How a report is produced — and what our evidence proves
Each report answers one bounded question from public sources and shows where every finding came from. Owner tests are not sales
or independent benchmarks. Independent customer results: none verified yet.
Pipeline
QuestionOne supplied company, vendor or product page and one bounded question — never open-ended discovery.
Acquire public source dataFetch only company-linked public pages and supported public feeds, within a fixed request, byte and time budget.
ExtractRead the specific fields the question needs (for example a job’s publication field or a plan’s displayed price).
NormalizeConvert values to a stable JSON shape: UTC timestamps, canonical URLs, explicit nulls for missing values.
VerifyKeep a finding only when the checked source supports it; otherwise mark the outcome partial, unknown or invalid.
Attach evidenceEach finding carries its source URL, check time and a SHA-256 of the checked source body where the schema provides it.
Return JSONOne report item in the shape of the published report schema, stored in the run’s dataset and OUTPUT record.
Charge for an eligible resultAt most one report event per run, and only for the eligible outcomes listed on each service page.
Source handling
Only public pages linked from the supplied company or product are read. robots.txt is checked; a disallowed or
unavailable robots file is reported (for example ROBOTS_DISALLOWED) instead of being bypassed. No logins, forms,
checkouts or paid data sources. Every run stops at a fixed budget:
Request bounds per service (from the public capability index)
UTC time the source was checked — observation freshness, not publication time.
published_date
The verified date read from the source field named in evidence_field, with the raw value kept in raw_published_date.
date_meaning
What that source date means: last_published, first_published, date_posted or
released. A last_published date can reflect a republication, not the first posting.
Time zone
All timestamps are UTC. Lookback windows are UTC calendar days.
Undated listings cannot establish recency and are not counted as verified recent openings.
Partial coverage
coverage_complete: true means every discovered supported source was checked within the budget — it does not mean the
whole company or the whole web. coverage_scope states exactly what a completeness or negative claim covers. When a
budget, output cap or blocked source stops a check early, the report says so and stays partial or unknown.
Conflicts
Where sources disagree or duplicate evidence conflicts, the finding is not silently resolved: it is kept out of the verified set
and the outcome stays partial or unknown. Conflicting fields remain unknown rather than being guessed. (Handling of conflicting
duplicate evidence was tightened in the 8 October 2026 Hiring release; see release notes.)
Unknowns
Unknown is not negative. A blocked page, an unsupported source or missing documentation produces an unknown outcome and no
eligible report event — never a “not hiring”, “no integration” or “no contact” claim. Each service page has the full status table;
the statuses per service:
These classes are never pooled. A result in one class does not stand in for another. There is no uptime figure and no SLA.
Evidence class 1 of 4
Local fixtures
Offline candidate checks before the reliability release: Hiring 131, Contact 67 (plus client and workflow-code checks). MCP argument templates and the recorded examples were validated offline against their schemas. This site’s build re-validates every recorded example and minimal input against its schema.
What it proves
Parsing, classification and schema logic behave as intended on fixed inputs.
Recorded reports and minimal inputs validate against the published schemas.
Instruction templates carry the stated build, memory, timeout and cap.
Hiring Signals cloud validation cohort, (build
CTATHD9BRo5auMxMb) — Owner-funded sample, small, not a precision benchmark; coverage depends on supported ATS/careers sources.
Measure
Value
Companies in sample
24
Runs that succeeded
100%
Runs with an eligible result
41.7%
Schema-valid reports
100%
Recorded reasons for non-eligible or limited results
DISCOVERY_PAGE_LIMIT 6
NO_SUPPORTED_JOB_SOURCE 4
DETAIL_CHECK_LIMIT 4
ROBOTS_* (disallowed 1, unavailable 2) 3
HTTP_403 2
MALFORMED_SITEMAP 1
OUTPUT_LIMIT 1
PLACEHOLDER_LISTING 1
Hiring Signals release tests, 2026-10-08 (build 2.0.18) — The Stripe result is a partial positive, not exhaustive company coverage.
Case
Status
Report events
Ashby company
VERIFIED
1
Greenhouse company (Stripe)
PARTIAL
1
Unsafe input
INVALID_INPUT
0
Reports may precede final event charging; their billing fields alone do not establish a settled debit.
What it proves
The published Actor ran on Apify cloud with the stated build, input and cap, and produced a schema-valid report.
For each recorded run: what the checked sources showed at that time.
Unsafe or invalid input produced no report event in the recorded tests.
What it does not prove
Sales, demand or revenue — the publisher paid for these runs.
Independent benchmark accuracy or broad coverage.
Current results for any other company or for the same company today.
Evidence class 3 of 4
Read-only metadata checks
Last checked with an anonymous
GET https://api.apify.com/v2/acts/{actorId} per Actor (no token, no run). Checked: public visibility,
default build, event price, pricing model. Recheck the Apify Store page before purchase.
What it proves
On the check date, each of the seven Actors was public, with the stated default build, event price and pricing model.
Prices and builds shown on this site match that check.
What it does not prove
That the values are unchanged now — there is no continuous monitoring.
That a run will succeed, or anything about report quality.
Availability or uptime.
Evidence class 4 of 4
Independent customer results
None verified yet.
Confirmed customer revenue: $0.00.
Interest, clicks, copies and installs are not revenue.
What it proves
Nothing yet — this class is empty.
What it does not prove
Interest, page views, clicks, copies and installs are not purchases and are not counted here.