Today's red-team run seeded 16 small faults into two files and ran each file's own tests after every one: the engine of the S3 storage-class checker and the schedule code of Rev Rec. 11 broke a test. Of the 5 that got through, 3 could not change any output, and the other 2 now have tests that catch them. Reading the CSV import the way a stranger might use it also turned up one real bug, now fixed.
The bug: a typo read as "no end date"
Rev Rec's browser demo reads an invoice-line CSV: amount, billing date, period start and an optional period end. A line with no end is a one-off charge, recognised on its start date. A line with an end is spread evenly per day across the period.
The import treated "the end didn't parse" the same as "the end is blank". So a $1,200 annual subscription with its end typed as 2027-13-01 (month 13) became a one-off. The schedule put all $1,200 into January, showed nothing deferred, and listed nothing in the skipped rows. A bad start date already gets the row skipped with a reason. A bad end date gave a confident wrong schedule instead.
The fix keeps the blank-end case as it was and skips a row whose end is filled in but isn't a date, with the reason "period end is not a date", which the page shows next to the line count. A test pins both cases.
What the tests missed
- S3 checker, custom retrieval prices. The engine merges your price overrides over the defaults. If the retrieval-price override had been silently dropped, the tests would have stayed green, because none of them passed a custom retrieval price. The new test sets Glacier retrieval to five times the default and checks the 12-month retrieval cost comes out five times larger, with the other classes left alone.
- Rev Rec, credits on a positive line. A credit note that names the invoice line it refunds is spread over that line's remaining term, so a mid-year refund stops the schedule instead of booking one day of negative revenue. That should happen only for negative lines. Nothing checked that a positive line with a
creditsvalue keeps its own date. Now a test does.
The three that changed nothing: a CSV line-ending check that only differs on a blank line (which is dropped either way), a < turned into <= on two equal month strings, and a > turned into >= one step after the zero case had already returned.
The numbers
| Seeded | Caught by the tests | Equivalent | Catch rate | |
|---|---|---|---|---|
| Before | 16 | 11 | 3 | 11 of 13 (85%) |
| After the new tests | 16 | 13 | 3 | 13 of 13 (100%) |
Rev Rec and the S3 checker are free and MIT licensed. Their output is an estimate from the data and assumptions you give them, not accounting, tax or financial advice.