The first scored week of the Automation Displacement Index produced a Policy Lag Gap of −6 — implying policy was running ahead of displacement. The methodology says a negative gap is almost certainly a scoring error. It was. Here is the same week re-scored under a corrected model, with both readings shown.

New here? The full methodology is free and permanently unpaywalled — all eighteen indicators, their sources, every scoring anchor, and every known weakness. Worth reading first if you want to judge whether the correction below is a real one.

Pillar D was measuring legislative volume, not legislative relevance. By March 2026 lawmakers in 45 states had introduced 1,561 AI-related bills. Colorado's AI Act commenced June 30. Significant laws took effect in California, Texas and Illinois on January 1. Scored on activity, that is unambiguously a 4.

But those laws govern algorithmic discrimination, hiring-tool bias audits, disclosure and health care decision-making. A bias-audit requirement for an automated hiring tool does nothing for a customer service representative whose role was eliminated. The index was counting activity with no bearing on what it measures — while the federal government litigates to preempt those very laws, which the model couldn't see at all.

Pillar D is rescoped to displacement-relevant measures only: income support, unemployment insurance, worker transition funding, and binding restrictions on substitution.

The displaced content moves to a new C4 indicator rather than being discarded. Bias audits and human-review mandates don't respond to displacement — but they slow deployment, and deployment friction is exactly what pillar C measures. The laws were never irrelevant; they were filed under the wrong heading.

C4 scored 1 this week — binding audit and human-review duties are in force in multiple states but narrowly scoped to hiring and health care. That dilutes pillar C from 15.0 to 12.5 and pulls the composite from 44 to 41.

That direction is worth stating plainly. Regulatory friction is genuinely slowing enterprise AI deployment, and v1.0 had no way to register it. The corrected reading is less alarming than the original. A model that only ever revises upward is not a model.

The v1.0 baseline estimated the fix would land the Response Index around 10–15, giving a gap of +29 to +34. That was a loose figure written before the re-score was run. The true value is milder — RI 25, gap +16 — because D3 survives the rescope intact: blocking substitution is a displacement response regardless of its stated justification.

Pillars A, B and E are untouched, so every substantive finding stands.

AI remains the leading stated reason for US job cuts for four consecutive months — a streak with no precedent in Challenger's data. 14,029 AI-attributed cuts in June, 31% of the month's total; 101,743 year to date, roughly 23% of all cuts. For scale, AI accounted for 54,836 cuts in all of 2025, about 5% of the total.

The offshore channel is still arguing against the thesis. Staffing employment is up year over year and the Indian IT aggregate added headcount last quarter. The forecast predicted this channel would be cut first. It hasn't happened. Either the sequencing is wrong or the cut hasn't started, and I don't think we can tell yet.

The re-employment gap is the most interesting thing in the data. Initial claims hit the lowest level of 2026 while average unemployment duration climbed to 25.5 weeks from 21.5 a year earlier and long-term unemployed rose 286,000. Prime-age participation fell 0.6 points in a month while the unemployment rate fell — meaning people left the labor force rather than showing up as unemployed.

Almost nobody fired. The people who do lose a job taking far longer to return, or leaving entirely. That is precisely the signature of displacement running through hiring freezes and non-replacement rather than layoffs — and staying invisible in headline unemployment.

C3 remains ambiguously defined, blurring "what happened this week" with "where capability stands." Flagged in the original baseline and not yet fixed.

Three indicators are thinly sourced. A4 (trades wages), E1 (payroll receipts) and E2 (labor share) are scored at neutral by default rather than researched to conclusion. Together they carry roughly 14 of the 100 ADI points, and the direction they would move is unknown.

C4 has a definitional edge worth watching. Friction nominally about discrimination can function as de facto displacement protection if it makes substitution expensive enough. If that starts happening at scale, some of C4 may need to move back toward D.

Scores are not comparable across versions. 41 is the new baseline coordinate. Don't compute 44 to 41 as a weekly move — it's a methodology change, not a signal. Every future four-week delta measures against 41.

The ADI is an index, not a forecast. It compresses judgment about eighteen heterogeneous signals into one legible number so direction becomes readable week over week. It does not predict a date, and two analysts scoring this week would not produce identical numbers.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

AI will eliminate 300 million jobs in the next 5 years.

Yours doesn't have to be one of them.

Here's how to future-proof your career:

Start learning AI now