-
A random test found two bugs in how ai-bookkeeper sells an asset
700 made-up assets and disposals showed ai-bookkeeper booking too little depreciation before a sale, and a cash-flow statement that did not tie.
-
ai-bookkeeper now reads Amazon's order-history report
ai-bookkeeper imported zero rows from Amazon Business Order History CSVs. Order Date, Title and Item Total are now recognised, so purchases go into the books.
-
A bank file that reads two ways: ai-bookkeeper now says when it guessed the date order
Ten real spreadsheet statements from OCA's bank-statement-import all read, but one file's dates (01/04, 02/04) were silently booked to January and February.
-
Twelve ways a bank writes -4.50, and the one we checked wrong
ai-bookkeeper read twelve spellings of one amount (Unicode minus, Swiss apostrophes, DR/CR, euro decimals) to the same cents. Our one fault was arithmetic.
-
What a demo bank feed taught the bookkeeper's keyword rules
Of six open-source accounting projects, only Bigcapital shipped a bank-feed sample; 121 rows read cleanly and showed what the built-in keyword rules miss.
-
How much do you have to sell a month? Break-even straight from your bank statements
AI Bookkeeper now computes monthly break-even and margin of safety from a profit and loss built from your bank statements.
-
Testing break-even with 800 made-up businesses
A property test throws 800 random profit and loss statements at ai-bookkeeper break-even and days-outstanding figures. No fault found.
-
A credit note in a different currency than its invoice
ai-bookkeeper revalues a credit note against an invoice in the same foreign currency. For different currencies it works at base value, and a new test pins that.
-
AI Bookkeeper now shows who your money comes from
A new payees report ranks customers and suppliers by share of the period's money and warns when one payer is a quarter or more of your income.
-
Why an eBay payout is not your sales: a worked month
A synthetic eBay month where $145.00 of sales and $5.00 of collected tax became a $111.70 payout, and how AI Bookkeeper splits it.
-
Crypto exports date rows in Unix milliseconds, and ai-bookkeeper skipped every one
Sweeping four open-source repos' sample files through ai-bookkeeper's bank reader found one fault: epoch timestamps were rejected as dates. Fixed, with a test.
-
ai-bookkeeper now matches a euro invoice to euro payments in euros
A EUR invoice paid in two EUR deposits used to leave the receivable open because deposits were matched on their dollar value. They now match in euros.
-
ai-bookkeeper: two more checks that a bank file's layout can change without changing the books
An unused column, or a signed amount split into debit and credit columns, must not change what ai-bookkeeper reads. Both hold on every fixture.
-
ai-bookkeeper read every Fidelity cash line as 'No Description'
Sweeping another open-source project's test exports found that Fidelity's cash-management CSV lost its payee text, and its buys were booked as spending.
-
800 random euro invoices later, the books still balance
A seeded property test throws random foreign invoices, bills, rates and instalments at ai-bookkeeper's settlement code. It found nothing, and now guards it.
-
Free cash flow on the bookkeeper's cash flow statement
ai-bookkeeper's cash flow statement now ends with free cash flow: cash from operations less what you spent on equipment.
-
A truncated column header booked a 110 rupee debit as a 30,523 credit
Sweeping an open-source finance app's Indian bank fixtures through ai-bookkeeper found three import faults, one of which changed the amount.
-
How long do your customers take to pay? A two-number answer in the books
ai-bookkeeper's report now shows days sales outstanding and days payable, computed from the open receivable and payable.
-
ai-bookkeeper: the IRS publications can be read again, and the 1099 figures hold
Publication 1099 (2026) confirms the 1099-K thresholds and the 24% backup withholding rate that ai-bookkeeper had coded from memory.
-
Two more bank-file faults found in KMyMoney's sample files: empty descriptions and a day-first QIF read as nothing
Sweeping KMyMoney's sample CSV and QIF through ai-bookkeeper found a header it did not know and a QIF whose dates it could not read.
-
ai-bookkeeper now reads its own books the way a lender does
A ratios report for ai-bookkeeper: margins, working capital, current and quick ratio and debt to equity, with n/a instead of a made-up number.
-
Why a lender's loan schedule can differ from yours by dollars
600 random loans through ai-bookkeeper's schedule: every invariant holds, and the last payment absorbs rounding drift that compounds on long, high-rate loans.
-
AI Bookkeeper shows cash flow month by month
The cash flow statement in AI Bookkeeper now has a monthly table and a runway figure based on the last three months, not the whole-period average.
-
An OFX file with no posting date lost its interest line
Kresus's checking.ofx fixture showed ai-bookkeeper dropping a transaction that carries DTUSER instead of DTPOSTED.
-
Reading only the payee loses what the categorizer needs
Two sample bank files from the Fava project read fine but dropped half the description. ai-bookkeeper now keeps both columns.
-
ai-bookkeeper now uses Plaid's own category as a hint
Plaid transaction JSON carries a personal_finance_category. ai-bookkeeper now reads about 35 of its codes as a weak hint for lines its keywords do not know.
-
A bank file format we silently read as empty: QFX with values on the next line
Sweeping another importer's test files through ai-bookkeeper found OFX/QFX statements with each value on the line after its tag; they imported zero rows.
-
AI Bookkeeper now reports profit and loss by project
Tag bank lines and invoices to a job, and ai-bookkeeper shows revenue, cost and margin for each one, with the overhead left visible.
-
Reading the company name off a receipt from its registration number
ai-bookkeeper's receipt reader now trusts the line above a printed registration number, lifting merchant accuracy on 626 real receipts from 83.2% to 84.7%.
-
One bracket cost a receipt reader 20 points on shop names
ai-bookkeeper misread the merchant on 22 AEON receipts: a registration number looked like symbols. Fixing it lifted merchant accuracy from 63.6% to 83.2%.
-
Reconciling 300 random statements: the bank rec hints hold up
A seeded property test runs ai-bookkeeper statement reconciliation on 300 random statements and checks the difference, the hints and untouched books.
-
Checking a sales tax return against 150 random ledgers
A seeded property test runs the sales tax return on 150 random ledgers at three frequencies and checks every state due-day rule.
-
Schwab's checking export has note lines under the header, and our reader warned about them
Sweeping 33 public bank-file fixtures through ai-bookkeeper's reader found one fault: Schwab's newer checking CSV was not recognised and gave false warnings.
-
ai-bookkeeper stops booking 10 shares as $10
A bank-style CSV with an amount like '10 AAPL' used to import as ten dollars. It is now reported instead of booked.
-
Checking sales tax due dates against the state's own pages
ai-bookkeeper's sales tax worksheet presets filing due dates for dozens of states; three are now read from the state's own page, and one was wrong.
-
A stray comma at the end of every line moved a bank statement's dates
A new check adds an empty last column to every CSV fixture. One header-less Australian export read April instead of March.
-
Closing the books on 200 random ledgers
A seeded property test runs ai-bookkeeper's reports and year-end close on 200 random ledgers and checks the books still balance.
-
An interval that admits where it is optimistic
The changepoint detector's growth-change range holds the true bend 94% of the time on clumpy counts, but 82% on steep bends. The CLI and JSON now say so.
-
A tax estimate for a year the IRS has not published yet
AI Bookkeeper's federal estimate now takes a year's brackets from a JSON file, so 2027 works the day the figures appear and refuses tables that cannot be right.
-
A multiplicative season did not fix the airline passengers series
changepoint-detector on the classic monthly airline passengers data: 21 false steps with an additive season, 22 with a multiplicative one, 8 after taking logs.
-
Heavy-tailed noise fooled the Bayesian changepoint view too: clipping fixed it
On t3 noise with no change, changepoint-detector's probability view flagged a change in 47% of 200-point series. Clipping at 2 robust sds cut that to 1%.
-
Fewer false alarms on volatile data: a bootstrap penalty for changepoint-detector
changepoint-detector's --bootstrap sets the meanvar penalty from block-resampled residuals, cutting false alarms on GARCH-style noise from 51% to 16%.
-
Two ways to keep the tail in a bootstrap penalty, and neither worked on the franc
changepoint-detector tried a robust scale for its bootstrap penalty on real daily FX returns: one changed nothing, the other erased the 2015 Swiss franc shock.
-
Changepoint detector: the count bootstrap stops working when you add a trend
A measurement of --bootstrap on clustered counts combined with --trend, a weekly season and negative-binomial noise.
-
Changepoint detector warns when drifting-season steps pile up
With a drifting seasonal shape, a cluster of steps usually means the shape is still changing, not that regimes did. The detector now says so.
-
How often does the changepoint detector's drift note cry wolf?
On 1,600 simulated seasonal series the drift note fires on 2-3.5% with no change, but on 11.5% of monthly series with two real steps.
-
A GARCH filter fixes false alarms in changepoint detection, then breaks real ones
Dividing a series by a fitted GARCH volatility cut false change alarms from about 50% to 1% but hid a real variance break 94 times in 100. A negative result.
-
Holidays do not break the count bootstrap, a growth trend does
changepoint-detector on simulated daily counts with known holidays: the block bootstrap cuts false alarms from 85% to 21% on clustered counts.
-
A season that grows with the level: changepoint-detector gets --multiplicative
A seasonal swing that grows with the level reads as steps when removed as a fixed amount. A multiplicative option helps monthly data and hurts quarterly.
-
A changepoint tool on a series where the answer depends on the scale
UK quarterly gas 1960-1986 through changepoint-detector: a slope bend near 1970 on the raw series, none on the log series, and a seasonal swing cut into steps.