fix(reports): make every tab agree on what spending means

[KenTaniguchi-R]

Closes #142.

Clicking through the Reports tabs gave four different answers to "how much did I spend?", all under one Last 3 months chip:

Tab	Before	After
Spending	$25,495.49	$25,495.49
Income vs Expense	$28,494.34	$25,495.49
Trends	$9,905.63	$25,495.49

Measured against the dev database, range Jun 2 – Sep 2.

What was wrong

spendingBaseConditions (src/lib/spending-helpers.ts:26) is the canonical rule — a negative, settled, non-transfer, non-income transaction, summed by magnitude. Four queries diverged from it.

getIncomeVsExpense summed ABS() over all non-income rows, so every refund and credit counted as spending. The discrepancy was exactly the period's credits: $28,494.34 − $25,495.49 = $2,998.85 = $2,600.00 of rent that only received credits + $398.85 of refunds.

getIncomeExpenseByCategory did three things at once:

• isNotNull(transactions.categoryId) dropped uncategorized, so 61% of spending was missing from the table explaining the tile directly above it.
• Summing signed amounts made a net-positive category (Rent/Mortgage) render as a positive "expense" among negatives.
• Sorting those signed totals ranked expenses backwards — the largest expense was the most negative, so it sorted last and Home Insurance ($5.06) sat at the top.
• percentOfTotal divided by the signed net of category totals (−$6,906.78), a number shown nowhere, producing shares like -38% that summed past 100.

getCategoryTrends skipped null-category rows, so its "Total Spent" tile was silently total-minus-uncategorized — $25,495.49 − $15,589.86 = $9,905.63, exactly.

getCashFlowSankey excluded uncategorized twice: an explicit isNotNull, and a notInArray that evaluates to NULL rather than TRUE for a null category_id. The second one is why the first fix silently didn't work — the new test caught it.

What changed

All four share one definition. Expense rows sum to their tile, expenses sort largest-first as positive magnitudes, percentages are positive and sum to 100 within each pool, and uncategorized is a first-class row everywhere.

IncomeExpenseCategoryRow.categoryId and CategoryTrendRow.categoryId are now string | null; the category table and its drill-down handler carry the null through, which DrillDownSheet already reads as uncategorized.

Deliberately not in scope

Income aggregation is unchanged. Making it symmetric drops Salary and Investment Income entirely — they net −$7,563.92 signed in this dataset, while ABS() overshoots the Total Income tile. Gross vs net income is a product decision, so this PR leaves it alone rather than settling it silently. The income rows still undershoot the tile, which counts uncategorized credits as income; worth its own issue.

Testing

tests/integration/report-consistency.test.ts — 11 new tests that pin the definition of spending itself rather than any one tab's arithmetic: cross-tab totals, refunds excluded, credit-only categories excluded, uncategorized present in every surface, rows summing to their tile, sort order, and percentage sign.

Three tests in report-queries.test.ts asserted the old behavior (negative totals, uncategorized excluded) and were updated with the reasoning recorded.

Typecheck and lint clean. Full suite 1042/1043 — the one failure, investment-queries.test.ts > returns dayChange from holdings_history, is pre-existing hardcoded-date rot and fails identically on a clean main (verified by stashing).

🤖 Generated with Claude Code