Version 1.2 · reviewed 2 Oct 2026
How we know what we say we know.
Every number, badge and score on ASI has a rule behind it. This page holds all of them in one place: what each page is built from, how a figure is computed, and what it does not claim.
Prototype notice. The companies, people, investors and figures on ASI are fictional while the product is in development. The methods below describe how real data will be treated, and how the prototype data is labelled.
Start here / The shared vocabulary
Six states a fact can be in.
- Verified
An editor checked the fact against a primary source, or against two independent reputable ones. The date of the check is shown.
- Disclosed
The company, an investor or a regulator states it. ASI records it as stated.
- Reported
A credible third party reports it and it is not yet confirmed by a primary source. It is counted, and labelled.
- Modeled
ASI calculated it from other facts, such as hiring data. Directional only, with the formula on the page.
- Estimated
ASI judged it from partial evidence. Labelled and timestamped, never mixed into a total without saying so.
- Unknown
Not publicly available. Shown as unknown, and never counted as zero.
Confidence · 80–100
High
Several agreeing sources, at least one primary.
Confidence · 60–79
Medium
Credible sources, partly corroborated.
Confidence · Below 60
Low
Thin or single-source evidence. Shown, and kept out of views that filter to medium and high confidence.
01 / Foundations
v1.2 · reviewed 2 Oct 2026
Evidence states, confidence and unknowns
The rules every page follows before it shows a fact, an amount or a score.
Built from
- Primary sources: company notices, filings, investor announcements, regulator records
- Secondary sources: credible reporting, labelled as such
- ASI calculations, labelled Modeled or Estimated
How it works
- A fact keeps its state
- Each fact carries one of six states (Verified, Disclosed, Reported, Modeled, Estimated, Unknown). A total never silently mixes states: where it does, the page says which are included.
- Unknown is not zero
- A missing amount, headcount or date is shown as unknown and left out of averages. It is never filled with zero or a guess.
- Confidence measures evidence, not quality
- A confidence score from 0 to 100 describes how complete and consistent the evidence is. It does not rate a company, an investor or a market.
- Sources travel with the number
- Where a record shows an amount or a status, it lists the sources behind it and the date they were last checked.
- Currency
- Amounts are normalised to US dollars at the announcement-date rate. Euro and sterling views convert that figure at a fixed display rate; they are not a second source.
- Dates
- Periods are stated on the page (for example 1 Jan to 2 Oct 2026). Comparisons use a window of the same length, or the same dates a year earlier for year to date.
What this does not claim
- ASI does not infer gender, ethnicity, nationality or motive from names, photos or behaviour.
- High confidence is not a guarantee that a figure is right.
- All data on the site is currently a fictional prototype dataset; each page is marked.
Changes
- 2026-10-02Added the six-state vocabulary and the confidence bands to one place.
02 / Companies & capital
v1.2 · reviewed 2 Oct 2026
Company directory, profiles and comparison
How a startup becomes a record, what “active” and “stage” mean, and how badges and comparisons are produced.
Built from
- Company websites and notices
- Funding announcements, deduplicated by deal ID
- Operator interviews and hiring pages
- Editorial review
How it works
- Covered company
- A unique startup record with one primary sector and dated operating evidence in the last 12 months. It is not a census of all startups.
- Company stage
- Early (pre-seed and seed), Scale-up (Series A) or Growth (Series B and later), taken from the latest known round. It reflects financing, not revenue maturity.
- Verified record
- An editor reviewed it against at least one primary source or two reputable secondary ones. A Reported record has not yet had that review.
- Status badges
- Derived from the record, never typed by hand: Unicorn (a sourced valuation of $1bn or more, not shut down or acquired), Shut down, Acquired, Listed, and Recently funded (a round in the last 180 days).
- Valuations
- Shown only when a source reports one. Otherwise the field stays empty.
- Comparison
- Compares two to four companies on the same fields. A difference is highlighted when values differ; good and concern tones come from rules on the fields, and the insights are computed from the data, not written by hand.
- Duplicates
- Matched on normalised name and website domain before a record is added.
What this does not claim
- A badge is a fact about the record, not a judgement of the company.
- Headcount is an ASI estimate shown as a range.
- Social links and documents are listed only when the source is known.
03 / Companies & capital
v1.2 · reviewed 2 Oct 2026
Funding, deals and the Deal tracker
How a transaction is recorded, scored and summed, in both the Funding & deals explorer and the Deal tracker.
Built from
- Company announcements
- Investor confirmations and portfolio pages
- Facility and regulatory filings
- Credible press reports
How it works
- One round, one record
- Each round is recorded once with its announcement date, closing date (if known), lead investor, participants, instrument and amount. Duplicates are merged by deal ID.
- Confirmed or Reported
- Confirmed means a primary source states the amount. Reported means a credible third party does and it is not yet confirmed. Reported rounds are counted and labelled.
- Confidence score
- Starts from the amount’s state (Disclosed about 91, Reported about 79, Modeled about 67, Estimated about 58) and moves with the number of independent sources. Bands: High 80–100, Medium 60–79, Low below 60.
- Capital figures
- Sums use disclosed and reported amounts. Undisclosed amounts count towards the deal count only. Project debt, grants and acquisitions are marked and can be filtered out.
- Median round
- The middle value of round sizes in the view, shown beside the total because one very large round can lift an average.
- Large for its group
- A round more than 1.5 interquartile ranges away from the middle half of rounds, on a log scale. It can be switched off, and then it is left out of every figure.
- Cross-border
- The lead investor is based in a different country from the company. Each round is allocated once, by its lead’s domicile.
- Change figures
- Compared with a window of the same length before it. Year to date compares with the same dates a year earlier. Where no earlier window exists, the change shows “—”.
- Weekly chart
- Capital announced in each week from Monday, under every filter except the period.
What this does not claim
- Round size is not company quality or ecosystem health.
- A quiet week can mean few announcements, not a missing record.
- Discovery and disclosure lag mean the newest weeks are partial.
- Capital raised is not money spent, and not investor loss.
Changes
- 2026-10-06Deal tracker added; it reads the same records and rules as Funding & deals.
04 / Companies & capital
v1.2 · reviewed 2 Oct 2026
Investors and investor activity
How investor records are verified and how activity measures such as cheque size, lead rate and reserves are estimated.
Built from
- Fund announcements and regulatory filings
- Portfolio pages and deal disclosures
- Direct confirmation from the investor
- The transaction records
How it works
- Verified investor record
- Combines fund announcements, filings, portfolio pages, deal disclosures and direct confirmation. Estimates are labelled, timestamped and confidence-scored.
- Cheque and deployed capital
- A lead takes two shares of a round and each other participant one, so cheques and deployed capital are estimates.
- Lead rate
- Lead slots over all slots the investor held.
- Later rounds
- Slots in a company’s later rounds. The reserve (Modeled) is half of later-round capital.
- Repeat co-investors
- Rounds that include investors who had co-invested before.
- Time between deals
- The median days between one investor’s consecutive deals.
- Dry powder (Modeled)
- Deployed capital multiplied by 1.5 to 2.4. A range, not a fund balance.
- Activity score
- 40% deals, 30% capital, 15% lead rate, 15% later rounds. A description of the period, not a ranking of quality.
- Emerging and inactive
- Emerging: deployment grew from the first half of the period to the second, for investors with three or more deals and $3m or more early. Inactive: two or more earlier deals and none for ten months.
What this does not claim
- Participation is not endorsement, and counts overlap across syndicates.
- Network diagrams show the investors who share the most rounds, not the whole market.
- Fund-raising signals and fund launches are tracked as events, not as committed capital.
05 / Markets & places
v1.2 · reviewed 2 Oct 2026
Intelligence dashboards and filters
How the overview’s headline figures behave when you change a filter.
Built from
- Monthly funding, deal and investor series
- A pool of recent rounds
- Country and subsector breakdowns
How it works
- Data states
- Disclosed, Reported, Modeled and Estimated figures can be shown together or filtered to one state.
- Filtered views
- Market, sector, stage and data-state filters apply fixed, modeled shares to the example series. They show how a view would shift; they are illustrative, not separate datasets.
- Compare
- Adds the previous period beside each figure and chart tooltip.
- Saved views and alerts
- Kept in this browser. Nothing is emailed until member accounts open.
What this does not claim
- A filtered figure on the overview is a modeled share, unlike Funding & deals, which recomputes from records.
06 / Markets & places
v1.2 · reviewed 2 Oct 2026
Markets and sectors: momentum and market ranges
How a sector score is built and why market sizes are ranges.
Built from
- Quarterly funding by sector
- Company formation counts
- Employee-growth proxy from hiring data
- Operating survival and time between rounds
How it works
- Momentum score
- A weighted sum of five parts, each scored 0–100: funding velocity 30%, growth proxy 25%, survival 20%, investor diversity 15% and reach 10%. It describes the last period and does not predict the next.
- Capital efficiency
- Funding raised per unit of modeled operating cost, shown as a multiple. Above 1.0× means a sector turns capital into headcount and revenue faster than the set average.
- Concentration
- The share of funding going to the top five companies in the set.
- Market sizes
- Ranges, never point estimates. The low case counts sites already served by a comparable product; the high case adds sites a new financing model could reach. Each range lists its inputs and a version.
- Formation
- New companies counted by founding year or quarter. The most recent quarter is an estimate.
- Filters
- Market, stage and confidence filters apply fixed modeled shares to each sector.
What this does not claim
- Scores are descriptive, not predictive.
- Revenue and growth are modeled where direct figures are unavailable.
- A market range is a scenario with assumptions, not a forecast or a total addressable market claim.
07 / Markets & places
v1.2 · reviewed 2 Oct 2026
Sector directory and sector profiles
How companies are placed in a sector, and what sector counts do and do not include.
Built from
- Reviewed company records
- Funding announcements
- Hiring pages and operator interviews
- Self-reported impact statements
How it works
- One company, one primary sector
- Aggregate company and funding totals use a single reviewed primary sector. Secondary tags make a company discoverable but never add a second company or duplicate a deal. Mixed-sector rounds inherit the company’s primary allocation.
- Taxonomy
- Ten sector families, each with subsectors, inclusion rules and boundary cases. Version 1.2. A boundary case is reviewable.
- Deal change
- The tracked deal count in the period minus the count in the prior window, shown only when the baseline is large enough. Families with too little evidence are withheld from the maturity chart.
- Data completeness
- High, Medium or Developing: how much of a family’s evidence has been reviewed. It is not a measure of business quality.
- Impact
- Reported example metrics only. Capacity is installed nameplate, not energy generated. No avoided-emissions total is built from installed capacity; a total needs a baseline, a system boundary and a measurement date.
- Scenarios
- Market-size scenarios multiply stated inputs (sites by annual spend) and carry a version. They are sensitivities, not forecasts.
What this does not claim
- Covered startups are not total industry size.
- Funding concentration does not measure ecosystem value.
- Sector comparison needs matched operating models: assets, subscriptions and transactions are different businesses.
08 / Markets & places
v1.2 · reviewed 2 Oct 2026
Countries and ecosystems
How country figures are counted and why capital-city data is not national data.
Built from
- Company headquarters records
- Funding announcements
- Investor domicile
- Public policy and programme notices
How it works
- Country assignment
- A company belongs to the country of its headquarters. A company’s operating markets are recorded separately and are never summed as unique companies.
- Hubs
- Cities are counted from headquarters records. Records that are only partly geocoded are grouped as “Other”.
- Investor origin
- The domicile of the investor, not the founder’s nationality. A round is allocated once, by its lead.
- Newly formed companies
- Counted by founding year, not by the year the record was added. Discovery lag makes the latest year partial.
- Workforce
- An estimate from a modeled subset of records. It is not a national jobs figure and is not extrapolated to the whole country.
- Briefings
- Country briefings and snapshots are generated from the same figures as the page, with the filters and date in the image.
What this does not claim
- Coverage counts describe the directory, not the whole market.
- A rise in documented startups can mean better discovery, not a stronger ecosystem.
- Policy and regulatory notes are reading guides, not legal or investment advice.
09 / Markets & places
v1.2 · reviewed 2 Oct 2026
Signals and watchlists
What a signal is, how strong it is rated, and how opportunity and risk scores are labelled.
Built from
- Hiring postings
- Leadership changes
- Product releases
- Expansion notices
- Partnerships and news
- Risk indicators
How it works
- Signal
- A dated event about a company in one category (hiring, leadership, product, expansion, partnership, news, risk), with its evidence and a confidence label.
- Heatmap intensity
- The sum, over a company’s signals, of recency times confidence. Recency halves every 20 days, and a repeated signal is weighted ×1.15. High is 2.0 or more, Medium 0.9 or more, Low above zero.
- Confidence labels
- Verified (85 or more with two or more primary sources), High (80 or more), Medium (60 to 79) and Reported (below 60).
- Opportunity and risk scores
- Modeled from the same evidence and labelled directional, never a forecast. Scores shown for named companies are editorial examples.
- Alerts
- A rule matches signals by category, confidence and company set. Each rule shows what it would have flagged in the last week.
What this does not claim
- A signal is evidence of change, not proof of outcome.
- Alert precision is measured on past flags and can change.
- Lists, rules and digests are kept in this browser; nothing is emailed yet.
10 / Markets & places
v1.2 · reviewed 2 Oct 2026
Research reports and datasets
How a report separates evidence from analysis, and how it is cited.
Built from
- The transaction records
- Cohort samples
- Primary reporting and interviews
How it works
- Evidence colours
- Charts mark amounts as disclosed, reported or estimated, using separate colours, so a total shows its mix.
- Source notes
- Every report lists its sources and a methodology panel, and offers APA and BibTeX citations.
- Chart controls
- Year, region and sector selectors scale totals by fixed shares in the prototype; the downloads give the data behind the chart.
- Premium material
- Workbooks, chart packs and some appendices are members-only. The executive summary and key figures stay free.
- Revisions
- Reports carry a version, a revision history and a corrections section.
What this does not claim
- Headline samples and chart totals can differ where a report says so.
- Analysis is labelled as analysis and names who holds the view.
11 / People & work
v1.2 · reviewed 2 Oct 2026
People and founders
What a person profile may contain and how it is kept fair.
Built from
- Public professional information
- Company and investor records
- Published interviews
How it works
- Public professional facts only
- Role, organisation, career, investments and board seats. No personal contact details unless the person published them for business use.
- Self-identified background
- Shown only where the person has said it publicly. ASI never infers it from a name or a portrait.
- Network
- Shows recorded links (co-founders, co-investors, boards). A link is a fact about the record, not an endorsement.
- Corrections
- Takedown and correction requests are handled within 7 days.
What this does not claim
- People with no portrait show initials, never a stock image.
- Counts cover the people ASI has recorded, not everyone in the ecosystem.
12 / People & work
v1.2 · reviewed 2 Oct 2026
Startup jobs and hiring
How listings are sourced and checked, and what a hiring figure means.
Built from
- Employer careers pages
- Named public sources
- Employer identity checks
How it works
- Listings link out
- ASI hosts no applications, collects no CVs and never charges candidates. Each job links to its employer’s page or a named source.
- Last checked
- Every listing shows when ASI last checked its source. An identity badge confirms the careers domain belongs to the company, not that a vacancy or its terms are real.
- Open roles
- Active listings only. Expired roles stay on record for reference and are excluded from totals and results.
- Salary
- Shown only where the employer states it, in its original currency. Salaries are never converted or estimated.
- Hiring signal (Modeled)
- A trend in public postings over a recent window. It is not headcount growth, filled roles or a forecast, and repeat postings can change it.
- Departments
- A grouping of role families: Technical, Product & analytics, Field operations, Commercial, Customer success and Finance & people.
What this does not claim
- Team size is an ASI estimate, not an employment term.
- Employer copy is shown as the employer wrote it and is not verified fact.
13 / Learning & corrections
v1.2 · reviewed 2 Oct 2026
Startup Graveyard
What counts as a documented closure and how reasons are handled.
Built from
- Company operating-end notices
- Corroborating investor, board or partner statements
- Customer-transition notices
- Public postmortems
How it works
- Confirmed closure
- Primary evidence that operations ended, plus independent corroboration. A quiet website or a lapsed domain is not evidence.
- Four outcomes, never mixed
- Confirmed closures, discontinued products, dormant or unconfirmed records, and acquired or integrated companies. An acquisition is not a failure.
- Three dates
- Announced, operating end and last verified. The operating end is recorded apart from the announcement.
- Reasons
- Shown only when attributed to a named source, with a source key such as [S3]. ASI does not infer causes and does not score failure.
- Lifespan
- Months between the operating dates. Year-only dates produce no lifespan and are left out of lifespan figures.
- Cohort lifespan
- A restricted mean over a five-year window that includes firms still operating (right-censored). It is not a survival forecast.
- Disputed records
- Move out of the confirmed view and every headline count while under review, and show the conflict.
What this does not claim
- The archive is documented history, not a census and not a ranking of ecosystems.
- English-language and venture-funded records are over-represented.
- Funding is capital raised, not losses.
14 / Learning & corrections
v1.2 · reviewed 2 Oct 2026
Corrections, right of reply and revisions
How a record is challenged, changed and kept honest.
Built from
- Correction requests
- Responses from named companies and people
- Editorial review
How it works
- Request a correction
- Name the field, give dated evidence and explain the change. Every correction form opens an email to ASI today; nothing is stored by the page.
- Right of reply
- A company, founder or investor named in a record may send a public response, published beside it. ASI invites one before publishing a sensitive record.
- No purchase of verification
- Submitting a correction, a deal or a listing does not buy verification, ranking or placement.
- Revision log
- Material changes are logged on the record with a version and a date. Previous versions remain accessible.
- Timelines
- Verified errors are fixed within 48 hours of a verified report; person takedowns within 7 days.
What this does not claim
- Email delivery, stored submissions and alerts start when member accounts open.
Reference / Glossary
Terms we use the same way everywhere.
- Covered startup
- A reviewed company record with one primary sector and recent operating evidence. Not a census.
- Disclosed funding
- Equity amounts stated by a company or investor. Undisclosed amounts, acquisitions and project finance are excluded unless noted.
- Illustrative
- A fictional prototype figure or entity, used to show how a page works. It must never be read as a real fact.
- Primary source
- The company, a filing, a regulator or the investor itself.
- Right-censored
- A firm still operating at the end of an observation window, counted as operating for the whole window.
- Restricted mean
- An average over a fixed window (here 0 to 5 years), used so firms still operating are not ignored.
- Modeled score
- A figure ASI computes from a stated formula. It is directional and carries its weights.
- Window
- The dated period a figure covers. Changes compare with a window of the same length.
Version 1.2 · reviewed 2 Oct 2026
Disagree with a method? Say so.
Methods change when evidence or readers show a better way. Every change is dated here and on the pages it touches.