How Leadiy turns observations into publishable evidence.

Definitions set the denominator

Opportunity

A public, actionable signal that an organization or buyer may need software work, delivery capacity or a technology-related hire. Public aggregates count eligible imported vacancy/opportunity records, not every page seen by a crawler.

Vacancy

An opportunity record centred on a role, contract, project, RFP or selected hiring signal with enough public context to assess relevance. Hiring Pulse currently uses this imported vacancy/opportunity class.

Lead

A company or contact-centred record that may support direct outreach. Direct leads and their people, contact routes, source URLs and raw rows are excluded from public Hiring Pulse datasets.

Technology mention

A canonical catalog label matched from normalized technology evidence. Categories are multi-label: one opportunity may count once in several technology buckets.

Provenance fields are separate on purpose

Source classes, not private source rows

Eligible evidence comes from configured public opportunity surfaces: project and contract listings, procurement and RFP notices, vacancy and hiring pages, and structured opportunity feeds. Source access may change, fail or lag. Public datasets name the source classes but do not expose source URLs, raw rows, contact data, outreach copy or individual posting details.

Cadence and verification state

Leadiy checks configured sources on a 30-minute cycle. An item’s source publication time, access conditions, processing and matching can make discovery later than publication. Public editions advance only after a successful verified sync. If a check fails, Leadiy retains the last good aggregate, keeps its original verifiedAt, and marks freshness as delayed rather than presenting request time as new evidence.

Deduplication before counting

1. Prefer a stable provider or external record identifier when one is available.
2. Normalize canonical source URLs by removing non-identity fragments and known tracking parameters.
3. Where neither stable key exists, compare a conservative fingerprint of source class, organization, normalized title or project label and timing evidence.
4. Treat a changed version of the same record as an update, not a new opportunity, unless its stable identity demonstrates a new posting.
5. Within a record, count each canonical technology at most once even when it appears in several fields or aliases.

Ambiguous collisions are suppressed from derived claims until they can be resolved; deduplication reduces obvious repeats but cannot guarantee that two differently worded listings never represent the same underlying need.

Canonical technology matching

Technology evidence is normalized against Leadiy’s controlled technology catalog. Matching handles approved aliases, punctuation and common naming variants—for example, a catalog identifier remains stable when a source writes a display label differently. Exact phrase and token-boundary rules protect short labels from substring collisions. A record can match several technologies, but repeated mentions of the same canonical identifier still contribute one count.

Technology incidence is calculated as the published technology count divided by all eligible opportunities in the selected edition. Because categories overlap, shares do not sum to 100% and must not be called market share. The public edition does not infer technology co-occurrence unless a separately validated cross-tab is published.

Small-bucket suppression

Public breakdowns require at least 5 observations. A smaller bucket publishes count: null, sharePercent: null and suppressed: true; HTML renders <5. Leadiy does not replace suppressed values with zero, a midpoint or an estimate. Totals may therefore not reconcile exactly when one or more components are withheld.

Suppression reduces disclosure risk and discourages over-reading tiny groups. It does not mean the category is absent.

Budget outlier review preserves both views

Budget values are published only in comparable currency-by-unit groups. Fixed, hourly, daily, monthly, yearly and unspecified values never share a scale. Median and interquartile range describe the comparable group; only fixed-value groups can publish totals.

For a currency-unit group with at least 8 raw observations, Leadiy sets the upper-tail cutoff to the greater of Q3 + 3 × IQR and median × 100. An observation is flagged only when it exceeds that cutoff and contributes at least 20% of the raw midpoint total. Flagged values remain in the raw aggregate QA fields but are excluded from comparable median, IQR and totals. This prevents one very large tender from dominating a quotable summary while preserving evidence that the value existed.

The flag does not assert an error or prove that a large tender is unreal. If observation-level outlier metadata is unavailable in an older frozen edition, the page reports not-evaluable rather than claiming no outlier.

Known limitations

• Coverage depends on configured public sources, their availability and the observation process.
• Classification can miss implicit, novel or ambiguously named technologies.
• Multi-label incidence does not measure exclusive stack choice or technology market share.
• Organization counts, engagement intent and budgets are not technology-specific unless a page explicitly publishes a validated cross-tab.
• Unstated delivery preference remains unknown; it is never reassigned to solo or company.
• Historical trend language requires at least three stored snapshots spanning 14 days. A single edition supports dated level statements only.
• Suppression, deduplication and aggregate QA reduce risk but cannot eliminate every source error or identity ambiguity.

License, attribution and citation

Unless a dataset says otherwise, Leadiy’s published aggregate JSON and the original aggregate tables on these pages are licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Attribute Leadiy, link the canonical data page, name the edition, and preserve observedThrough, verifiedAt, denominator and suppression state.

This license covers Leadiy’s aggregate selection and expression. It does not relicense source postings, third-party text, brands, personal data or provider databases. Do not reconstruct suppressed or individual records from aggregates.