File adequacy
Event date, first-seen date, and a revision flag. A zero is not a missing value unless the manifest says so.
01 Trial signal memo
Score the trialbefore anyone talks about alpha.
Send a trial file and a list of companies. Get a memo you can rerun: Kill, Renegotiate, or Continue. Every figure ties back to the rows. The fund keeps the alpha model.
- 39% Trials die for lack of a signal
- 27% Buyers convert more than a quarter of trials
- 4–6 mo Onboarding time for 36% of buyers
Neudata buyer surveys, 2026. Investment managers spent about $2.8bn on alternative data in 2025.
02 Method
A file in. A verdict out.
The same inputs rewrite the same memo.
-
01
File
A trial history plus the buyer’s list of companies. Two drops of the same weeks are better than one.
-
02
Map
Names resolve to a FIGI. Ambiguous stays ambiguous and counts as a miss.
-
03
Checks
Coverage, point in time, known events, and reported revenue after the filing date.
-
04
Memo
Kill, Renegotiate, or Continue. The rule id and the rows sit under every number.
A model may propose the column map. It does not score the file, and it does not see the fund’s alpha model.
03 Checks
Six tests. No blended score.
These are the checks a data team already runs by hand in the first weeks of a trial. The public benchmark is reported revenue, using the SEC filing date. There is no price panel and no Sharpe ratio.
Point at a test. The sample memo marks the figure.
Entity match
Weighted coverage of the buyer’s names. A fuzzy guess can propose a FIGI. It cannot write one.
Coverage shape
Whether the panel stays stable through time, including names that later delisted.
Point in time
Two drops of the same week agree, or the change carries a new first-seen time.
Known events
Closures, outages, or layoffs move the series the way the public record says they should. The event list is frozen before the run.
Public benchmark
Rank correlation with reported revenue, counted only once the filing date says the data was knowable.
04 Verdicts
First matching rule wins.
Renegotiate is the useful outcome in a live trial. It says what to fix, what to cut from the price, and which history must not be backtested.
01
Kill
The weighted match is too thin, or known events fail in the wrong direction. Stop the trial and send the misses back.
02
Renegotiate
A relationship may exist, but the history was rebuilt, the panel jumped, or the fit shrinks once publication lag is applied.
03
Continue
Mapping holds on small names, revisions are logged, and a pre-registered public test still points the right way after lag.
05 Who it’s for
Vendors first. Funds second.
Card spend is the account. Job postings are the demo. The memo format stays the same. The event library changes.
Now
Data vendors
You are mid-trial and the buyer cannot see a signal. The memo goes in the data room: what matched, what was rebuilt, and what to change before the quarter ends.
Next
Buy-side data teams
The same checks on every inbound file, inside your cloud. Re-run them when the live feed revises a past week.
Stays with you
The alpha model
We do not estimate a book, search theses, or certify a return. After Continue, that work is yours.
The trial dies while someone is still mapping names. Score the file, or say it is not evidence yet.All Data Labs
06 Pricing
Less than one feed you kill.
Design-partner memos are $2k to $15k. A dataset license is often $50k to $500k a year. There is no free information-coefficient preview.
Design partner
Trial memo
$2k–$15k
One dataset, one run you can rerun
- Six checks and a rule id
- Match table, with ambiguous names listed
- Point-in-time diff when you send two drops
- Known-event results for that dataset family
- Revenue comparison after the filing date
- A hash of the files and the thresholds
Later
Inside the fund
Same checks
On every inbound trial, in your cloud
- Your ticker list stays on your side
- Re-run when a past week is revised
- No upload of the alpha model
- Additivity against your book only after Continue
07 Landscape
The labs start after the file is clean.
Discovery catalogs list the dataset. Research platforms backtest a ticker-mapped panel. The trial usually dies while that panel is still being built.
| Job | Who already does it | What we do |
|---|---|---|
| Find a dataset | Neudata, Eagle Alpha | We start after you have a file |
| Backtest a mapped panel | AltHub, Exabel, Neuravest, CloudQuant | We score the file those labs assume |
| Scrape and structure the web | Bright Data, Oxylabs, Kadoa | We do not collect the trial for you |
| Read filings and transcripts | AlphaSense, Hebbia, internal tools | We do not write the research note |
08 Design partners
Have a trial that is stalling?
Send a note. The first memos are design partnerships on a live file, not a badge for a backtest.