Christopher Ward · Evidence ← MoneyLedger on the Product door

MoneyLedger — The Artifacts

A private, offline finance ledger in one HTML file — and the eight screens that show it exercising judgment rather than running validation.

Rev 1.6.0 · captured 1 August 2026 · synthetic dataset, 352 transactions across 19 months

The decision every artifact below is testing Notice it, price it, and then decline to assume. Software that finds a problem usually does one of two things — stays silent, or declares you wrong. This app does a third: it states what it sees, quantifies what the mistake costs, and then names the case where the user is right after all. That pattern is the product, and it repeats on every screen here — in the integrity check, in the split that reconciles to the cent, in a lock screen that admits there is no way back in.

01 The integrity check — the one screen where it disagrees with you

Decision A warning that only says something looks wrong pushes the work back onto the user. Name the amount, name the consequence in the units the user actually cares about, and concede the case where they meant it.
The app's data-check dialog reporting one issue: a credit card called Some Day VISA holding a positive balance of $500.00, warning that net worth is overstated by $1,000.00 until the opening balance is entered as owed, and noting that if the card was genuinely overpaid the figure is correct. Behind it: an automatic snapshot with a restore button, a never-exported backup warning, and a shred-all-snapshots control.
Data check — one issue, three moves in a single sentence. It notices a credit card holding a positive balance — you just wait a minute, mister. It prices the error — $500 with the wrong sign costs $1,000.00 of net worth, because the money is simultaneously not counted as debt and counted as an asset. Then it declines to assume: “If you have genuinely overpaid the card, it is correct.” Overpaying a card is a real thing people do, so the app says what it sees and stops short of calling it a mistake. The surface behind the dialog is the argument — an automatic snapshot taken without being asked, an honest “last workbook export: never”, and a control to shred every stored copy before you hand the device on.

This is also the rule the documentation had backwards. The shipped guide told users to enter what they owe as a positive number; the app was enforcing the opposite in software, and the guide was corrected to match it. The product was right and the manual was wrong — which is the correct direction for that error to run.

02 At a glance — and one decision visible without a caption

Decision Spending-by-category is a solved chart. The unsolved part is that money spent on credit is not money spent from an account — most tools quietly merge the two.
MoneyLedger dashboard: five KPI cards for net worth, debt, income, expenses and net for the month, a recent-activity list, a spending donut broken out by category with each slice's amount in the legend, a debts panel and a six-month income-versus-expenses chart.
Dashboard — July 2026. Net worth $76,094.45 shown with its own working underneath ($88,387.95 assets − $12,293.50 debt) rather than as a number you have to trust. The strongest thing in the frame is “On credit $558” carrying its own outlined ring in the donut — a spend that has not left an account yet, given a visually distinct treatment instead of being folded in with the rest. Every figure here was verified against the source file before this shot was promoted, not eyeballed.

03 Eleven themes, and whose pictures they are

Decision A tool you open every week should not feel like a tax form — but a theme system is only a craft argument if the imagery is yours. Scraped wallpaper in a portfolio about authorship is an own goal.
The Appearance theme picker showing eleven themes in a two-column grid — Ledger, Matrix, Terminal, Gulf, Riviera, Ocean, Wings, City, Route, Clouds and Landscape — each with a live preview rendering the same seed ledger and a written description of its color decisions.
Appearance — eleven themes, each previewed on the same seed ledger. The three non-photographic themes lead (Ledger, Matrix, Terminal), then the eight image-backed ones. The descriptions explain the color decisions rather than naming them — “Two hues taken from the same set and given opposite jobs”, “the frames carry no color, so the accent is chosen.” Ledger sits first for a structural reason as well as a visual one: it supplies the default palette and is the fallback the theme loader falls back to.
Why this screen took a week longer than the others Four of these themes were built on wallpaper of unknown origin. Christopher raised it himself rather than being told — “Some of the themes hold screen-scraped images. Should we include those anyway?” — and the fix was not to crop them out of frame. Three were rebuilt from generations he art-directed; one was deleted outright. Every frame in this grid is now either a photograph he took or an image he directed. Cropping the grid to hide the others would have traded a provenance problem for an honesty one.

04 One receipt, three budgets, reconciled to the cent

Decision A split editor that merely accepts three numbers proves nothing. One that runs the total and tells you whether it reconciles is doing the work.
The split-transaction editor: a Costco purchase of $187.43 divided across three categories — Groceries $112.80, Miscellaneous $49.63 and Personal Care $25.00 — with a running line reading Allocated $187.43 of $187.43, fully allocated.
Splitting a single receipt across three budgets. The line carrying the shot is “Allocated $187.43 of $187.43 · Fully allocated.” The same rule is enforced deeper down — the integrity check in §01 carries a split total ≠ amount condition — so the modal is surfacing a constraint the file format holds itself to, not a bit of interface politeness. A spreadsheet makes this painful: one row, one category, or a formula nobody maintains. Having every report, chart and budget line account for the three parts correctly is the domain depth the case study claims.

05 It names the entry that explains the gap

Decision Most reconciliation UIs stop at “you are out by $49.06.” Finding which transaction accounts for the difference is the part that is actually tedious — so do that part.
The reconcile screen for a credit card: a typed statement balance of -8,561.09, a cleared total, the remaining difference, and a hint identifying the single unticked transaction whose amount would close the gap.
Reconcile — the “closes the gap” hint. The hint has three forms and only one is worth showing: it names a single entry when exactly one unticked transaction matches, otherwise it says “N entries would each close this” or “no single entry accounts for this.” Reaching the good one meant deriving the target from the data — a cleared total of −$8,512.03 against a statement of −$8,561.09 leaves a gap of exactly $49.06, the effect of one unticked charge. Every intermediate figure in the frame matches that derivation.

06 A period is not an instant

Decision Income and expenses are flows measured over a span. Balances are a fact about one moment. Printing today's debt on a report headed “March 2025” would be quietly misleading — so refuse to.
A year-to-date report rendered as a centered document: income, expenses, net and savings rate for the period, followed by a Where you stand balance sheet listing assets, debts and net worth, stamped as of August 1.
Reports → Year-to-date, August 2026. Flow above — income $48,321.99, expenses $19,084.40, net +$29,237.59, a 61% savings rate. Balance sheet below, stamped “as of Aug 1.” The section is deliberately suppressed on any report whose period ends before the current month, which is why this had to be shot in August: a July report hides it entirely rather than printing a present-tense balance under a past-tense heading. That distinction is ordinary in accounting and rare in personal-finance software. The report renders as a ~720px sheet centered on the app background — it reads as paper on a desk, which the Print / Save as PDF button beside it makes literally true.

07 Density handled without clutter

Decision Twelve months by twenty-two categories is 264 cells. Showing everything and shouting about nothing is a typography and restraint problem, not a data problem.
The budget year grid in Actual mode for the 2025 financial year: twenty-two category rows against twelve month columns plus a total column, with populated categories showing amounts and unused categories showing a dim mid-dot instead of repeated zero values.
Budget → full financial year, actuals for 2025. Every total was verified against the source file before promotion, not eyeballed: expenses $32,989.71, income $79,846.57, Rent $19,800.00 across twelve untouched months. The nine empty categories are the point of the shot. Education, Travel, Health & Medical and the rest show a dim · rather than a wall of $0.00 — while the Total column still reads $0.00 where the zero is genuinely the answer. That restraint is the density claim.

08 What the product says about itself

Decision An app that holds the only copy of something should say so, unprompted — and an app with no password reset should say that before you need it, not after.
Unprompted, on any fresh restore
A banner across the top of the dashboard reading that the user has never exported a backup, and that a workbook is the only copy that survives losing access to this app.
The lock screen, stating the dead end
A full-page lock screen with a passphrase field and no surrounding application chrome, footed by a note explaining there is no password reset, that an encrypted json backup is locked with the same passphrase, and that the xlsx workbook remains readable and is the way back in.
Left: “You have never exported a backup. A workbook is the only copy that survives losing access to this app.” Most software that holds your only copy of something never mentions it. Right: “Forgot it? There is no reset.” — stated flatly, and then the way out in the same breath: the encrypted backup is locked with that same passphrase, but the spreadsheet export stays readable and is the route back in. Most software either implies a recovery path that does not exist or leaves you to find the dead end yourself. The full-page takeover is deliberate: when the app is locked there is no app, only the lock — nothing is dimmed-but-visible, because nothing is available.

09 What assembling this evidence found

Shooting these captures was not a photo session — it was a review pass, and the same thing happened that happened on the Homestead set. A full audit ran immediately before capture and the static half came back clean: every interaction handler accounted for, tokens complete across all eleven themes, 344 assertions green. The capture session then found defects anyway, because driving a product for an outside audience asks questions that normal use never asks.

Defects logged
4
on capture day, 1 Aug
Found by a person
3
one caught by a linter
Takes on the hero
4
to get one frame right
Shots skipped
1
claim carried better elsewhere
What was wrongHow it surfacedWhy it mattered
The month arrows had never been visible Christopher, being told how to change the month for a screenshot — “there are no arrows” The control existed and drew nothing, so the app had behaved for weeks as though it had one month
The budget grid could not be widened Trying to fit a fourteen-column grid in frame and finding the browser made no difference A width cap that was not relative to the window — the boundary looked like a layout choice and was a bug
A placeholder payee in the most-read panel Reading the dashboard capture before promoting it TestJson / TestJson sitting in Recent activity — exactly the leak the capture plan warns about
Ten stray commas left by a reordering A linter, in the pre-shoot gate — the one machine catch of the day The app would not open on ten of its eleven themes; pushed and live for four hours
Why this section is in a portfolio Three of the four were found by a person looking at output and refusing to accept it. The fourth was found by a linter — and it is in the table for that reason. The honest claim is not that automation catches nothing; it is that automation catches the class of fault it was written for, and a control that renders nothing, a cap that cannot be widened, and a placeholder in a screenshot are all outside it. Knowing which answer is wrong when nothing has failed is the judgment the job is for.

10 Reading these images honestly

What was staged, what was computed, and what a careful reader would want to know — stated here rather than left to be discovered.

In the imagesStatus
The dataset is synthetic Deliberate and non-negotiable. 352 transactions across 19 months, 4 accounts, 22 categories, ~30 invented payees. Christopher runs this app on his real finances daily; none of that appears here or ever will. The volume is realistic — the money is not real.
§01 and §04 are staged Both, and neither was saved into the record. The source file contains no mis-signed account and no split transaction, so each had to be created to screenshot the behavior. The split modal was closed without saving; the test credit card was removed afterwards by restoring the file.
§05 shows a typed statement balance, not a committed reconciliation Typing a balance and reading the hint changes nothing. Finish was deliberately not clicked — it writes cleared flags to every ticked entry and would have broken the dataset for every later shot.
§06 had to be shot in August, and §07 in the 2025 financial year Both are consequences of correct behavior rather than workarounds. The balance sheet suppresses itself on a past period; the budget grid defaults to a mode where nothing has been budgeted yet, and to a financial year that is only seven months populated.
§03 is not eleven photographs Three of the eight image-backed themes were rebuilt from generations Christopher art-directed, not photographs he took. The distinction matters on a page arguing authorship, so it is stated rather than blurred. The claim is that nothing in the grid is of unknown origin — not that every frame came out of a camera.
§04's middle category is Miscellaneous A small loss, recorded rather than hidden. An earlier take read Groceries / Shopping / Personal Care — one receipt landing in three real budgets — which made the point more sharply than a catch-all does. The claim survives; the sharper version does not.
Several frames carry a lot of empty ground Working as intended. The report is a centered document and the lock screen is a card on an empty field; both were captured at the set width because a single rig width across the set is the rule, and cropping is a layout decision made later.
← Back to MoneyLedger on the Product door
MoneyLedger — a private, offline personal-finance ledger that lives in a single HTML file: no cloud, no account, no network, no third-party JavaScript. Concept, product design, UX, branding and AI direction by Christopher Ward; built with Claude against a written operating contract. In daily use on his own finances. Captures 1 August 2026, Rev 1.6.0, against a synthetic dataset.