How it works
A company name in. A cited verdict out.
MarketPrior resolves who a company legally is, sweeps 121 public sources in parallel, keeps only the matches it can tie to that entity, and scores risk and upside from cited records. The memo drops any sentence it can't cite.
01 · Resolve identity
Pin the legal entity first
Before any search runs, the company is resolved against company registers and global identifiers — LEI, national company numbers, SEC CIK — so every later lookup is about the right entity.
- Former names and consolidated subsidiaries come from the LEI record.
- Court and regulator searches try every name variant, most specific first — legal name, former names, the US subsidiary a foreign group litigates through, then the brand.
- Officers and beneficial owners are pulled from filings and screened as people, not just as names.
02 · Parallel sweep
121 public sources, all at once
Every retrieval step runs in parallel, dependency-ordered, under a per-host rate budget and a shared cache — a run takes as long as its slowest useful source, not the sum of all of them.
- The free screen — verdict, risk and upside — is computed first from whatever has landed and pushed to the report live.
- The full report follows on its own budget; slow upstreams finish in the background and their sections update when they arrive.
- Every source, its licence and coverage are listed on the public data sources page.
03 · Entity matching
A namesake is not a hit
List screens and court searches match on names, so most raw hits are someone else. Each watch-list hit is tiered against the identity already held for the company, its parent and its owners.
confirmed
Exact identifier
probable
Name plus corroboration
possible
Name only
04 · Scorers and forecast
Risk × upside, then a backtested forecast
Deterministic scorers turn the cited records into a risk score and an upside score; a transparent gate fuses them into one verdict. The language model never decides it.
Fused verdict
Four quadrants — high upside and low risk is the one you want. Confidence is the weaker of the two legs, and the verdict degrades gracefully when one is missing.
Outcome forecast
Two concrete questions — will the company file for new money within 18 months, and will it file nothing further within 24 — each with an 80% interval, the base rate and the model’s backtest metrics. No filing history, no forecast.
05 · Cited memo
Uncited sentences are dropped
Every finding becomes a numbered evidence entry with its source link and quote. The memo is written against that ledger and then checked sentence by sentence.
- A sentence with no valid citation is removed.
- A sentence with a number that appears in none of its cited evidence is removed.
- Citations to evidence that doesn’t exist are stripped, and a risk’s severity can’t exceed its evidence.
- With no language model available the memo is still produced — each evidence entry becomes a cited sentence.
06 · Monitoring and triggers
The run is where it starts
Monitored companies are re-screened daily, weekly or monthly. Only genuinely new events become alerts — each with its source link and quote — and the same event is never alerted twice.
The same trigger detectors — new private-offering filings, federal contract awards, officer changes — feed the signal-based lead feed. Set up continuous company monitoring.
Freshness
Served instantly, refreshed in the background
Reports are stale-while-revalidate. Opening one returns the cached sections immediately; any section past its time-to-live is refreshed in the background and updates in place.
- Each retrieval layer has its own time-to-live; a section is as fresh as its tightest layer.
- Roll-ups (verdict, memo) go stale as soon as anything they read is recomputed.
- Refreshes are single-flight: a second viewer, a re-run and the monitor all join the same job.
- Every section in the app and the diligence REST API reports its computed time, age and whether it is refreshing.
Time-to-live by layer
- sanctions — Sanctions lists, disqualified directors, owner screening
- litigation — Court dockets, gazettes, regulator enforcement feeds
- adverse media — Global news coverage
- hiring — Public job boards
- filings — Re-checked daily, recomputed only when a new filing lands
- people — Officers and founders from filings and public profiles
- web — Web traction, tech stack, GitHub, app stores
- profile — Company profile, competitors, identifiers, patents
FAQ
How the engine works
Does an AI model decide the verdict?
What happens to a name-only sanctions hit?
How are forecasts validated?
How fresh is a report?
What if a source is slow or down?
Next: read a how cited company diligence reports read, see pricing per report, or how we handle your data on the security and trust page.
Run cited diligence on your next company.
Name a company or founder and get a risk × upside verdict where every finding links to its source document.