---
type: "synthesis"
articles: ["a035", "a045", "a046"]
tags: ["evidence-base", "labor-economics", "calibration"]
id: "cross-canaries-shared-evidence"
sources: ["reskilling"]
sourceVaultSlug: "hbr-seg-reskilling"
originDay: 10
articleStem: "hbr-seg-reskilling"
sourceUrl: "(unified vault: 13 sources)"
sourceTitle: "HBR — People Ⅲ-B · Reskilling / L&D / talent / restructuring"
---
Three articles lean on the **same empirical spine** — Stanford's ADP payroll study, "Canaries in the Coal Mine" ([[evidence-stanford-canaries]]) — which is worth tracking as a single citation appearing in triplicate.

A046 cites it most precisely: a ~16% relative employment decline for workers aged 22–25 in the most AI-exposed occupations ([[claim-ai-displaces-early-career]]). A045 cites the widely-quoted **13%** entry-level decline from "highly accurate payroll data" ([[claim-ai-exposed-job-decline]]) — same dataset, rounded differently. A035 reports the demand-side mirror: −13% for automation-prone postings and +20% for augmentation-prone ([[claim-post-chatgpt-demand-shift]]), measured via [[concept-augmentation-score]] and [[framework-task-categorization-scoring]].

Crucially, A035's vault carries the broader corroboration lattice the other two lack: [[evidence-world-bank-labor-demand]] (strongest replication), [[evidence-anthropic-labor-study]], [[evidence-yale-budget-lab]] (the "too soon to tell" counter), and [[evidence-goldman-sachs-projection]]. **Calibration rule for the whole corpus:** the *direction* (early-career exposure, augmentation-vs-automation split) is robust and replicated; the *exact magnitudes* (−13/+20/16/50%) are article-level or scenario figures — the aggressive [[claim-50-percent-elimination]] (A045) sits far above observed data. Yale and Anthropic both temper any "economy-wide bifurcation" reading.