Problem
Two smoke cases are labelled in the source as regression guards for real
past bugs:
- Sweden / AB Volvo - disposal-group held-for-sale liabilities falling
outside the current+non-current derivation
- Singapore / Keppel - a "net assets" presentation misread as the
non-current liabilities subtotal
Neither guard can currently fire. Both filings leave one of the three
balance-sheet totals to be derived, and the identity check correctly
refuses to run in that case, because verifying assets = liabilities + equity against a total computed as assets - equity is circular. Both
therefore report not_applicable and pass unconditionally.
Their present value is limited to proving extraction does not crash.
What would actually close it
Pinned reference facts in src/openfilings/benchmarks.py, which is the
mechanism built for exactly this: exact figures that fail if normalized
output drifts.
That file states its own rule clearly, and it must be honoured:
Add a benchmark only after an independent source-document review; never
populate this list from the extractor output.
So this needs the figures transcribed from Volvo's and Keppel's published
annual reports by hand, not copied from what the library currently returns.
Related
benchmarks.py currently covers two issuers in total (Tesco and ASML) - see
the companion issue on broadening it.
Problem
Two smoke cases are labelled in the source as regression guards for real
past bugs:
outside the current+non-current derivation
non-current liabilities subtotal
Neither guard can currently fire. Both filings leave one of the three
balance-sheet totals to be derived, and the identity check correctly
refuses to run in that case, because verifying
assets = liabilities + equityagainst a total computed asassets - equityis circular. Boththerefore report
not_applicableand pass unconditionally.Their present value is limited to proving extraction does not crash.
What would actually close it
Pinned reference facts in
src/openfilings/benchmarks.py, which is themechanism built for exactly this: exact figures that fail if normalized
output drifts.
That file states its own rule clearly, and it must be honoured:
So this needs the figures transcribed from Volvo's and Keppel's published
annual reports by hand, not copied from what the library currently returns.
Related
benchmarks.pycurrently covers two issuers in total (Tesco and ASML) - seethe companion issue on broadening it.