> **This is our internal working methodology, published unedited.** > > It is not a tidied-up public summary. It still contains the corrections and > retractions we made along the way — including two findings we published > internally and then withdrew, and the reason each was wrong. That is > deliberate: a methodology that shows only its final state is asking to be > trusted rather than checked. > > Two things to know while reading it. Numbers prefixed with `#` are issues in > our private tracker and will mean nothing to you; ignore them. Paths like > `apps/marketing/...` are files in our repository, named so we can find them > again — the only one that matters to you is the source artifact this dataset > is derived from, and every published number here traces to it. > > Dataset: https://efficientdollar.com/data/leftover-money-after-essentials/ > Article: https://efficientdollar.com/blog/leftover-money-after-essentials/ > Licence: CC BY 4.0 --- # Leftover money after essentials — methodology **Artifact:** `apps/marketing/data/processed/leftover-after-essentials.json` **Generator:** `apps/marketing/scripts/build-leftover-after-essentials.py` **Raw:** `apps/marketing/data/raw/ce_labstat/`, `apps/marketing/data/raw/cpi_labstat/` **Issues:** #961 (this dataset) under #905 (the research post). Post: #962. Distribution: #963. The question: **how much money is left after the essentials are paid, and has that grown or shrunk?** Nationally, over the longest span the data honestly supports. --- ## 1. Why this source and no other Four candidates were researched (#905, 2026-08-14). Three were rejected on structure, not convenience: | Source | Verdict | |---|---| | BEA national accounts (PCE + disposable personal income) | **Rejected.** Imputations make the residual meaningless — ~15.8% of PCE is imputed, including $2.5T of owner-occupied rent BLS books as household consumption. Mean DPI per household 2024 was $162,594 against a Census median of $83,730. Naive use yields "the typical household overspends by $45k/yr", which is false. | | Fed Distributional Financial Accounts | **Rejected.** Wealth only. No income flow, no consumption, no saving. | | Fed Survey of Consumer Finances | **Rejected for the subtraction.** A balance-sheet survey; it never collected an expenditure total. Spending-vs-income is categorical, with no dollar amount. | | **BLS Consumer Expenditure Survey** | **Chosen.** The only source measuring after-tax income *and* category spending **on the same households** — the only way to compute a residual without mixing distributions. | The decisive property is the last one. Subtracting a mean basket from a median income, or one survey's spending from another's income, produces a number about no household that exists. ### Aggregates, not microdata We use the **published quintile means** from the LABSTAT bulk time series, not the CE public-use microdata. The repo holds both. PUMD (`data/raw/ce_pumd/`) is the right source for *within-band distributions* — that is what `process-ce-leftover.py` builds. It is the wrong source for a long mean series: it is per-year, behind a bot wall, and would have to be downloaded and harmonised eleven times to answer a question BLS already publishes. --- ## 2. Construction ``` leftover(quintile, year) = income after taxes − essential spending ``` Both terms are CE published annual means for the same quintile of consumer units in the same year. **Essentials (broad) — following the St. Louis Fed:** food at home, housing, transportation, healthcare, personal insurance & pensions. The residual swings by more than 2× on definition, so whoever chooses the categories chooses the answer. Following a Federal Reserve bank's published treatment means the normative call is owned by an institution with no stake in our conclusion, and it makes our figures directly comparable to theirs. > **Precision about what is and is not verified (#962, 2026-08-16).** This section > previously said the five categories were "adopted **verbatim**" from the primer. > The verifiable claims are narrower, and the wording now matches them: > > - **VERIFIED.** The primer's plain-English framing of essentials, quoted from the > article's *"What Is Discretionary Income?"* section: *"rent or mortgage, groceries, > utilities like electricity and water, insurance, car payments and taxes."* Note this > names **taxes** (we handle those by starting from after-tax income, not by > subtracting a spending line) and does not name **healthcare** in those words. > - **VERIFIED, and it is the load-bearing evidence.** These five CE lines reproduce all > three of the primer's published 2023 figures within the rounding it printed (§5). An > independent derivation from raw LABSTAT files does not land on three fabricated > numbers by accident, so the published figures are real and the basket that > reproduces them is the right one. > - **NOT VERIFIED.** Whether the primer itself names these five CE line items anywhere > below the definitional section. `stlouisfed.org` returns HTTP 403 to every automated > fetch tried from this repo's sandbox across two sessions, and `web.archive.org` is > unreachable, so only the section a human transcribed by hand is confirmed. > > **Write "following the St. Louis Fed", never "adopted verbatim", until someone reads > the whole primer.** The distinction matters because "verbatim" is the sentence that > transfers the normative judgment to the Fed, and it is exactly the sentence a hostile > expert reader would check first. **Essentials (narrow) — the same list minus personal insurance & pensions.** That category is 12.5% of spending and is mostly Social Security contributions and retirement deposits, which are arguably *saving* rather than consumption. It is the single largest definitional lever in the whole construction. Both variants ship; neither is privileged. ### CE `HOUSING` is broader than a housing payment It aggregates shelter, utilities, household operations (which includes childcare), housekeeping supplies and household furnishings. Any comparison against a narrower "rent plus bills" basket must say so — see §6. --- ## 3. Windows, and why the obvious chart is forbidden | Series | Span | Form | |---|---|---| | Dollar residual | **2013–2023** | real (constant 2024$) and nominal | | Essentials share | **1984–2024** | ratio | **There is no 40-year dollar chart, and there must never be one.** CE's coverage of national consumption fell from ≈0.795 of PCE in 1984 to ≈0.536 in 2024 (external; Bee/Meyer/Sullivan, NBER w18308). Under-captured spending mechanically **manufactures** rising leftover money. Measured from this pull, the third quintile's naive 1984→2024 residual in constant 2024 dollars reads **+65.4%** (`INCBFTAX − broad essentials`, $12,686 → $20,980). No after-tax version is even constructible, because after-tax income stops at 2023. Most of that +65.4% is plausibly artifact rather than gain, and no honest way exists to say how much. This is the credibility bomb in the dataset and the reason the dollar window is short. > **A #905 figure retired here.** The research checkpoint reported this as *"+86.4% > naive, +59.9% chain-linked."* Neither reproduces from these sources — every > construction tried lands elsewhere (`TOTALEXP − broad` gives **−16.7%**, which is > the opposite sign). The figures are dropped rather than carried, and the > reproducible +65.4% replaces them. The argument does not depend on the magnitude: > a coverage ratio that fell by a third is disqualifying at any residual. The ratio form cancels part of that drift, which is why the 41-year series is published as a share and never as dollars. ### Why the dollar window starts in 2013 Not data availability — 2012 and earlier are fully populated. It is a methodology regime boundary: 1. **Pre-2004**: income statistics came from *complete income reporters* only while expenditure statistics came from all consumer units. Different households in the numerator and the denominator. Structurally verifiable: the complete/incomplete reporter breakdowns exist in the LABSTAT catalog only through 2003. 2. **2013**: taxes became TAXSIM-modeled rather than reported. Real personal taxes step $3,041 → $10,008 across 2012→2013 — a 3.3× artefact, not a tax rise. 3. **2014**: CE health insurance carries its own question-redesign break. The published health-insurance line (all consumer units) moves $2,229 → $2,868: **+28.7% nominal**, **+26.6%** in constant 2024 dollars. (#905 reported +26.2%, which reproduces from neither and carried no citation.) ### The window is permanently closed After-tax income is **discontinued after 2023** — NBER stopped updating TAXSIM, so BLS publishes no CE after-tax income for 2024 onward. Expenditures continue. This construction can therefore never become a recurring series, which settles the one-off-post decision rather than merely supporting it. Gate G6 asserts the absence so that if BLS ever resumes the series a human widens the window deliberately instead of the build quietly picking up a new regime. --- ## 4. Deflation BLS publishes CE as **nominal** and has no house method for deflating it. Its own reports are inconsistent about this — recent ones subtract inflation *rates* from spending growth rates rather than rebasing, while conceding "there is a range of possible real-spending estimates that are not presented here." So the deflator is **our** choice and must be labelled as ours wherever a real figure is published. We use CPI-U annual averages (`CUUR0000SA0`, period `M13`), base year **2024**. The #905 research used R-CPI-U-RS, which is what Census uses for long-run real income. That reasoning does not bind a 2013–2023 window: across that span the two series are built from the same current methods and do not meaningfully diverge. R-CPI-U-RS would matter for a series reaching before 2000; ours does not. Choosing a better index does not touch the coverage drift in §3 — that is a nominal-side problem and survives any deflator. --- ## 5. Validation gates Nine gates run inline in the generator and **abort the build** rather than emit a suspect artifact. Each was verified by weakening the thing it guards and confirming it reds — a green gate is not evidence. | Gate | Guards | |---|---| | **G1** | The St. Louis Fed oracle — three published 2023 figures reproduced exactly | | **G2** | Published-total oracle — 2023 all-CU average annual expenditures = $77,280 | | **G3** | Sum-of-parts — five categories sum to broad; narrow = broad − insurance/pensions | | **G4** | Quintile ordering — leftover strictly increases Q1→Q5 in every year | | **G5** | Deflator **value** — the loaded CPI-U matches pinned published index levels, and the resulting factor has the right direction *and* magnitude | | **G6** | Closed-window guard — after-tax income must stay absent after 2023 | | **G7** | Completeness — no gaps in any series across its declared span | | **G8** | Estimate reliability — no series read here is footnoted RSE ≥ 25% | | **G9** | Regime-break confirmation — BLS's own footnotes span exactly the years §3 documents | ### The oracle St. Louis Fed, *"That Extra Money: A Primer on Discretionary Income"* (August 2025), 2023, nominal, broad basket: | Figure | Published | Ours | |---|---|---| | Mean after-tax income, all consumer units | ~$88,000 | **$87,869** | | Leftover, all consumer units | ~$27,000 | **$27,491** | | Leftover, lowest quintile | ~−$11,000 | **−$10,648** | Three of three, to the rounding the Fed printed. This is both the strongest available validation and the post's credibility anchor: our arithmetic is the same arithmetic a Reserve Bank published, extended across time. **How much that actually discriminates, measured.** Enumerating all 32,767 subsets of the 15 top-level CE lines, **200** reproduce both published figures at ±$1,000 and **42** at ±$500. So 99.4% of possible baskets fail — the match is not a coincidence of a rounded number. But it is not unique either: every passer with seven or fewer components contains these exact five as its core, and the alternatives are these five *plus one small line* (+personal care, +alcohol, +tobacco, +reading). **Ours is the only five-component basket that passes at any tolerance**, and it is the closest passer. The claim the evidence supports is therefore: the core is robustly identified, the edge is not, and the tie-break is that the Fed named five categories and this is the parsimonious set. Anyone restating this as "only our basket reproduces it" overstates it. **G2 is an independent second anchor.** 2023 all-CU average annual expenditures = $77,280, which is exactly the constant `process-ce-leftover.py` already hard-codes from the CE news release. Two unrelated pipelines, one from microdata and one from published aggregates, agreeing to the dollar — which is precisely the check that would catch a vintage or weighting error in either. ### A correction to the #905 research The #905 checkpoint recorded the third validation point as *"Q3 $27,491"*. **$27,491 is all consumer units, not the third quintile** — the third quintile's 2023 leftover is $14,572. The number was right and the label was wrong. More seriously, **the #905 quintile trend table does not reproduce from these sources**, and its stated methodology does not produce its own table (its figures sit nearer a four-category basket than the five it says it used, and its levels run 5–20% high throughout — most likely a microdata run). The five-category published-aggregate path reproduces a Reserve Bank's printed figures three times out of three, so it is the one that stands. **Every direction survives; only magnitudes move.** Notably the headline gets *stronger*: the bottom quintile's deterioration is −$1,583, not −$956. --- ## 6. Reconciliation against `leftover-money-benchmark.json` Two artifacts in this repo compute "leftover money" from CE. They are different constructions of the same concept and **must not silently disagree** — this is the #199 failure ("Take-Home Pay meant two different things across two calculators") and it must not repeat. | | `leftover-after-essentials` (this) | `leftover-money-benchmark` | |---|---|---| | Source | CE published aggregates (LABSTAT) | CE PUMD microdata (2022+2023) | | Statistic | **mean** | **median** + percentile curve | | Grouping | income **quintiles** | income **dollar bands** | | Period | annual, 1984–2024 | one pooled period, monthly | | Basket | 5 CE categories | 6 fields (housing payment, home bills, groceries, transportation, healthcare, childcare) | | Pensions/SS | inside the basket | removed from **income** | The pension treatment is equivalent — both remove it from leftover. The baskets differ: CE `HOUSING` already contains childcare (inside household operations) and additionally contains furnishings and housekeeping supplies, so **this artifact's basket is slightly broader**, and its leftover correspondingly lower. **Where they can be compared, they agree:** | Check | This | Benchmark | |---|---|---| | Bottom-group take-home, monthly | $1,348 (Q1 mean) | $1,307 (`<25k` median) | | Bottom-group leftover sign | negative (−$887/mo) | negative (−$407/mo median, 66% of households negative) | | Leftover rises with income | yes, strictly | yes, strictly | The remaining gap is mean-vs-median plus the broader basket, both in the expected direction. `apps/marketing/src/__tests__/leftover-after-essentials-data.test.ts` asserts the agreements above so the two cannot drift apart unnoticed. --- ## 7. Blind spots that must be disclosed wherever these numbers are published These are not caveats to bury in a methodology note. Two of them have already destroyed a published index. 1. **Employer-paid health insurance appears in neither income nor spending.** The employer share of family coverage was **$20,143 in 2025** (KFF) — about **3.25× the entire CE healthcare line**. The employer's Social Security match has the same blind spot. This exact unit mismatch accounted for more than half of the Cost-of-Thriving Index's headline finding and is what discredited it. 2. **CE excludes mortgage principal** (treated as asset accumulation), so "total expenditures" is not money out the door. 3. **The reference household changed.** Age of reference person 46.7 → 52.2 across the span (reproduces from this pull). Consumer-unit size 2.6 → 2.4 is carried from BLS summary tables — LABSTAT publishes earners, children and vehicles per CU but no persons-per-CU series, so that half is **not** verifiable from these files. Older, smaller households need less; part of any rising residual is demography, not affordability. 4. **The decomposition trap.** Middle-quintile non-essential spending falls from **27.4% to 20.0% of budget** (1984→2024; `(TOTALEXP − broad essentials) / TOTALEXP`, third quintile). That looks like a squeeze until it is decomposed. Of the total real decline across the non-essential lines that actually fell, in constant 2024 dollars: | Line | Real change | Share of the decline | |---|---|---| | Apparel | −$2,192 | **61.3%** | | Food away from home | −$412 | 11.5% | | Alcohol | −$314 | 8.8% | | Tobacco | −$297 | 8.3% | | Reading | −$280 | 7.8% | | Miscellaneous | −$83 | 2.3% | **Apparel alone is nearly two-thirds of it** — clothing got dramatically cheaper, which is a price story, not an affordability story. Anyone who does not decompose publishes the wrong one. > **Two more #905 figures retired here.** The checkpoint reported *"17.7% → 13.4%"* and > *"62% of the dollar decline is alcohol, tobacco and reading … strip those three and it > is −6.5% real."* Neither reproduces: the share is 27.4% → 20.0%, and those three are > **24.9%** of the decline, not 62%. The warning survives and is if anything sharper — > but its subject is apparel, not smoking and print. This section's whole point is that > undecomposed numbers mislead, so it could not itself ship an unreproducible one. 5. **"Essential" is a spending CATEGORY, not a necessity** (found 2026-08-16 by the founder while reviewing #962's chart 3; not previously recorded anywhere). Every dollar spent inside the five categories counts as essential, however discretionary it actually was. In 2023 the highest quintile spent **2.5×** the lowest on food at home ($9,198 vs $3,707), **3.1×** on housing ($43,897 vs $13,943) and **5.1×** on transportation ($25,279 vs $4,917). Nobody eats 2.5× the calories — that is quality and quantity choice landing inside an "essential" line, and the same is true of a larger mortgage and a nicer car. **Consequence: a high essentials SHARE is not by itself evidence of financial pressure.** For the lowest quintile, spending 165.8% of take-home on essentials alone, it plainly is. For the highest it partly reflects buying expensive versions of necessary things. Any share-based claim must not be phrased as though the two were the same condition, and any chart of essentials-as-a-share inherits this — including ours. 6. **Top-quintile leftover is noisy.** In constant dollars it runs $80,108 (2015) → $97,079 (2016) → $80,998 (2017). No chart may narrate a one-year move as a trend. --- ## 8. What the data actually says Leftover money, constant 2024 dollars, broad basket: | Quintile | 2013 | 2023 | Change | |---|---|---|---| | Lowest | −$9,379 | −$10,962 | **−$1,583** | | Second | $1,602 | $2,084 | +$481 (+30.0%) | | Third | $13,783 | $15,002 | +$1,218 (+8.8%) | | Fourth | $29,476 | $35,918 | +$6,442 (+21.9%) | | Highest | $79,483 | $99,275 | **+$19,792** (+24.9%) | | All | $22,978 | $28,302 | +$5,324 (+23.2%) | Essentials as a share of after-tax income, 2023: lowest quintile **165.8%**, second 95.0%, third 78.1%, fourth 66.6%, highest 54.3%. Essentials as a share of total spending, the 41-year series: **72.3% (1984) → 78.7% (2024)**, rising in every quintile. > **⚠️ WHO the share-squeeze fell on REVERSES on the basket, and only the dollar findings are > robust (#962, 2026-08-16).** This was found by the founder asking why the top quintile's > discretionary share looked implausibly small. It is the single most important caveat in this > document, because the broad-basket answer is the flattering, quotable one and it is not > load-bearing. > > | | broad (counts insurance & pensions) | narrow (excludes it) | > |---|---|---| > | spread across the five, 1984 | 3.7pp | 11.6pp | > | spread, 2024 | **3.1pp — CONVERGING** | **18.0pp — DIVERGING** | > | rose most | Q4 **+7.9pp**, Q3 +7.3pp | **Q1 +7.0pp** | > | Q5 | +6.3pp | **+0.5pp — flat across 41 years** | > > Personal insurance & pensions is mostly Social Security contributions and retirement deposits. > Counting saving as an expense pushes high earners' essentials share up toward everyone else's, > which manufactures the convergence. **§2 says both baskets are published and neither is > privileged — so neither answer may be published alone.** The defensible claim is the narrower > one: essentials take a growing share of what every quintile spends, and whether that squeeze fell > evenly across the distribution is not something this data settles. > > **The DOLLAR findings survive the same swap and are therefore what the post leads on.** Leftover > money still rises for four of five quintiles and still falls for the lowest (−$123/mo narrow vs > −$132 broad), and the raise-absorption gradient stays monotonic: **148.2 / 83.7 / 73.4 / 42.3 / > 28.2%** narrow against **151.8 / 91.7 / 87.7 / 62.0 / 46.2%** broad. > **A claim retired here — it was wrong, and three files carried it (#962, 2026-08-16).** > Earlier text in this section, the #961 hand-off comment on #962, and the 2026-08-15 > `idea-log.md` entry all said essentials take a larger share **"than at any point in four > decades" / "a 41-year high."** They do not. **2020 is the peak in every single quintile:** > > | | all | Q1 | Q2 | Q3 | Q4 | Q5 | > |---|---|---|---|---|---|---| > | **2020 peak** | 79.2% | 80.6% | 80.4% | 80.5% | 79.7% | 77.6% | > | 2024 | 78.7% | 78.9% | 80.1% | 80.0% | 79.7% | 77.0% | > > 2020 is a pandemic artifact — discretionary spending collapsed, so the essentials *share* > spiked. The accurate claim is **"the highest in 41 years apart from 2020"** (2021 77.9%, > 2022 78.0%, 2023 78.1%, 2024 78.7%). It is also the better line to publish, because it > answers the obvious objection before a reader raises it. Nobody had checked the maximum; > the endpoint was assumed to be it. **The two series point in apparently opposite directions, and that tension is the finding.** Essentials claim a larger share of the budget than in any year except 2020, while the dollar residual grew — because incomes outran spending. **Where the growth went is a two-part answer, and reporting only one part is a defect.** In *dollars* the gains are extremely lopsided: per month, Q1 **−$132**, Q2 **+$40**, Q3 **+$102**, Q4 **+$537**, Q5 **+$1,649**. In *percent* they are not, and are mildly progressive: Q2 **+30.0%** grew faster than Q5's **+24.9%** (Q3 +8.8%, Q4 +21.9%, all +23.2%). Both are true; dollars is the honest lead, because a percentage of a small base is not spendable. Any published version must not assert the dollar story as if the percentage story agreed with it. **The mechanism, which is monotonic and is the strongest finding in the dataset.** The share of each quintile's real 2013→2023 after-tax income gain that essential spending absorbed: | Quintile | Real income gain | Essentials gain | Share of raise absorbed | |---|---:|---:|---:| | Lowest | +$3,059 | +$4,642 | **151.8%** | | Second | +$5,780 | +$5,299 | 91.7% | | Third | +$9,872 | +$8,653 | 87.7% | | Fourth | +$16,961 | +$10,519 | 62.0% | | Highest | +$36,769 | +$16,977 | **46.2%** | | All | +$14,580 | +$9,255 | 63.5% | The lower the quintile, the more of its raise the essentials took. This explains the dollar lopsidedness mechanically rather than asserting it, and it explains why a decade of genuine real income growth reads as "nothing happened" to most households. **Endpoint-robustness of the gradient (tested #962, 2026-08-16).** Re-run over every window whose endpoints are ≥8 years apart that the dollar series allows — 2013–2021, 2013–2022, 2013–2023, 2014–2022, 2014–2023, 2015–2023 — on **both** baskets. Strictly decreasing from Q1 to Q5 in **12/12**. Levels swing hard (Q1 absorption ranges 106.2%–208.9% across the twelve tests; Q5 18.3%–46.2%), the **ordering never breaks**. This is what earns it the spine. **The gradient is largely a restatement of starting position, and the post says so.** Across the five quintiles, absorption correlates with each quintile's 2013 essentials share of after-tax income at **r = 0.976**. Absorbed ÷ starting ratio: 0.90 / 0.96 / 1.15 / 0.92 / 0.83. That is the mechanism, not a defect — the defensible claim is structural ("the further behind you start, the less of any raise you keep"), not decade-specific. > **⚠️ RETIRED before publication — do not resurrect.** The #962 checkpoints of 2026-08-16 > proposed promoting **"the middle quintile is the only one above 1.0"** (cost growth ÷ income > growth; equivalently the only quintile whose essentials share of after-tax income *rose*, > 76.5% → 78.1% broad, 68.5% → 69.2% narrow) to the post's centerpiece, in place of the > gradient. **It does not survive an endpoint test and is not in the post as a finding:** > > | Window (broad) | Quintiles with ratio > 1.0 | > |---|---| > | 2013–2023 | **Q3 only** ← the proposed finding | > | 2014–2023 | **Q2 only** | > | 2015–2023 | Q1, Q2, Q3 | > | 2013–2022 | Q3 only | > | 2014–2022 | **none** | > | 2013–2021 | **none** | > > Move the start year by one and the "only quintile" changes identity. Q3's broad share of > after-tax income across the decade — 76.5 / 78.7 / 77.0 / 77.5 / 79.1 / 78.4 / 78.4 / 71.6 / > 73.5 / 77.6 / 78.1 — shows 2013 sitting near the bottom of the non-pandemic range, which is > the whole of the "+1.6pp". The post carries it in its Methodology only, as a worked example > of how easily a finding is manufactured. **The bottom quintile needs all three of its facts stated together.** Its dollar gap widened (−$9,379 → −$10,962), its essentials share of income *eased* (169.0% → 165.8%, peak 182.9% in 2019), and its real after-tax income grew **+22.5% — the fastest of any quintile** (Q5 +20.4%, all +19.2%). "It went further underwater" alone is half the truth. The full truth is that when essentials start 69% above income, even faster-than-average income growth widens the absolute hole. The honest headline is not a collapse. It is a widening. --- ## 9. Prior art to name - **St. Louis Fed** published the one-year cut using our exact definition. We extend it across time; we did not invent it. - **The Hamilton Project** charted CE spending shares 1984–2014 and stopped one subtraction short. - **LISEP** (Ludwig Institute for Shared Economic Prosperity) publishes the closest thing to this that exists: ***Leftover Earnings After Basic Living Expenses***, nominal annual dollars, 2001–2024, for eight household types. *Leftover Income* is the local (50-MSA, 2005–2022) variant of the same residual expressed as a share of earnings. The differences from ours are: **(1) pre-tax** — no taxes subtracted, no transfers added ("In LISEP's main number, taxes are left out", TLC Methodology Feb 2026); **(2) assumed, not measured** — every adult is presumed a full-time worker earning the median wage, priced against a *normative* basket of minimal adequate needs (housing, medical care, transportation, food, childcare, technology, miscellaneous), so the household exists in no survey; **(3) cut by family structure** where ours is cut by income quintile. LISEP measures a **standard**; we measure a **population**. Novelty is claimable on the quintile-by-quintile after-tax cut of real households — **not** on "dollars rather than a ratio." > **A claim retired here — verified false (#962, 2026-08-16).** This bullet previously > read: *"Theirs is pre-tax earnings **deflated by** a cost index — a purchasing-power > ratio, not a dollar residual … (deflating it flips the sign, +64.9% since 2001 > becoming −6.9%)."* **LISEP's leftover metric is a dollar residual, the same > construction as ours.** From their own published workbook (`TLC Data 2024.xlsx`, > sheet "Earnings Left After Expenses"): single person $12,195.23 (2001) → $20,111.65 > (2024); single parent + 3 children −$12,387.57 → **−$30,218.00**. Negative values in > dollar magnitudes are dispositive — no deflated index behaves that way. The sentence > was also self-refuting: the "+64.9% → −6.9%" parenthetical was computed *from that > very dollar series*. The ratio description belongs to a **different** LISEP metric, > *Change in Spending Power* (median earnings ÷ their TLC/MQL cost index). The > `#962` done-means required confirming this before claiming novelty; it did not hold. **Never cite a LISEP metro figure.** Their own numbers disagree across three of their own surfaces for the same city-year — San Francisco is 18.2% (2023 press release), 29.5% ("Rankings 2022" sheet) and 0.8% ("Fact Sheet 2022 Data" sheet); San Jose 25.4% / 33.0% / −5.0%. The two workbook sheets disagree for **all 50 MSAs**, zero exact matches, and no published local methodology exists to reconcile them. National series only. **The asset here is defensibility, not originality.** Well-resourced institutions have stood next to this question and not built it.