, precisely the window where private equity returns are made.
Unpriced Scientific Moats
Private companies often have deep ties to academic research. We identify which scientific relationships translate into durable competitive advantages.
Portfolio Construction Clarity
We help VCs answer: “Is this technical thesis actually grounded in the state of science?” — a question that separates category leaders from also-rans.
Work With Us
Ready to invest at the frontier?
We work with a small number of investment partners. If you are a VC, family office, or institutional investor seeking a durable scientific edge, we would like to speak with you.
The dataset is listed on Eagle Alpha. Access is by invitation — send us your email and we will open it for you.