# Gate-1 Backtest & Falsification Exhibit — Labor Index

*Generated 2026-07-12T20:54:43.336Z. Pre-registration: docs/19-preregistration-labor-backtest.md.*


## Verdict: ❌ FAIL

- Cohorts and thresholds per docs/19 Amendment 1 (2026-07-12): validation majors {15, 27, 43}; V1 = ≥2/3 with major 43 required; F1 unchanged at ≥4/5.
- Non-gating diagnostic — sustained ≥4-quarter run overlapping the onset window but with onset OUTSIDE it: major 15 (run onset 2021-Q1), major 27 (run onset 2021-Q1), major 43 (run onset 2021-Q4), major 49 (run onset 2021-Q4), major 31 (run onset 2022-Q1), major 35 (run onset 2021-Q3). Informational only; the gating V1 rule is onset-in-window.

## Registered tests

- **V1 (validation — collapse appears)** — ❌ FAIL. 0/3 validation majors show a ≥4-quarter >2σ decline with onset in 2022-Q4–2024-Q4 (threshold ≥2; required major 43: NOT qualifying).
- **V2 (magnitude sanity)** — ✅ PASS. Mean validation cycle-adjusted deficit 16.4pp over 2023-Q1–2025-Q4 (threshold ≥5pp).
- **F1 (falsification — controls quiet)** — ❌ FAIL. 1/5 falsification majors stay quiet in the AI era (no ≥4-quarter >2σ run overlapping 2022-Q4→end; pre-2022Q4-resolved COVID runs out of scope per Amendment 3); threshold ≥4, marginal = 3.
- **F2 (within-segment placebo)** — ✅ PASS. Placebo ratio (A05 deficit / entry deficit) = -0.131 over 2023-Q1–2025-Q4 (pass <0.5, marginal 0.5–0.75). Entry deficit 0.146, A05 deficit -0.019.

## Registered thresholds (restated verbatim from docs/19)

- **Baseline (D2):** 2015–2019 same-quarter average per family (2015-Q1–2019-Q4).
- **σ:** estimated per family from 2015–2019 residuals of the cycle-adjusted series.
- **V1 (per Amendment 1):** ≥4 consecutive quarters >2σ below baseline, onset in 2022-Q4–2024-Q4, in ≥2 of 3 validation majors, with major 43 required among the qualifiers. The through-window column is a non-gating diagnostic.
- **V2:** mean cycle-adjusted entry deficit ≥5pp over 2023-Q1–2025-Q4.
- **F1:** no ≥4-quarter >2σ run in ≥4 of 5 falsification majors.
- **F2:** A05 (35–44) deficit < half the entry deficit over 2023-Q1–2025-Q4 (pass <0.5, marginal 0.5–0.75).
- **Outcomes:** Pass = all four hold. Partial = (V1 or V2) holds while F1/F2 are marginal. Fail = anything else.

## Data window & coverage

- **Usable quarters:** 2015-Q1 → 2025-Q3 (43 quarters).
- **Balanced panel (docs/19 Amendment 2):** 49 of 51 requested states. Excluded terminated QWI series: state 02 (last published 2016-Q2); state 26 (last published 2021-Q4).
- **NAICS sectors:** 19. **Age bands:** A02, A03, A04, A05.
- Quarters after the latest fully-covered panel quarter are dropped as the QWI release frontier (see coverageByQuarter in the JSON report).

## Validation cohort (high-AI-exposure) — per family

| SOC | Family | μ_base | σ | breach<br/>(μ−2σ) | longest<br/>run | qualifying<br/>onset | through-window<br/>(diagnostic) | mean cycle<br/>deficit (pp) | entry<br/>deficit | A05<br/>deficit |
|---|---|---|---|---|---|---|---|---|---|---|
| 15 | Computer & Mathematical (software) | 1.001 | 0.0281 | 0.944 | 19 | — | 2021-Q1 | 17.71 | 0.167 | -0.010 |
| 27 | Arts/Design/Media | 1.001 | 0.0335 | 0.934 | 19 | — | 2021-Q1 | 15.77 | 0.157 | 0.001 |
| 43 | Office & Administrative Support | 1.001 | 0.0337 | 0.934 | 16 | — | 2021-Q4 | 15.70 | 0.114 | -0.048 |

## Falsification cohort (low-AI-exposure controls) — per family

| SOC | Family | μ_base | σ | breach<br/>(μ−2σ) | longest<br/>run | qualifying<br/>onset | through-window<br/>(diagnostic) | mean cycle<br/>deficit (pp) | entry<br/>deficit | A05<br/>deficit |
|---|---|---|---|---|---|---|---|---|---|---|
| 47 | Construction & Extraction | 1.000 | 0.0127 | 0.974 | 0 | — | — | -4.20 | 0.065 | 0.103 |
| 49 | Installation, Maintenance & Repair | 1.001 | 0.0353 | 0.931 | 16 | — | 2021-Q4 | 12.78 | 0.127 | 0.001 |
| 31 | Healthcare Support | 1.001 | 0.0275 | 0.946 | 15 | — | 2022-Q1 | 10.85 | -0.108 | -0.242 |
| 35 | Food Preparation & Serving | 1.005 | 0.0784 | 0.849 | 17 | — | 2021-Q3 | 21.30 | 0.176 | -0.040 |
| 53 | Transportation & Material Moving | 1.002 | 0.0380 | 0.926 | 10 | 2023-Q2 | 2023-Q2 | 13.51 | 0.067 | -0.075 |