What the research
actually measured.
Every figure QuikSignal states openly is computed from rows a reader can open. This page states what was researched, what the numbers count, and what they do not support.
One research run, 1 October 2026, across four markets.
Five stages, and what each one is allowed to conclude
The stages are not a diagram. Each one has a limit on what it can claim, and the limits are why the numbers above are small.
Research
The agents research each company and market you choose across named, credible sources, reading the document itself rather than a summary of it.
Assess
Each source is weighed for relevance and for what it actually is - an official record, the company’s own statement, or independent reporting - before anything is built on it.
Cross-check
A finding is checked against independent reporting, and the number of independent sources carrying it is counted and shown. Agreement is reported as agreement, never as proof.
Analyse
Ninety days of history are read backwards first, near-duplicate coverage of one story is merged, and what changes nothing is held back into a log you can open.
Synthesise
What survives is written into structured business intelligence - a short brief, each line tracing to its dated document.
A document is not a source
This is the distinction that makes the rest of the page readable, and the one most often collapsed elsewhere.
- A document read is a research input: one page, filing or posting that was fetched and assessed. It is a count of work done, not of evidence gathered.
- An independent source is an outlet that carries a claim on its own account. Reprints of one press release collapse into one source, however many outlets ran it, and the count shown beside a finding is the count after that collapse.
- A finding worth printing is one that survived the suppression stage. What was held back is logged with the score that held it, so the gap between the two numbers above is readable rather than asserted.
- Agreement between sources is reported as agreement. It is never reported as proof, and no finding on this site is described as verified.
Of 866 documents read, 50 were worth printing. That ratio is only meaningful because the numerator counts findings and the denominator counts inputs. Treating every reprint as a separate source would inflate both and tell a reader nothing.
Three source classes, behaving completely differently
An aggregate hides this. Filings produce most of what is printed; pricing pages produce almost nothing, because they genuinely do not change often.
| Market | Companies | Documents read | Worth printing | Researched |
|---|---|---|---|---|
| B2B SaaS | 4 | 214 | 14 | 1 October 2026 |
| Fintech and payments | 3 | 182 | 6 | 1 October 2026 |
| AI infrastructure | 5 | 174 | 14 | 1 October 2026 |
| Cybersecurity | 7 | 296 | 16 | 1 October 2026 |
What this evidence does not support
Stated here rather than in a footnote, because these are the sentences that decide whether a figure can be quoted.
- One research run. There is no earlier run to compare against, so nothing here measures change, direction or trend, and no index is computed from it.
- A cohort of 19 companies across four technology markets. It is not a sample of any market and no market-wide figure is inferred from it.
- Filings coverage is United States federal: SEC EDGAR and the Federal Register. Other jurisdictions are outside what was researched.
- Hiring comes from six applicant tracking systems. A company hiring elsewhere is invisible, and that absence is reported rather than filled in.
- Postings are not hires, and a pricing page change is not a revenue change. Each is evidence of a decision having been taken, nothing more.
- Nothing here forecasts, because nothing here keeps a record of whether past forecasts were right.
What has been worked up, and what has not
How much market research is noise
The print rate above, written up with its cohort, its method and the four pages a reader can add up themselves.
The Hiring Signal Index
Designed, and deliberately not computed yet. The page sets out the calculation and names exactly what is missing before a figure could be defended.
Run the same method on your own market
The four markets above are read by the same agents that run inside a workspace.