Diffusion lens

Last three status changes

1. 2026-09-10enterprise_multi_homing_share○unmeasured →to ○not yet measurableconf — → 45 · evaluate

   First reading. Menlo's 2023 enterprise survey put multi-model use at 60% (restated January 2024); later waves state the pattern without a share and the 2025 mid-year update gives an 11% vendor-switch rate instead. One numeric point, so not yet measurable.

   obs:5a3caabe
2. 2026-09-10neocloud_debt_terms○unmeasured →to ○not yet measurableconf — → 60 · evaluate

   First reading. CoreWeave's Q2 2026 10-Q states effective rates of 15% (DDTL 1.0), 11% (DDTL 2.0), 9% (DDTL 5.0) and 7% (non-recourse DDTL 4.0), a simple average of 10.5%, on $35.6B of total indebtedness; the August DDTL 5.5 facility prices at Term SOFR plus 5.5%. One quarter-end, so the direction rule cannot run: not yet measurable.

3. 2026-09-10depreciation_useful_life_changes○unmeasured →to ○not yet measurableconf — → 65 · evaluate

   First reading. Amazon's FY2025 10-K states a subset of servers and networking equipment moved from six to five years from 1 January 2025; Meta lengthened to 5.5 years, Alphabet holds six and Microsoft states two to six. One annual point per filer, so the direction rule cannot run yet: not yet measurable.

   obs:81efc38c

What would change our mind

• METR 50% time horizon: Normal = doubling time ≥ 7 months (≈213 days, METR's 2019–2025 trend); fast = ≤ 4 months (≈122 days), the AI 2027 assumption. Between the two is `emerging`. The bands are about the current pace, so the input is a log-linear fit over the 2024-onward window, the most recent window METR reports (89 days in the TH1.1 post of 29 Jan 2026), recomputed from METR's corrected data. On the 2023-onward window METR's published figure is 128.7 days, in the gap between the bands: the reading is window-sensitive and both numbers are shown.
• METR 80% time horizon: Same bands and same 2024-onward window as the 50% horizon, applied to our own fit because METR publishes doubling times for the 50% series only. The 2023-onward fit (horizon_doubling_days_80) sits in the gap between the bands.
• 50%/80% horizon ratio: Normal = ratio ≥ 5 (gap holding or widening); fast = ratio ≤ 2. The falsifying condition (≤ 2 AND 80% horizon > 8 h) is compound and lives in the thesis monitor, not the band.
• US labour productivity, year on year: Post-war nonfarm productivity growth averages about 2.1% a year, so anything up to 2.5% is trend. Goldman's AI scenario adds about 1.5 points a year; sustained prints of 3.5% or more would be that. Between is `emerging`.
• US total factor productivity, private nonfarm business: Long-run private nonfarm TFP growth is about 1% a year; Acemoglu's AI estimate is at most 0.66% over a decade, invisible at annual resolution. Goldman-scale is roughly a point a year on top of trend. Normal = at most 1.2%; fast = 2% or more.
• Developer productivity uplift (METR RCTs): Normal = uplift of at most about 20%, setting-dependent (the range of controlled studies to date); fast = broad uplift above 40%, the scale the productivity-boom claims need.
• Enterprise pilots with measurable P&L impact: Normal = at most 10% of pilots with measurable P&L impact, the typical enterprise-IT hit rate; fast = more than 30%.
• Share of work hours assisted by generative AI: Normal = single-digit share of hours in year three (PCs and the internet were here at the same age); fast = more than 20% of hours assisted.
• US firms using AI (Census BTOS): Normal = under about a third of firms in year three, the pace at which PCs and e-commerce reached half of firms over a decade or more; fast = more than half. Ramp's card-spend index (>50%) is a different instrument and not comparable.
• US businesses paying for AI (Ramp AI Index): Ramp's customers are younger and more tech-forward than the Census universe (56% here against 22% in BTOS at the same date), so the BTOS edges (a third, a half) do not transfer. Normal = under 45% of Ramp customers paying in year four; fast = 65% or more, i.e. most of even this sample paying. Between is emerging. The trend carries more information than the level: +0.4 points in August 2026 after a year of roughly a point a month.
• Agent-workdays per human workday in frontier research (self-reported): Below one agent-day per human-day, agents are tools inside a human-run process (normal); at three or more, most research effort is agent effort, the regime the AI 2027 'R&D multiplier' assumes (fast). Between is emerging. Tier 7 evidence caps the status at emerging whatever the number says.
• Human interventions on 4–8 hour agent tasks (self-reported): Humans steering most long tasks (half or more need intervention) is the normal-technology picture; a quarter or fewer is the regime where long agent runs are autonomous. Between is emerging. The reported value is a floor ('over half'), stored as 0.5.
• Safety brakes inside the labs (events, trailing year): Normal = at least one announced brake a year (the limits bind); fast = none, the pace the AI 2027 scenario assumes on the invention stage. Lab self-reports are tier 7, so the status is capped at emerging until an independent record exists.
• Frontier training compute growth (doubling time): Normal = doubling no faster than every 300 days (about 2.3x a year, the pre-2010 pace of compute growth); fast = every 182 days or less (4x a year or more, Epoch's 2010-2025 trend). Between is emerging.
• Training power draw growth (doubling time): Normal = doubling no faster than yearly (grid-connection pace); fast = every 230 days or less (3x a year, the pace a runaway buildout needs). Between is emerging.
• Hardware price-performance growth (doubling time): Normal = doubling no faster than every 500 days (roughly the +49% a year Epoch reports); fast = yearly or faster. Between is emerging.
• Inference price at fixed capability (halving time): Normal = halving no faster than every 230 days (about 3x a year, the hardware pace); fast = every 110 days or less (10x a year or more; Epoch's reading at this threshold is faster still). Between is emerging.
• Public text data exhaustion year (Epoch projection): Normal = exhaustion in 2028 or later (the paper's compute-optimal case); fast = 2026 or earlier (the overtrained case, which is what the AI 2027 scenario's compute path implies). Between is emerging.
• ARC-AGI-2 best score: Normal = under half (the 2025 frontier); fast = 85% or above (the ARC Prize grand-prize bar). Between is emerging.
• Continual-learning ladder (highest production rung): Appendix E's rule: L0-L3 (memory, population-level online learning, per-customer adapters) is normal; L5-L6 (persistent per-session weight updates, the six-month-employee test) is fast; L4 (organisation knowledge in weights) is emerging.
• FDA-authorised AI-enabled devices, growth: Normal = the cumulative count growing under 50% a year (the 2019-2024 pace); fast = doubling or faster, which is what an ungated regulated domain would show. Between is emerging.
• US state AI bills introduced (legislative year): Normal = five hundred or more bills a year (the 2024-2025 volume, legislatures reacting at their own speed); fast = under two hundred, legislatures standing aside. Between is emerging.
• Active RL-environment vendors (directory count): Normal = a few dozen vendors (a specialist market); fast = 150 or more (a commodity market forming in months). Between is emerging.
• Ord half-life misfit (observed 50/80 ratio vs constant hazard): Normal = misfit of 1.5 or more (observed ratios of 5-10x against the constant-hazard 3.1x, the pattern since 2024); fast = within 10% of the constant-hazard prediction, which would mean long-task reliability tracking short-task success. Between is emerging.
• FDA-authorised devices built on a large language model: Normal = a handful of LLM-based authorisations by end-2026, given a 510(k) path of roughly six months and a first clearance in December 2025; fast = ten or more, the regulatory throughput a fast scenario needs. Between is `emerging`.
• Employment in AI-exposed occupations, year on year: Normal = within one percent of flat, ordinary occupational churn; fast = a 3% or worse decline in exposed occupations, which is what a broad AI-attributable displacement looks like in payroll data. Between is `emerging`.
• Entry-level employment shortfall in AI-exposed occupations: Normal = a shortfall under 10 points, inside the range youth employment swings over a cycle; fast = 25 points or more. The current reading sits between, so it is `emerging`.
• Exposed vs unexposed occupation employment gap (Revelio): Normal = a relative gap of at most 5 points over nearly four years, within the drift between occupation groups in any period; fast = a 10-point or larger gap. Between is `emerging`.
• California unemployment claims from AI-exposed occupations: Normal = monthly moves of up to 2% in the smoothed series, the noise floor of claims data; fast = 5% or more month on month, sustained. Between is `emerging`.
• Recent-graduate unemployment rate (NY Fed): Normal = up to 6%, the top of the 2010s range for recent graduates; fast = 8% or more outside a recession. Between is `emerging`.
• Cross-tracker concordance (labour): Normal = at most one tracker past its threshold; fast = two; falsifying = three or more concurrently, per the thesis rules in the brief.
• Augmentation vs automation share (Anthropic Economic Index): Normal = augmentation at least half of classified use; fast = automation clearly dominant (augmentation under 40%). Between is `emerging`.
• US labour share of income, year on year: Normal = year-on-year moves of at most 2%, the size of ordinary cyclical swings in the index; fast = 5% or worse a year, roughly the pace that takes the share from 60% to 45% by 2030 (the extreme scenario). Between is `emerging`.
• Customer-support productivity uplift from a generative-AI assistant: Normal = an average uplift under 20%, the range prior process tools (CRM, knowledge bases) delivered in the same job; fast = 50% or more, the order of magnitude the fast scenario needs from a single tool. Between is `emerging`.
• Executives reporting no AI impact on their own firm: Normal = at least 70% of firms report no measurable impact three years in, as with PCs at the same age; fast = a majority reporting impact (no-impact share under 40%). Between is `emerging`.
• Frontier models completing expert legal tasks end to end (Harvey LAB): Normal = the best frontier model completes under 30% of expert tasks end to end; fast = above 70%, where substitution rather than assistance becomes the product. Between is `emerging`.
• Agent reliability across repeated runs (tau-bench pass^k): Normal = under 70% of tasks pass all four runs, so the best agent still fails one attempt in three; fast = 90% or more at k of four or higher, the reliability a workflow can be built on. Between is `emerging`.
• Progress against the AI 2027 quantitative predictions: Normal = 60% or less of the predicted pace (a delay that compounds to years); fast = 90% or more (on schedule). Between is `emerging`.
• Work-related share of ChatGPT consumer messages: Normal = under 40% of consumer messages are work-related, as personal use led with PCs and the web; fast = a work majority, meaning the consumer product has been pulled into production workflows. Between is `emerging`.
• Usage as a share of theoretical exposure: Normal = under 60% of exposed workers use the tools weekly three years in; fast = 80% or more, the gap closed. Between is `emerging`.
• Workers highly exposed to AI with low adaptive capacity: Normal = under five million workers (about 3% of the workforce), the scale of the China trade shock spread over a decade; fast = ten million or more. Between is `emerging`.
• Occupational-mix dissimilarity since ChatGPT (Yale Budget Lab): Normal = within one and a half times the pre-AI baseline's reading at the same age (the January 2021 baseline sat near 5 pp forty months in); fast = 10 pp or more, twice the pre-AI pace. Between is `emerging`.
• Chatbot effect on earnings and hours in exposed occupations (Denmark): Normal = earnings and hours effects within two percent two years in; fast = a decline of five percent or more for exposed workers. Between is `emerging`.
• Freelance writing earnings after ChatGPT (Upwork): Normal = declines under 10% in the most substitutable freelance categories, a tool effect at the margin; fast = a 20% or worse collapse in freelance earnings. Between is `emerging`.
• Wage premium for AI skills (PwC AI Jobs Barometer): Normal = a premium between 25% and 75%, large but of the size prior scarce IT skills carried; fast = a premium of 100% or more, wages doubling for a skill. Between is `emerging`.
• Expert-data market run-rate (Mercor): Normal = the largest vendor under one billion gross, the scale the data-labelling leaders reached over a decade; fast = two billion or more at a single vendor, doubling within a year. Between is `emerging`.
• Internal-to-public deployment gap at the frontier: Normal = a gap under six months, ordinary red-teaming and productisation; fast = a year or more, a lab running internally on a model the public does not have. Between is `emerging`.
• Waymo paid rides per week: Normal = under a million paid rides a week, well under one percent of US ride-hail trips and scaling one city at a time; fast = five million or more, a national-scale substitution. Between is `emerging`.