- The run timeout is now a BUDGET, not a backstop. Before this, one company
could book about 3,070 s of retry ladders (USAspending ~1,114 s, OSHA
~1,462 s, SEC ~491 s) on top of ~3,080 s of shared index downloads — WARN's
seven state files alone booked ~2,537 s — against a 3600 s run timeout at a
default of 200 companies. A run the platform kills never gets to write a
status message, so it delivered a blank run card; that is exactly what
osha.gov's CloudFront outage did to the sibling actor on 2026-09-06/07. Every
attempt is now planned against the time that is actually left, every url has
its own cap, and the run is carved into phases with the most expensive and
most droppable source (the state WARN feeds) last.
- No single company can eat the run any more: each gets a fair share of what is
left, recomputed as the run goes, so a fast company donates its slack to the
ones behind it.
- A source the clock never asked is now its own answer. "The source
answered and this company is not in it", "the source refused" and "we never
looked" are three different facts, and the report keeps them apart: a new
sourcesNotQueried field per row, a per-section skipped-time-limit /
partial-time-limit status, and a run status that NAMES the companies it
never screened and the sources it never reached. A company no source could be
asked about is not shipped as an empty report at all — it is named as
unscreened.
- The FDA warning-letter table is paged newest-first, so a download the clock
cut short is reported as partially read: letters it found are real, an empty
FDA section from it is "not fully checked", never "clean". State WARN feeds
the clock never reached are listed separately from feeds that failed and from
Indiana's permanent platform block.
- A 4xx that is not a rate limit now leaves the retry ladder immediately instead
of being retried four times to be told the same thing.
- Rows already collected always ship: a run stopped by the clock still charges,
pushes and exits with an honest
partial-time-limit status.
- Every run now finishes with a status message that names the federal sources it
actually read. An empty risk section can mean "nothing found" or "the source
did not answer", and only one of those is a finding — the message now lists
the sources that answered, names any that did not, and states that their
sections are "unavailable", never "clean". A run where no federal source
answered produces no report at all rather than five empty sections.
- Indiana is named on every run, not only when something fails: the Apify
platform does not resolve Indiana's WARN site, so those notices can never be
read from a run here. That permanent gap is now disclosed in the status
message instead of being invisible.
- A run stopped by its charge limit now states that the unscreened companies
were NOT checked and must not be read as clean, and input rows with no usable
company name are counted in the message rather than dropped in silence.
- README: the company-report price is now stated as a number — $0.03 per report,
$30 per 1,000, with the real plan-discount steps and a worked "100 suppliers
costs $3.00" example. It previously explained the billing rule without ever
giving the figure.
- WARN coverage is now six state feeds, not seven: Indiana has been dropped.
The Apify platform does not resolve the Indiana WARN site, so that feed could
never be read from a run here. It is now named as skipped instead of quietly
contributing nothing.
warnStatesCovered and the rest of the report are
otherwise unchanged.
First release.
One consolidated federal due-diligence report per US company, joined across five
official sources:
- Federal awards from USAspending.gov (contracts and financial assistance),
with the recipient's UEI, DUNS, registered address, business types and parent.
- OSHA inspection and violation history from the federal IMIS establishment
database, all states, with a link to each inspection.
- FDA warning letters from the published fda.gov table, including whether a
letter has been closed out.
- WARN layoff notices from seven state feeds (California, New Jersey, Maryland,
Indiana, Alaska, South Dakota, District of Columbia).
- SEC EDGAR filing presence: CIK, tickers, exchanges, SIC industry, state of
incorporation and recent material filings.
Plus a match-confidence score (0-100) with its full reasoning, and risk flags
you can filter on.
Matching is deliberately strict. Short and generic company names are held to a
higher bar than the default and are not billed when they cannot be joined; a
supplied UEI or CIK that resolves to a different entity is flagged rather than
ignored; a state you supply corroborates the identity but never filters the
findings; and every award row is re-checked against the matched entity before it
can enter a total.
Input accepts a pasted list, another run's dataset, or a CSV link. No API key.
