Bulk-export UCC (Uniform Commercial Code) financing-statement lien filings from Connecticut and Colorado state open-data portals — debtor, secured party, filing date, status, one normalized row per filing across both states in a single run. No API key, no login.
Three defects found in real customer runs after this Actor went public
and priced 2026-09-23 (30-day success rate flagged by the daily
report). All three verified live against data.ct.gov/
data.colorado.gov before and after the fix.
Fix: Colorado individual debtors billed with an empty
debtor_name. CO's debtor and secured-party datasets store people
in firstname/middlename/lastname/suffix when there is no
organizationname — map_co_row read only organizationname, so
every individual CO debtor shipped a row with debtor_name == ""
(still a billed result-emitted event). New
co_person_or_org_name() composes "Last, First Middle Suffix" the
same way CT's business_or_individual_name already did; applied to
both debtor_name and secured_party_name, and to the
debtor_name_contains filter so individual debtors are searchable
too.
Fix: no sort order — a capped run returned the oldest filings in
the window, not the newest. Neither state's Socrata query passed
$order, so a maxResults cap smaller than the date window
returned whatever Socrata's undefined default order produced —
verified live to be oldest-first. client.soda_page now accepts an
order param; both adapters request <date column> DESC
(dt_accept DESC for CT, filingdate DESC for CO), so a capped run
surfaces the newest filings — the use case this Actor is sold on.
Fix: dates shipped as full timestamps, not the YYYY-MM-DD the
README documents. Both states' date columns are Socrata "floating
timestamp" types and come back as 2026-09-08T00:00:00.000 over the
wire. New adapters._to_iso_date() normalizes filing_date and
lapse_date on every row for both states, matching the documented
output schema.
0.1 — pre-publish QA fixes (2026-09-22)
Two defects found by a pre-publish deep QA run (NYA3N3Qv8pab9rYgw,
states=[CT,CO], daysBack=120, maxResults=200) before this Actor
was ever made public. Not shipped to any customer.
Fix: CO adapter dropped extra debtors on multi-debtor filings.build_co_joined_filings kept only the first joined debtor row per
fileid, silently discarding the rest. It now emits one row per
(filing, debtor) pair — matching CT, whose source dataset already
ships one raw record per debtor per filing. Decided this is the
correct contract (not "aggregate to one row per filing"): it's CT's
native shape, it's the more useful shape for the lead-gen use case,
and it's what the deep run actually exercised. README, spec.md, and
Store copy corrected to say "one row per filing-debtor pair"
throughout, including a billing callout next to the price — PPE's
result-emitted event bills per row, so a multi-debtor filing costs
more than one row.
Fix: a 2-state request could starve the second state of the
entire max_results budget.main._run handed the whole
remaining cap to whichever state was fetched first; with
states=[CT,CO] and maxResults=200, CT alone consumed all 200
before CO was ever queried, even though CO had matching filings in
the same window (verified directly against data.colorado.gov).
max_results is now split evenly across cfg.states up front, with
a state's unused share rolling over to the next state so the global
cap is never wasted.
0.0 — scaffold (2026-09-22)
Actor skeleton created (actor-scaffolder). Boots, reads input, pushes one
{"placeholder": true, ...} dataset row. No real scraping logic yet.
See docs/specs/ucc-lien-filing-scraper/tasks.md for the implementation
plan (T02-T10).
