Federal Posting Signals

Federal Postings Register Policy Shocks

JAAT is applied to 3.09 million USAJOBS postings from January 2017 through March 2026. Federal posting volume falls sharply in 2025, so the policy-shock analysis tracks each signal as a share of active postings rather than as raw counts. The main finding is that the content of federal postings shifts in line with AI, sustainability, and DEIA executive actions: DEI keyword intensity falls after January 2025, while AI keywords and strict AI skills recover after an initial dip.

Count convention: the dashboard uses the full USAJOBS corpus currently available to JAAT, January 2017 through March 2026. Restricting the corpus to January 2018 through March 2026 yields 2.85 million postings.

Indexed Policy-Shock Signals

Figure C.1. USAJOBS Signals Indexed to 2021-2025 Administration Mean

This indexed view is presented to compare direction and timing across signals that have different raw units. Values above 100 are above each signal's February 2021-January 2025 baseline; values below 100 are below it.

Note: Each series is indexed to its own mean over February 2021 through January 2025. The AI keywords series uses the deduplicated any-keyword aggregate from raw USAJOBS AI concept maps.

Table C.1. Post-EO 14148 movement in posting-content signals

Note: Values are indexed to the February 2021-January 2025 mean for each signal. DEI keywords include diversity, equity, inclusion, and equal opportunity; AI keywords use the deduplicated any-keyword aggregate across raw USAJOBS AI concept maps. Strict and lenient AI codes use AIMatch code membership from the public AI taxonomy.

Core Policy Timing

The top-line timing emphasis follows the manuscript: federal AI, sustainability, and DEIA requirements under the prior administration, followed by January 2025 rescissions and replacement actions.

Signal Intensity and Baseline Comparisons

Figure C.2. USAJOBS DEI, Green, STEM, and Strict AI Code Intensity

This raw-intensity view is presented to show the scale of each posting-content signal before indexing. Because the units differ by signal family, the indexed figures are better for cross-signal comparison.

Note: DEI Keywords values are keyword mentions per 1,000 active postings; strict AI codes are strict-score AIMatch taxonomy codes per 1,000 active postings. Green and STEM task sets use prompt v2 two-model LLM agreement procedures; inclusion requires a two-model unanimous YES. The green-task procedure reviewed 5,839 complete task IDs, retained 1,056 task IDs, and produced 97.5% raw agreement and Cohen's kappa of 0.920. STEM has one unresolved adjudication error documented in the audit trail.

Green Task Agreement Procedure

The green-task series in Figures C.1-C.3 uses task IDs retained by a separate two-model agreement procedure, then counts LLM-validated green task mentions in monthly active USAJOBS postings. This makes the green signal a task-domain measure rather than a keyword-only sustainability vocabulary count.

Green task agreement summary

The procedure reviewed 5,839 complete task rows and retained 1,056 green task IDs. Most retained green tasks come from the canonical O*NET green-task universe; the discovery pathway adds a smaller set of green task IDs outside that canonical universe. Download the underlying workbook here: green task agreement data. The prompt and additional agreement-workbook links are in the Agreement and Audit Trail.

Figure C.3. USAJOBS Signals Indexed to January 2025

This supplemental index is presented to isolate post-transition movement. A value of 1.20 means the signal is 20% above its January 2025 level.

Note: Each series is divided by its own January 2025 value. This supplemental view isolates movement after the January 2025 transition.

Additional Policy Timing Context

These actions are not the main narrative spine for Figure C.1, but they are relevant context for interpreting federal posting language around equity, climate, energy, hiring, and merit.