Every indicator, every scoring anchor, and every known weakness. Published free and unpaywalled, permanently — because a number you can’t interrogate isn’t worth paying for.

This page describes v1.2, current as of October 2026. What changed from v1.1, and why, is at the bottom.

Nineteen signals, scored 0 to 4 each week against fixed written anchors, rolled into two numbers.

The ADI measures escalation, 0–100. The Response Index measures policy response on the same scale. Their difference is the Policy Lag Gap — positive and widening means displacement is outrunning the response.

Keeping response out of the composite is deliberate. Response is not escalation, and folding them together would hide the one number that matters most.

Pillar points = (sum of scores ÷ (4 × indicator count)) × weight. Equal weighting within a pillar is deliberate — unequal weights would imply a precision the signals don’t support.

Policy Lag Gap bands: 0–5 response tracking · 6–15 mild lag · 16–30 clear lag · 31+ severe lag. A negative gap means policy is running ahead of displacement, which historically almost never happens — investigate it as a scoring error before treating it as real.

A1 · Entry-level vs senior posting ratio — weekly. Source: Indeed Hiring Lab, LinkUp, or a manual count across a fixed occupation basket.

0 ratio rising, junior hiring recovering · 1 flat within normal variance · 2 falling steadily 3+ weeks · 3 down sharply, junior reqs pulled · 4 junior hiring near-frozen in 2+ exposed occupations

A2 · Prime-age (25–54) participation rate — monthly. Source: BLS Employment Situation, table A-1.

0 rising · 1 flat within 0.1pt · 2 down 0.1–0.2pt over three months · 3 down 0.3–0.5pt over three months · 4 down 0.5pt+ with no cyclical explanation

A3 · Continued claims, four-week average — weekly. Source: DOL ETA 539, Thursdays.

0 falling · 1 flat · 2 rising modestly, no concentration · 3 rising with information and professional-services concentration · 4 sharp rise concentrated in exposed sectors

A4 · Skilled trades wage growth vs trend — monthly. This is the shelter-flooding test.

Series: BLS Current Employment Statistics, average hourly earnings of all employees in construction (CES2000000003), cross-checked against installation, maintenance and repair occupations. Compared against all-employee average hourly earnings (CES0500000003) and CPI-U. The test is whether trades wage growth holds its premium over the broad wage index, not whether it is positive.

0 accelerating, trades still scarce · 1 on trend · 2 decelerating slightly · 3 clearly below trend, applicant volume up · 4 flat or negative real growth in the trades

B1 · Indian IT/BPM headcount — quarterly. Source: TCS, Infosys, Wipro, HCL results.

0 net adds accelerating · 1 positive but slowing · 2 roughly flat · 3 net reductions at two or more · 4 reductions across all four, or guidance cut citing automation

B2a · Office and clerical staffing hours — weekly. Source: ASA Staffing Index, office–clerical segment, year-over-year.

0 rising · 1 flat · 2 down 1–3% year over year · 3 down 4–7% sustained · 4 down 8% or more, or accelerating

B2b · Aggregate staffing volume — weekly. Source: ASA Staffing Index, all segments.

0 rising · 1 flat · 2 declining modestly · 3 declining with IT and clerical concentration · 4 sharp decline plus guidance cuts citing automation

B3 · Named BPO/CSR/MSP losses attributed to automation — continuous. Source: trade press, earnings calls, client announcements.

0 none · 1 isolated mention · 2 two or three named accounts in a month · 3 multiple large accounts, a provider warns · 4 a major provider restructures or exits a service line

C1 · Enterprise agent deployments at large employers — continuous. Source: earnings calls, press releases, vendor case studies.

0 pilots only · 1 production in narrow low-stakes functions · 2 production in a core function at one large employer · 3 multiple large employers, core functions, headcount tied · 4 explicitly framed as headcount replacement at scale

C2 · Layoffs explicitly attributed to AI — monthly. Source: Challenger Gray monthly job cut report, WARN filings, layoffs.fyi. Scored on absolute counts and their trend, not on whether attribution exists at all.

0 none attributed · 1 scattered attribution, plausibly covering other causes · 2 consistent monthly attribution with counts flat or falling · 3 counts rising month over month · 4 multiple 1,000+ actions attributed and rising within one quarter

C3 · Frontier agentic capability — continuous. Source: lab release notes, long-horizon agentic benchmarks. Scored on current deployed state, not on the most recent announcement.

0 no meaningful movement · 1 incremental gains · 2 notable jump on long-horizon tasks · 3 credibly performs multi-day autonomous work · 4 capability jump plus immediate enterprise availability

C4 · Regulatory friction on deployment — continuous. Source: bias-audit and disclosure duties, human-review mandates, preemption litigation. Scored in the escalation direction: high means little friction.

0 binding audit or human-review duties in force across multiple states, broadly scoped · 1 binding duties in force but narrowly scoped (hiring, health care) · 2 duties adopted somewhere but weakly enforced · 3 little effective friction, preemption gaining · 4 no effective friction, deployment legally unimpeded

Timing rule for C4: a law scores when it is in force, not when it is signed. Enacted-but-not-yet-operative measures are noted in the issue and scored on their effective date.

E1 · Payroll tax receipts vs prior-year trend — monthly.

Series: amounts withheld from workers’ paychecks, as reported in the Congressional Budget Office’s Monthly Budget Review, fiscal-year-to-date against the same period a year earlier. The CBO publication is used rather than the Monthly Treasury Statement directly because it separates withholding from other individual income tax receipts and states the year-over-year change explicitly.

0 above trend · 1 on trend · 2 slightly below · 3 below trend three months running · 4 below trend with employment flat — wage-base erosion

E2 · Labor share — quarterly.

Series: BLS nonfarm business sector labor share, from the quarterly Productivity and Costs release. Total compensation as a share of nonfarm business output, available back to 1947.

This replaces the BEA NIPA compensation share of gross domestic income named in v1.1. The BEA series remains a useful cross-check, but the BLS series is quarterly with a longer history, ties directly to the productivity data this indicator is meant to be read against, and is published with explicit historical context. The two are not interchangeable and produce different levels. Readings before October 2026 were scored at a neutral default and should not be treated as measurements of either series.

0 rising · 1 flat · 2 declining slightly · 3 declining while productivity rises sharply · 4 sharp decline, gains landing on capital

E3 · Real wage growth, production & nonsupervisory — monthly. Source: BLS CES real earnings.

0 positive and accelerating · 1 positive, on trend · 2 slowing but positive · 3 flat in real terms · 4 negative in real terms while output rises

Displacement-relevant measures only. Bias-audit, disclosure and discrimination law is not response — it scores under C4 as deployment friction.

D1 · Federal displacement-response legislation — income support, UI, worker transition only.

0 nothing, or the federal posture is actively deregulatory · 1 rhetoric only · 2 a displacement-focused bill introduced · 3 hearings held, a bill advances · 4 enacted

D2 · State displacement-response legislation — excludes bias-audit and disclosure bills.

0 nothing addressing displacement · 1 scattered introductions · 2 multiple states with active bills · 3 a state enacts · 4 several states enact, pattern forming

D3 · Restrictions on substitution — licensing, scope of practice, operator mandates.

0 none · 1 discussion only · 2 a board or agency opens a proceeding · 3 a binding restriction adopted in one state · 4 restrictions across multiple states or federally

D4 · Organized labor action on automation

0 none · 1 statements only · 2 automation language raised in negotiations · 3 protections won in a major contract · 4 strike or major action centered on automation

Twenty to forty minutes. Carry forward last week, then update only what has fresh data: A1, A3, B2a and B2b weekly; the continuous signals as news warrants; monthly and quarterly indicators only in their release weeks.

Check the staleness figure. Above 40% carried-forward weight, the reading is low-confidence and should be published with that flag attached.

A single week is noise. Only a four-week move of six points or more is worth writing about. Only a twelve-week trend justifies changing a conclusion.

Watch the gap, not just the number. The ADI can sit flat at 45 for a year while the Response Index falls — a meaningfully worsening situation the composite alone would hide.

Score conservatively. Between two levels, take the lower one. Upward drift is the standard failure mode of hand-scored indices, because escalation makes better copy than stability.

Three changes, all made because the v1.1 version of an indicator was giving the wrong answer.

B2 split into B2a and B2b. A single staffing-volume indicator averaged away the only divergence that mattered. Aggregate staffing was growing while office and clerical hours fell every month, and one score could not say both. Splitting them took pillar B from three indicators to four, which is why pillar B arithmetic changed.

C2 rescored on absolute counts. The v1.1 anchor asked whether multiple thousand-plus actions had been attributed within a quarter. Once that threshold was crossed it stayed crossed, so the indicator held at 4 through months when AI-attributed cuts were actually falling. The rescored version registers direction, and caught the reversal in the first month it was live.

C3 scored on current state. It was drifting upward on announcements rather than on what is deployed.

A4, E1 and E2 carried named sources in v1.1 but were scored at a neutral 1 by default, because I had not actually pulled the series. That is roughly 14 of 100 points resting on an assumption, for five months, which I flagged in five consecutive issues before fixing it.

All three now have an explicit series, documented above. Two things to note about the result.

First, the placeholders were not neutral. Once sourced, E1 scored 0 and E2 scored 3 — opposite ends of the scale, where an identical default had sat for both. A neutral default is not a conservative choice; it is a guess that happens to be invisible.

Second, E2 changed series. The v1.1 source was BEA NIPA compensation share of gross domestic income; the indicator now uses the BLS nonfarm business labor share, for the reasons given above. That is a change to what the indicator measures, not just to where the number comes from, and it is recorded here rather than absorbed quietly.

Scores are not comparable across versions, and readings before October 2026 should be read with A4, E1 and E2 treated as unmeasured.

The v1.0 model produced a Policy Lag Gap of −6 on its first scored week, implying policy was running ahead of displacement. Its own instructions said to investigate that as a scoring error. It was one.

Pillar D was measuring legislative volume rather than relevance. The 1,561 state AI bills that produced a high D2 score govern algorithmic discrimination, hiring-tool bias audits, disclosure and health care decision-making — none of which respond to displacement.

The fix has two parts. Pillar D is rescoped to displacement-relevant measures. The displaced content moves to C4 rather than being discarded, because bias audits and human-review mandates don’t respond to displacement but they do slow deployment — which is exactly what pillar C measures. The laws were never irrelevant; they were filed under the wrong heading.

Re-scored under v1.1, the July 26 2026 baseline reads ADI 41 (was 44), Response Index 25 (was 50), gap +16 (was −6). Note that C4 pulled the composite down: regulatory friction is genuinely slowing deployment and v1.0 couldn’t register it.

This is an index, not a probability or a forecast of any date. Scoring is subjective and two analysts will not produce identical readings. Several indicators have no clean public weekly series and depend on a hand-built basket, which introduces construction bias. The pillar weights encode a specific thesis about how displacement propagates; if that thesis is wrong, the weights are wrong with it.

And the model cannot see the thing it most wants to measure — work that is never posted, never offshored and never announced, because the role simply stops being backfilled.