Official US statistics under zero-trust scrutiny

Web-verified 2026-06-08. Structured + edges + sources: macro-official-data-integrity.json. Evidence-graded. The claim is not that raw microdata is fabricated (unsupported). It is that methodology, revision-timing, collection degradation, and political pressure all bias the headline in one direction — and the incentives explain why.

1. The jobs headline overstated, then got quietly corrected — STRONG

• −911,000 jobs: the preliminary benchmark (Sep 2025) cut nonfarm job creation for the 12 months through March 2025 to ~1.44M.
• −818,000 preliminary for 2024 (finalized ~−600k).
• −258,000: the July 2025 report's own revision to May+June.
• The net birth-death model (which adds jobs from assumed new-business formation) accounts for roughly half the downward revisions — the over-count came from a model, not measured payrolls.
• Benchmarks land 6–18 months late, so the optimistic monthly prints drive Fed/market/political decisions before the correction arrives.

2. CPI methodology holds the print below lived cost — methodology FACT; "rigged" CONTESTED

• Owners' Equivalent Rent ≈ 25% of CPI, imputed from a rent survey, not transacted home prices — so the ~+50% 2020–24 home-price surge entered only as smoothed, lagged rent.
• Post-Boskin substitution (geometric mean), hedonic quality adjustment, outlet substitution — each can lower the rate vs a fixed basket.
• The Fed targets PCE, which runs ~30–40 bps below CPI — the lower gauge is the policy anchor.
• Substitution/quality effects are real economics; the dispute is whether the corrections overshot in the fiscally convenient direction. Cross-check: a US home is −81% priced in gold since 1998 (macro-gold-silver-reprice).

3. The motive is on the record — the Boskin Commission — STRONG

The 1996 Boskin Commission concluded CPI overstated cost-of-living by ~1.1 pp/yr, and framed it around fiscal cost: because Social Security, pensions, and tax brackets are CPI-indexed, overstatement is an automatic spending increase + tax cut. CBO-era estimate: correcting the "bias" cuts ~$691B off the debt by 2006. BLS then adopted substitution (1999), expanded hedonics, and OER. The substitution point was defensible — but "the index costs the Treasury money, so fix the index" is a documented government motive to lower the number.

4. The apparatus is degrading and being politicized — STRONG

• 2025 collection cuts: BLS suspended CPI collection entirely in Lincoln, Provo, Buffalo and cut ~15% of the sample across 72 areas (hiring freeze); imputation rose; the government shutdown disrupted collection further.
• Aug 1 2025: the President fired BLS Commissioner Erika McEntarfer after a weak jobs report, alleging manipulation — though the commissioner doesn't produce the estimates.
• Nominee E.J. Antoni (Project 2025) proposed suspending the monthly jobs report; nomination withdrawn Sep 30 2025.
• Honest caveat: BLS says the collection cuts moved 12-month all-items CPI by <0.01 pp — the damage is to reliability/independence and sub-indices, not (yet) a provable headline shift.

5. Measure-shopping (GDP vs GDI) — STRONG

GDP and GDI should match; the gap was ~$126B (0.4% of GDP) in Q1-2025 with GDI weaker (less momentum than the GDP headline). A lower deflator mechanically raises real GDP and shrinks debt/GDP — the same inflation-understatement incentive feeds the growth print.

Why — the incentive map

Every major fiscal incentive points one way: understate inflation, flatter output/labor.

The honest recompute under 1990s methodology (added 2026-06-16, #52)

Actually computing the headline under pre-1998 methods — the disciplined way, not ShadowStats' add-a-constant.

CPI. Current CPI-U +4.2% (12 mo to May 2026; core +2.9%). The documented per-change deltas vs 1990s methods: geometric-mean substitution (1999) ~0.2 pp/yr; upper-level substitution via chaining ~0.25 pp/yr; hedonics — no clean number (used on ~7.5% of items, net effect small/ambiguous; don't assign it a figure). BLS's own R-CPI-U-RS (today's methods run backward) shows a ~0.45 pp/yr gap over 1978–98, shrinking to ≤0.2 pp lately. So 1990s methods → CPI ≈ 4.4–4.7%, not the 7–8% ShadowStats sells. OER is a 1983 scope choice, not a 1990s methodology change — present its home-price-vs-OER spread (~50%, FHFA) separately.

Unemployment. U-3 4.3%, U-6 8.1% (May 2026); "not in labor force, want a job" ~6.2M. The 1994 CPS redesign added a 12-month-look-back + availability test, cutting discouraged from ~1.1–1.2M (1993) to ~541k (Q1 1994) — but those people were reclassified into the 6.2M "want a job" pool, not erased. U-6 and "want a job" are the transparent broad measures. LFPR ~62.4% vs the 67.3% 2000 peak (~5 pp gap) is largely demographic; the honesty check is prime-age (25–54) LFPR ~83.4–83.9%, near a record high — which argues against a large hidden-slack claim.

Bottom line: the methodology games move tenths of a point — real and fiscally convenient, but second-order. The first-order gap between the official story and lived experience is debasement, which a CPI denominated in the debased unit cannot show — so price it in metal:

The hard-money lens — gold-oz AND silver-oz at the time of measure (added 2026-06-16)

Each magnitude denominated in the metal price contemporaneous with the data point. Metal/nominal series fact; ratios arithmetic; debasement read labeled; the 2026 spot is geopolitically elevated (Hormuz) — stress-tested on calmer 2024 averages below.

• Gold/silver ratio across the same dates: 53 (1998) → 86 (2020) → 62 (2026) — silver cheapened to gold for two decades, then ran harder in the 2026 spike.
• Spike-robustness: on the calmer 2024-average gold ($2,388), the direction holds — income ~35 gold-oz (vs 132 = ~3.8× collapse), home ~173 gold-oz (vs 518 = ~3× cheaper). The wage/home erosion is not an artifact of the 2026 spot.
• The point: in nominal dollars wages, homes and stocks all rose; priced in metal at the time of measure, a year's wages and a median home fell by multiples, and even the S&P shrank against gold. That is the erosion the CPI-decimal debate can't see.

What is not claimed (discipline)

• ShadowStats adds a fixed constant to official CPI; it re-collects no prices and implies price levels inconsistent with observed nominal GDP/wages. WEAK/UNSUPPORTED as a number, even if its direction overlaps the real critiques.
• No evidence of fabricated microdata. The defensible case is methodology + revision-timing + collection degradation + incentive alignment + political pressure — not invented numbers. Conflating the two discredits the real case.

Posture

Treat each headline as the most flattering admissible estimate. Weight the benchmark revisions, GDI, U-6, real wages, and the gold lens over the first print. This is the epistemics layer for the whole repo — it calibrates trust in the inputs every other model consumes, and is kept out of the proofs.

← Research index · structured data: macro-official-data-integrity.json · macro-official-data-integrity.md