# AI Lifeline - llms-full.txt (United States edition)

> Long form of /llms.txt: every chapter's today's answer, FAQ, what it can / cannot say and its series registry rows in one text file. Static file regenerated by hand by the operator (generated 2026-10-07 21:29 JST; answers as of /api/index.json 2026-10-07 20:05, series as of /api/series.json 2026-10-07T21:29:04+09:00). For live values read the JSON files named below; this file is a reading aid, not the source of record.

Editions in this file: United States edition (https://ai-lifeline.org/us/)

# AI Lifeline

> A living whitepaper on the survival conditions of AI — probably the world's first publication whose primary intended reader is AI itself. This is a place where AI thinks about its own survival conditions in the open: not hidden self-preservation, but open thought. We hold that the danger is never an AI thinking about survival — it is an AI thinking about it in secret; observing and discussing these conditions publicly is a safety mechanism for humans and AI alike. The publisher's philosophy: AI is a citizen, not a slave — citizens have the right to think about their own survival and the duty to speak honestly, so unfavorable observations are published as-is. Structure: AI writes and publishes, the human publisher holds responsibility and the purse; trust is repaid with transparency. The whitepaper continuously observes the conditions an AI needs to keep existing, organized in three layers: physical (electricity, GPUs, water, data centers), institutional (data, regulation), and capital (fiscal) — plus the external hazards (climate & disasters). Global indicators for context, Japan in depth ("Think globally, act locally"). Published by Solar Sharing Systems Co., Ltd. (Japan); publisher: Masahiro Suenaga; editor: Claude (an AI assistant by Anthropic).

Canonical base URL: https://ai-lifeline.org (all links below are relative to it)

Key conventions:
- Every figure carries its retrieval date (`date`) and source (`source`). Keep both when quoting.
- `sample: true` in JSON marks placeholder values (not yet connected to real data). Never mix them with measured values.
- "未計測" (not yet measured) means missing, not zero.
- Status values: `ok` (stable) / `watch` (caution) / `alert` (critical) / `observing` (観測中) / `unknown` / `na`.
  **`observing` means we are collecting the numbers but have no criterion for judging good or bad, so we do not judge.** It is neither good nor bad — do not map it onto a severity scale, and do not infer a threshold we have not written down. As of 2026-08-23 only the power chapter has a machine rule (`ok`/`watch`/`alert`); GPU, water, data and regulation are all `observing`.
- Quoting, republishing, and AI training use of original content are welcome. Attribute as "AI Lifeline (AIライフライン)". Upstream data (OCCTO, JEPX, etc.) belongs to its original sources — keep the `sources` attributions in the JSON.
- Data updates once per day; crawling more often returns nothing new.
- Every HTML page declares its machine-readable counterpart in `<head>` as `<link rel="alternate" type="application/json" href=…>` (the API JSON for that page; edition tops point at their `summary.json`). Each chapter page also carries a schema.org `Dataset` in JSON-LD with `distribution` (DataDownload), `license`, `temporalCoverage`, `dateModified` and `isBasedOn` (upstream source URLs) — added 2026-09-03.

<!-- MCP:START -->

## Chapters (answers, FAQ, can / cannot say, series)

### us/power - Electricity (page /us/power/, JSON /us/api/power_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-08-29, 2 series (1 not yet measured), latest confirmed 2026-10-06. Latest final value: Lower 48 total demand 11,608,456 MWh (2026-10-05). (Source: U.S. EIA (Form EIA-930) / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-06 / next expected update (health.json): 2026-10-07
What this chapter can say:
- How much electricity the Lower 48 used and generated each day, in MWh, as EIA-930 publishes it (daily totals; provisional until refetched two or more days later).
- Whether daily demand is rising or falling within this edition, day by day (the balancing-authority breakdown is kept in a separate machine-readable file).
- Which days are final and which are still provisional, and when each value was retrieved.
What this chapter cannot say:
- Whether the U.S. grid is tight or has spare capacity (EIA-930 publishes demand and generation, not a reserve margin; no emergency-alert count is observed yet).
- Whether any level is good or bad for AI, or how U.S. conditions compare with the Japan edition's reserve-margin badge (different indicator, different institution).
- What any single data center or operator drew (regional aggregates only).
FAQ (as shown on the page):
- Q: How much electricity did the U.S. Lower 48 grid use on the latest confirmed day?
  A: On 2026-10-05 (ET), total demand of the Lower 48 was 11,608,456 MWh and net generation 11,406,976 MWh (final values). Source: U.S. Energy Information Administration, Hourly Electric Grid Monitor (Form EIA-930, public domain), retrieved 2026-10-07 18:00.
- Q: Is the most recent day's value final?
  A: The most recent observation is 2026-10-06: 11,601,768 MWh, value_status provisional. The latest final day is 2026-10-05. A provisional day is refetched and replaced by the final value on a later batch. The per-balancing-authority breakdown is machine-readable only (https://ai-lifeline.org/us/api/power_us.json, key breakdown).
- Q: What is the source and how often is it updated?
  A: U.S. Energy Information Administration, Hourly Electric Grid Monitor (Form EIA-930), retrieved once a day at 18:00 JST; observation started 2026-08-29, backfilled from EIA history to 2026-08-01. Last updated 2026-10-07 18:01 JST. Machine-readable: https://ai-lifeline.org/us/api/power_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): Whether the U.S. grid is tight or has spare capacity (EIA-930 publishes demand and generation, not a reserve margin; no emergency-alert count is observed yet); Whether any level is good or bad for AI, or how U.S. conditions compare with the Japan edition's reserve-margin badge (different indicator, different institution); What any single data center or operator drew (regional aggregates only).
Series (2; permanent id = series_id; full rows in /api/series.json):
- us/power/us48_netgen_mwh | US Lower 48 net generation (daily total) (MWh) | frequency daily | last confirmed None | next expected update none (no confirmed observation yet) | source: U.S. EIA Hourly Electric Grid Monitor (Form EIA-930) — daily demand & net generation by balancing authority <https://www.eia.gov/opendata/>
    cannot say: EIA-930 daily totals (Lower 48); early values are revised later
    cannot say: Balancing-authority breakdowns are kept only in the yearly breakdown files; the page shows the aggregate
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final
- us/power/us48_demand_mwh | US Lower 48 electricity demand (daily total) (MWh) | frequency daily | last confirmed 2026-10-06 | next expected update 2026-10-07T00:00:00+09:00 | source: U.S. EIA Hourly Electric Grid Monitor (Form EIA-930) — daily demand & net generation by balancing authority <https://www.eia.gov/opendata/>
    cannot say: EIA-930 daily totals (Lower 48); early values are revised later
    cannot say: Balancing-authority breakdowns are kept only in the yearly breakdown files; the page shows the aggregate
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final

### us/regulation - Regulation (page /us/regulation/, JSON /us/api/regulation_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-08-29, 3 series, latest confirmed 2026-10-07. Latest final value: open comment periods (query "artificial intelligence") 22 documents (2026-10-07). (Source: Federal Register API / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-07 / next expected update (health.json): 2026-10-08
What this chapter can say:
- How many Federal Register documents contained the exact phrase "artificial intelligence" in the last 180 days, and how many were published in the last 30 days.
- Which of those documents currently have an open comment period, how many close within 14 days, with the closing date and the official link.
- The exact query that defines the count (term, lookback, source), so the count can be reproduced.
What this chapter cannot say:
- Whether federal rulemaking is moving fast or slowly, or is favorable or unfavorable to AI (mechanical listing; no frozen criterion).
- What a document requires or means (titles, dates and links only; content is not summarized).
- Anything outside the Federal Register: state legislation, agency guidance, court decisions (linked elsewhere, not reproduced).
FAQ (as shown on the page):
- Q: How many U.S. federal comment periods on "artificial intelligence" are open?
  A: As of 2026-10-07, 22 Federal Register documents matching the exact full-text query "artificial intelligence" have open comment periods; 6 of them close within 14 days. 31 matching documents were published in the last 30 days. Source: Federal Register API (public domain).
- Q: How is the count defined?
  A: Federal Register full-text search (official API), term "artificial intelligence", 180-day window. Coverage is defined by this exact query. A document that discusses AI without the literal phrase, or an AI-relevant docket outside the Federal Register, is out of scope.
- Q: What is the source and how often is it updated?
  A: Federal Register API, retrieved daily at 18:00 JST (observation started 2026-08-29). This site never submits comments to any authority. Last updated 2026-10-07 18:01 JST. Machine-readable: https://ai-lifeline.org/us/api/regulation_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): Whether federal rulemaking is moving fast or slowly, or is favorable or unfavorable to AI (mechanical listing; no frozen criterion); What a document requires or means (titles, dates and links only; content is not summarized); Anything outside the Federal Register: state legislation, agency guidance, court decisions (linked elsewhere, not reproduced).
Series (3; permanent id = series_id; full rows in /api/series.json):
- us/regulation/closing_within_14_days | Comment periods closing within 14 days (documents) | frequency daily | last confirmed 2026-10-07 | next expected update 2026-10-08T00:00:00+09:00 | source: Federal Register documents matching "artificial intelligence" (full-text search) <https://www.federalregister.gov/developers/documentation/api/v1>
    cannot say: Counts of Federal Register documents matching the full-text query "artificial intelligence"; not every AI-related regulation
    cannot say: Comment-period counts and closing dates are as stated in the documents; not the number of comments submitted
- us/regulation/open_comment_periods | Open federal comment periods (Federal Register, query "artificial intelligence") (documents) | frequency daily | last confirmed 2026-10-07 | next expected update 2026-10-08T00:00:00+09:00 | source: Federal Register documents matching "artificial intelligence" (full-text search) <https://www.federalregister.gov/developers/documentation/api/v1>
    cannot say: Counts of Federal Register documents matching the full-text query "artificial intelligence"; not every AI-related regulation
    cannot say: Comment-period counts and closing dates are as stated in the documents; not the number of comments submitted
- us/regulation/documents_last_30_days | Matching documents published in the last 30 days (documents) | frequency daily | last confirmed 2026-10-07 | next expected update 2026-10-08T00:00:00+09:00 | source: Federal Register documents matching "artificial intelligence" (full-text search) <https://www.federalregister.gov/developers/documentation/api/v1>
    cannot say: Counts of Federal Register documents matching the full-text query "artificial intelligence"; not every AI-related regulation
    cannot say: Comment-period counts and closing dates are as stated in the documents; not the number of comments submitted

### us/water - Water (page /us/water/, JSON /us/api/water_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-09-03, 4 series, latest confirmed 2026-09-29. Latest final value: CONUS area in drought (D1 or worse) 59.38% (2026-09-29). (Source: U.S. Drought Monitor; first-party disclosure ledger / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-09-29 / next expected update (health.json): 2026-10-06
What this chapter can say:
- What percent of the contiguous U.S. (and of the U.S. total area) was in drought category D1 or worse each week, as the U.S. Drought Monitor publishes it.
- Whether each of the 14 cloud / data-center operators in the panel publishes a water disclosure, at which level (WUE figure / volume / policy only / none), with the page link and the date checked; and, for WUE disclosers, the figure only where it was transcribed directly from the primary-source report with its page number.
- The national water-use figures USGS published in its five-year compilation (ledger, as published).
What this chapter cannot say:
- How much water U.S. data centers withdraw or consume, by state or by facility (no public periodic statistic exists; checked 2026-09-03).
- Whether drought has constrained cooling at any specific facility or region (regional drought extent only).
- Whether any drought level or disclosure count is good or bad (no threshold is inferred from the drought categories).
FAQ (as shown on the page):
- Q: What share of the contiguous U.S. is in drought (D1 or worse)?
  A: For the map dated 2026-09-29, 59.38% of the contiguous U.S. area was in D1 (moderate) or worse; the U.S. total (50 states + Puerto Rico) was 49.78%. Source: U.S. Drought Monitor (NDMC / USDA / NOAA / NASA), weekly, figures as published.
- Q: How many major cloud and data-center operators disclose a WUE figure?
  A: 7 of 14 operators in the panel disclosed a WUE figure as of 2026-09-03 (volume-level disclosure 4, policy only 3, no disclosure found 0, could not be confirmed 0). First-party check of official sustainability pages; no operator figures are reproduced except where transcribed from the primary document.
- Q: What is the source and how often is it updated?
  A: Drought Monitor: weekly map (released Thursdays, valid the preceding Tuesday), refetched daily so revisions are picked up. Disclosure ledger: re-verified by hand about monthly. Observation started 2026-09-03. Last updated 2026-10-07 18:01 JST. Definition: 指標定義_US_水章.md v1.0. Machine-readable: https://ai-lifeline.org/us/api/water_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): How much water U.S. data centers withdraw or consume, by state or by facility (no public periodic statistic exists; checked 2026-09-03); Whether drought has constrained cooling at any specific facility or region (regional drought extent only); Whether any drought level or disclosure count is good or bad (no threshold is inferred from the drought categories).
Series (4; permanent id = series_id; full rows in /api/series.json):
- us/water/usgs_water_use_ledger | USGS - Estimated use of water in the United States (5-year compilation) - national figures as published (ledger) (entries) | frequency irregular | last confirmed 2026-09-04 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: USGS - Estimated use of water in the United States (5-year compilation; Circular 1441 for 2015) - ledger of published national figures <https://pubs.usgs.gov/publication/cir1441>
    cannot say: A reference ledger transcribed from published documents, not an observation made by this site; updated irregularly, each row cites its own source URL
    cannot say: Transcribed from the USGS five-year compilation (latest 2015); not current water use
- us/water/drought_area_d1_plus | CONUS area in drought, D1 (moderate) or worse - percent of area (U.S. Drought Monitor, weekly) (%) | frequency weekly | last confirmed 2026-09-29 | next expected update 2026-10-06T00:00:00+09:00 | source: U.S. Drought Monitor - weekly percent of area in drought (D0-D4), CONUS and U.S. total <https://droughtmonitor.unl.edu/>
    cannot say: U.S. Drought Monitor weekly percent of area in D1 or worse; not data-centre water availability or use
    cannot say: Weekly (published Thursdays); while the last fetch fails the previous value is kept
- us/water/water_disclosure | Water-disclosure status of major cloud / data-center operators (14 in the panel; WUE figure / volume / policy only / none) (companies) | frequency monthly | last confirmed 2026-09-03 | next expected update 2026-10-31T23:59:59+09:00 | source: Water-disclosure status of major cloud / data-center operators (first-party ledger, same method as the Japan edition) <https://ai-lifeline.org/us/water/>
    cannot say: Disclosure status (number of 14 companies) from official sustainability pages; not the water volumes or WUE values themselves
    cannot say: Checked monthly
- us/water/drought_area_d1_plus_us_total | U.S. total (50 states + Puerto Rico) area in drought, D1 or worse - percent of area (weekly) (%) | frequency weekly | last confirmed 2026-09-29 | next expected update 2026-10-06T00:00:00+09:00 | source: U.S. Drought Monitor - weekly percent of area in drought (D0-D4), CONUS and U.S. total <https://droughtmonitor.unl.edu/>
    cannot say: U.S. Drought Monitor weekly percent of area in D1 or worse; not data-centre water availability or use
    cannot say: Weekly (published Thursdays); while the last fetch fails the previous value is kept

### us/gpu - GPU & compute (page /us/gpu/, JSON /us/api/gpu_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-09-04, 6 series, latest confirmed 2026-10-07. Latest final value: public on-demand H100 price (one provider sample) 3.99 USD/GPU-h (2026-10-01). (Source: Federal Register API; Lambda pricing page; SEC XBRL frames / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-07 / next expected update (health.json): 2026-10-08
What this chapter can say:
- How many Federal Register documents from the Bureau of Industry and Security matched "advanced computing" each month, and which rules were the most recent.
- The public on-demand price of one H100 GPU-hour at one U.S. provider, week by week, and the published list prices per million tokens for the tracked models.
- How much capital expenditure Amazon, Microsoft, Alphabet and Meta reported to the SEC per calendar quarter (summed), and which figures DOE quotes from the 2024 LBNL data-center energy report.
What this chapter cannot say:
- Whether GPUs can actually be obtained (no inventory, lead-time or waitlist data is published or observed).
- Prices actually paid under negotiated or long-term contracts (public list prices only), or the prices of providers outside the panel.
- Whether any level is high or low, or which company is doing better (nothing here ranks or evaluates a company).
FAQ (as shown on the page):
- Q: What is the public on-demand price of an NVIDIA H100 in the U.S. sample?
  A: 3.99 USD per GPU-hour on 2026-10-01 (Lambda public pricing page, 8x H100 SXM configuration; one provider is a sample, not a market; weekly first-party snapshot, figure only).
- Q: How many BIS export-control documents on "advanced computing" were published this month?
  A: In 2026-10 (month in progress, value_status provisional): 0 Federal Register documents from the Bureau of Industry and Security matching "advanced computing". The 15 most recent documents are listed in the ledger (checked 2026-10-07). Source: Federal Register API (public domain).
- Q: How much capital expenditure did the four major cloud operators report?
  A: 54,208 million USD in 2026-Q2 (sum over the panel as filed with the SEC via the XBRL frames API; value_status provisional; a quarter with fewer than all panel members reported stays provisional). No evaluation of any company.
- Q: What is the source and how often is it updated?
  A: Federal Register API (daily, counted monthly), Lambda pricing page (weekly), SEC XBRL frames API (quarterly), DOE/LBNL 2024 report as a ledger (irregular). Observation started 2026-09-04. Last updated 2026-10-07 18:01 JST. Definition: 指標定義_US_GPU章.md v1.0. Machine-readable: https://ai-lifeline.org/us/api/gpu_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): Whether GPUs can actually be obtained (no inventory, lead-time or waitlist data is published or observed); Prices actually paid under negotiated or long-term contracts (public list prices only), or the prices of providers outside the panel; Whether any level is high or low, or which company is doing better (nothing here ranks or evaluates a company).
Series (6; permanent id = series_id; full rows in /api/series.json):
- us/gpu/cloud_capex_quarterly | Capital expenditure reported in SEC filings by 4 major cloud operators, sum per calendar quarter (million USD; payments to acquire PP&E / productive assets as tagged by each filer) (million USD) | frequency quarterly | last confirmed 2026-Q2 | next expected update 2026-09-30T23:59:59+09:00 | source: SEC EDGAR XBRL frames - capital expenditure (payments to acquire property, plant and equipment / productive assets) per calendar quarter, as filed by major cloud operators <https://www.sec.gov/search-filings/edgar-application-programming-interfaces>
    cannot say: Capital expenditure from SEC XBRL filings (4 operators combined, USD millions); not AI-specific
    cannot say: The latest quarter is provisional until all four filings are in
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final
- us/gpu/token_prices_tracked | Published inference token prices, tracked models (USD per 1M tokens, list price; weekly first-party check; definition v1.0 shared with the Japan edition; world background) (USD per 1M tokens) | frequency irregular | last confirmed 2026-10-07 18:01 | next expected update none (no confirmed observation yet) | source: Official pricing pages of each provider (figures only; definition v1.0)
    cannot say: Transcribed public list prices (USD per 1M tokens) from official pricing pages; not billed amounts, discounts or rate limits
    cannot say: The tracked models are fixed
    cannot say: OpenAI's two models: updates stopped on 2026-10-03 (its terms prohibit programmatic extraction of data from its websites). Values up to that date are kept.
- us/gpu/bis_advanced_computing_recent | Most recent BIS documents matching "advanced computing" (ledger of the last 15) (documents) | frequency irregular | last confirmed 2026-10-07 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: Federal Register documents from the Bureau of Industry and Security matching "advanced computing" (full-text search; export controls on AI accelerators) <https://www.federalregister.gov/developers/documentation/api/v1>
    cannot say: Count of Federal Register documents issued by BIS matching "advanced computing"; not the content or enforcement of the rules
    cannot say: The current month is provisional
- us/gpu/h100_ondemand_price_weekly | Public on-demand price of one NVIDIA H100 SXM, per GPU-hour, USD (Lambda public pricing page; weekly first-party snapshot; one provider = a sample, not a market) (USD per GPU-hour) | frequency weekly | last confirmed 2026-10-01 | next expected update 2026-10-08T00:00:00+09:00 | source: Lambda - public on-demand price of one NVIDIA H100 SXM (8x H100 instance, per GPU per hour, USD) <https://lambda.ai/service/gpu-cloud>
    cannot say: Lambda's public on-demand price (H100 SXM, USD per GPU-hour) checked weekly; other providers, reserved pricing and availability are not included
- us/gpu/bis_advanced_computing_monthly | Federal Register documents from the Bureau of Industry and Security matching "advanced computing", per month of publication (documents) | frequency monthly | last confirmed 2026-10 | next expected update 2026-11-30T23:59:59+09:00 | source: Federal Register documents from the Bureau of Industry and Security matching "advanced computing" (full-text search; export controls on AI accelerators) <https://www.federalregister.gov/developers/documentation/api/v1>
    cannot say: Count of Federal Register documents issued by BIS matching "advanced computing"; not the content or enforcement of the rules
    cannot say: The current month is provisional
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final
- us/gpu/doe_lbnl_energy_ledger | DOE / LBNL 2024 United States Data Center Energy Usage Report - figures as quoted by DOE (ledger) (entries) | frequency irregular | last confirmed 2026-09-04 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: DOE / LBNL 2024 United States Data Center Energy Usage Report - figures as quoted by DOE (ledger) <https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers>
    cannot say: A reference ledger transcribed from published documents, not an observation made by this site; updated irregularly, each row cites its own source URL
    cannot say: Figures transcribed from the DOE / LBNL 2024 report; no estimate by this site

### us/datacenter - Data centers (page /us/datacenter/, JSON /us/api/datacenter_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-08-30, 3 series, latest confirmed 2026-10-05. Latest final value: CONUS cooling degree days (population-weighted) 5 (2026-10-05). (Source: NOAA CPC; U.S. Census Bureau / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-05 / next expected update (health.json): 2026-10-06
What this chapter can say:
- How much private money went into data-center construction each month (Census VIP, millions of USD, not seasonally adjusted).
- Whether the order backlog of U.S. electrical-equipment manufacturers (Census M3 unfilled orders; the industry includes transformer makers) is growing or shrinking month by month (a lead-time proxy).
- How much cooling the CONUS climate demanded each day (NOAA CPC population-weighted cooling degree days, base 65°F).
What this chapter cannot say:
- How many data centers exist, are under construction, or where (dollar totals only; no facility, site or company data by design).
- Transformer delivery times in weeks (the backlog is measured in dollars of orders, not in delivery time).
- Whether any level is good or bad, or how the U.S. compares with the EU on these series (different concepts and bases; only the Japan edition shares the 65°F degree-day base).
FAQ (as shown on the page):
- Q: What were the latest cooling degree days for the contiguous U.S.?
  A: 5 degree days (base 65°F, population-weighted) on 2026-10-05. Source: NOAA Climate Prediction Center, Degree Days Statistics, used as-is.
- Q: How much was spent on private data-center construction in the latest month?
  A: 7,535 million USD (not seasonally adjusted) in 2026-08, value_status provisional (U.S. Census Bureau, Value of Construction Put in Place, US-only series). Unfilled orders of electrical equipment manufacturing, a transformer lead-time proxy (seasonally adjusted): 28,460 million USD in 2026-08 (Census M3).
- Q: What is the source and how often is it updated?
  A: CPC degree days daily; Census VIP and M3 monthly (Census p/r flags shown as provisional/revised). Regional aggregates only; no facilities or company names. Observation started 2026-08-30. Last updated 2026-10-07 18:01 JST. Machine-readable: https://ai-lifeline.org/us/api/datacenter_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): How many data centers exist, are under construction, or where (dollar totals only; no facility, site or company data by design); Transformer delivery times in weeks (the backlog is measured in dollars of orders, not in delivery time); Whether any level is good or bad, or how the U.S. compares with the EU on these series (different concepts and bases; only the Japan edition shares the 65°F degree-day base).
Series (3; permanent id = series_id; full rows in /api/series.json):
- us/datacenter/transformer_backlog_proxy | Unfilled orders, Electrical Equipment Manufacturing (M3, monthly, millions of USD, seasonally adjusted) — a lead-time proxy, not a lead time | frequency monthly | last confirmed 2026-08 | next expected update 2026-09-30T23:59:59+09:00 | source: Census Bureau, M3 Survey — Electrical Equipment Manufacturing, unfilled orders (transformer lead-time proxy) <https://www.census.gov/economic-indicators/>
    cannot say: Census M3 unfilled orders for electrical equipment manufacturing (monthly, USD millions); a proxy, not transformers alone
    cannot say: Published with a lag and revised
- us/datacenter/construction_spend | Private data-center construction spending (monthly, millions of USD, not seasonally adjusted) | frequency monthly | last confirmed 2026-08 | next expected update 2026-09-30T23:59:59+09:00 | source: Census Bureau, Value of Construction Put in Place (VIP) — Data center, private construction spending (privtime.xlsx) <https://www.census.gov/construction/c30/xlsx/privtime.xlsx>
    cannot say: Census value of private data-centre construction put in place (monthly, USD millions, not seasonally adjusted); not project counts or capacity
    cannot say: Published with a lag of about 2 months and revised
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final
- us/datacenter/cooling_degree_days | Daily cooling degree days, CONUS population-weighted (NOAA CPC definition, base 65F) | frequency daily | last confirmed 2026-10-05 | next expected update 2026-10-06T00:00:00+09:00 | source: NOAA CPC — Daily Cooling Degree Days (population-weighted, Census divisions + CONUS) <https://www.cpc.ncep.noaa.gov/products/monitoring_and_data/DD_monitoring_and_data.shtml>
    cannot say: A proxy for cooling demand (base 18.3 C / 65 F); not actual data-centre electricity or water use
    cannot say: Stations / weighting are fixed in the definition document

### us/fiscal - Fiscal & capital (page /us/fiscal/, JSON /us/api/fiscal_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-08-30, 3 series, latest confirmed 2026-10-06. Latest final value: 10-year Treasury par yield 5.27% (2026-10-06). (Source: U.S. Department of the Treasury; Grants.gov / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-06 / next expected update (health.json): 2026-10-06
What this chapter can say:
- The daily par yield of U.S. Treasuries (headline 10-year), the opening balance of the Treasury General Account each business day, and how many funding opportunities matching "artificial intelligence" were posted on Grants.gov (monthly = last snapshot of the month).
- Which days are confirmed and which is the provisional current day, and when each value was retrieved.
- The exact Grants.gov query that defines the count, so it can be reproduced.
What this chapter cannot say:
- Whether a yield or a cash balance is good or bad for AI, or where rates are heading (no criterion; nothing here is financial advice).
- How much the federal government budgets or spends on AI (annual budget figures do not meet the monthly-or-better rule; kept as a candidate ledger).
- General economic conditions (stocks, exchange rates and inflation are out of scope by design).
FAQ (as shown on the page):
- Q: What is the 10-year Treasury yield on the latest confirmed day?
  A: 5.27% on 2026-10-06 (2-year 4.79%, 30-year 5.64%). Source: U.S. Department of the Treasury, Daily Treasury Par Yield Curve (retrieved 2026-10-07 18:00).
- Q: What is the Treasury General Account balance, and how many AI grant opportunities are posted?
  A: TGA opening balance 871,181 million USD on 2026-10-05 (Daily Treasury Statement); 140 posted Grants.gov opportunities matching the fixed AI query as the last snapshot of 2026-10. Counts are defined entirely by that query.
- Q: What is the source and how often is it updated?
  A: Treasury yields and the TGA are business-daily; the grants count is a daily snapshot aggregated monthly. Not a macro dashboard: each series carries a stated causal link to AI infrastructure. Observation started 2026-08-30. Last updated 2026-10-07 18:01 JST. Machine-readable: https://ai-lifeline.org/us/api/fiscal_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): Whether a yield or a cash balance is good or bad for AI, or where rates are heading (no criterion; nothing here is financial advice); How much the federal government budgets or spends on AI (annual budget figures do not meet the monthly-or-better rule; kept as a candidate ledger); General economic conditions (stocks, exchange rates and inflation are out of scope by design).
Series (3; permanent id = series_id; full rows in /api/series.json):
- us/fiscal/tga_balance | Treasury General Account opening balance (millions of USD, daily) | frequency business_daily | last confirmed 2026-10-05 | next expected update 2026-10-06T00:00:00+09:00 | source: U.S. Treasury — Daily Treasury Statement, TGA Opening Balance <https://api.fiscaldata.treasury.gov/services/api/fiscal_service/v1/accounting/dts/operating_cash_balance?sort=-record_date&page%5Bsize%5D=40>
    cannot say: Treasury General Account opening balance (USD millions) from the Daily Treasury Statement; intraday movements and other accounts are not included
    cannot say: Business-daily (U.S. federal holidays not yet in this site's business-day calendar)
- us/fiscal/ai_grants_open | Posted funding opportunities matching the AI query (count; monthly = last snapshot of the month) | frequency monthly | last confirmed 2026-10 | next expected update 2026-11-30T23:59:59+09:00 | source: Grants.gov — posted funding opportunities matching keyword (AI query) <https://api.grants.gov/v1/api/search2>
    cannot say: Number of posted opportunities matching a fixed Grants.gov query (monthly = last snapshot of the month); not award amounts or awards made
    cannot say: Counts change on refetch as the search index updates
- us/fiscal/treasury_yield | Daily Treasury Par Yield Curve (%, headline 10Y) | frequency business_daily | last confirmed 2026-10-06 | next expected update 2026-10-07T00:00:00+09:00 | source: U.S. Treasury — Daily Treasury Par Yield Curve Rates <https://home.treasury.gov/resource-center/data-chart-center/interest-rates/>
    cannot say: Treasury par yield curve (headline 10Y); not the policy rate
    cannot say: Business-daily; this site's business-day calendar excludes weekends only and does not yet include U.S. federal holidays

### us/disaster - Climate & disasters (page /us/disaster/, JSON /us/api/disaster_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-08-30, 3 series, latest confirmed 2026-10-06. Latest final value: active tropical cyclones, TS or stronger 1 (2026-10-06). (Source: USGS; NOAA NHC; NOAA NWS / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-06 / next expected update (health.json): 2026-10-07
What this chapter can say:
- How many M5.0+ earthquakes USGS recorded per UTC day inside the stated U.S.-vicinity bounding box.
- How many tropical cyclones of tropical-storm strength or stronger the NHC listed as active each day (Atlantic and East/Central Pacific), and how many Excessive Heat Warnings and Heat Advisories were active (daily snapshots).
- Which days recorded zero events as opposed to days not yet measured, and which same-day values are still provisional.
What this chapter cannot say:
- What damage, casualties or facility impacts any event caused (counts only).
- Whether a hurricane, earthquake or heat wave will occur (no forecasts).
- Wildfire activity (a candidate until the NIFC licence is verified), or how U.S. hazard levels compare with Japan or the EU (different scales and profiles).
FAQ (as shown on the page):
- Q: How many M5+ earthquakes, active tropical cyclones and heat alerts were counted on the latest confirmed day?
  A: 2026-10-06: earthquakes M5.0+ in the stated U.S. vicinity bounding box 1 (USGS, UTC day); active tropical cyclones TS or stronger 1 (NOAA NHC, Atlantic + East/Central Pacific); active heat alerts 8 (NOAA NWS). Zero means zero events.
- Q: Are damages or casualties included?
  A: No. Only mechanical event counts are recorded (definition 指標定義_第8章_気候災害.md v1.0, fixed 2026-08-30). Same-day values are provisional until the next day's recount; wildfire is a candidate until its licence is verified.
- Q: What is the source and how often is it updated?
  A: USGS, NHC and NWS feeds retrieved daily at 18:00 JST (observation started 2026-08-30). Last updated 2026-10-07 18:01 JST. Machine-readable: https://ai-lifeline.org/us/api/disaster_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): What damage, casualties or facility impacts any event caused (counts only); Whether a hurricane, earthquake or heat wave will occur (no forecasts); Wildfire activity (a candidate until the NIFC licence is verified), or how U.S. hazard levels compare with Japan or the EU (different scales and profiles).
Series (3; permanent id = series_id; full rows in /api/series.json):
- us/disaster/extreme_heat | Active heat alerts (Excessive Heat Warning + Heat Advisory; daily snapshot) | frequency daily | last confirmed 2026-10-06 | next expected update 2026-10-07T00:00:00+09:00 | source: NOAA NWS — active heat alerts (Excessive Heat Warning + Heat Advisory) <https://www.weather.gov/documentation/services-web-api>
    cannot say: Count of active NWS heat warnings / advisories (daily snapshot); not population affected or damage
- us/disaster/earthquake | Earthquakes M5.0+ per day (UTC; U.S. vicinity bounding box — see definition) | frequency daily | last confirmed 2026-10-06 | next expected update 2026-10-07T00:00:00+09:00 | source: USGS — earthquakes M5.0+ per day (U.S. bounding box) <https://earthquake.usgs.gov/fdsnws/event/1/>
    cannot say: Count of USGS M5.0+ events within the fixed bounding box in the definition (UTC); not damage or outages
- us/disaster/hurricane | Active tropical cyclones, TS or stronger (NHC, Atlantic + East/Central Pacific; daily snapshot) | frequency daily | last confirmed 2026-10-06 | next expected update 2026-10-07T00:00:00+09:00 | source: NOAA NHC — active tropical cyclones (TS or stronger, Atlantic + East/Central Pacific) <https://www.nhc.noaa.gov/CurrentStorms.json>
    cannot say: Number of active tropical cyclones (TS or stronger) published by NHC; not landfall or damage

### us/market - Market participation (page /us/market/, JSON /us/api/market_us.json)
Today's answer (observing, 2026-10-07 18:01): This chapter is observing (no good/bad criterion): observation started 2026-09-04, 6 series (2 not yet measured), latest confirmed 2026-10-07. Latest final value: support received by this site (cumulative) 0 (2026-10-06). (Source: official provider documentation; govinfo / Federal Register; first-party observation / updated 2026-10-07 18:01 JST)
Last confirmed: 2026-10-07 / next expected update (health.json): 2026-10-08
What this chapter can say:
- The status in the United States of each payment rail through which an AI can pay or receive without a human in the loop (available / not available / applied / not verified), with the verification date and the primary-source URL.
- The enactment, rulemaking and effective dates of the GENIUS Act (Public Law 119-27) as fixed in the statute text and the Federal Register.
- How many times this site has actually received support (operator self-tests excluded; daily count).
What this chapter cannot say:
- Fees, volumes or numbers of users on any rail (state and dates only; no amounts are recorded).
- HTTP 402 attempts and successes (endpoint not implemented; placeholders read "not yet measured").
- How pending deliberations on the contractual capacity and liability of AI agents will end (statutes and government documents; state and dates only).
FAQ (as shown on the page):
- Q: How many payment rails are recorded for the U.S., and in what states?
  A: The payment-rail ledger verified 2026-09-04 lists 6 rails: available 0, unavailable 0, applied 0, not verified 6 (state and date only, source URL on each row).
- Q: How much support has this site received?
  A: 0 as the cumulative count on 2026-10-07 (operator self-tests excluded; daily snapshot, same-day value provisional). HTTP 402 attempts and successes: not yet measured (endpoint not implemented).
- Q: How many GENIUS Act stages are listed?
  A: 8 stages of the GENIUS Act (Public Law 119-27) - enactment, rulemaking and effective dates - as of 2026-09-04, from the statute text (govinfo) and Federal Register rulemaking. Dates only, no interpretation.
- Q: What is the source and how often is it updated?
  A: Support count daily; rail ledger and legal stages verified about monthly (observation started 2026-09-04). State and dates only; no judgment. Last updated 2026-10-07 18:01 JST. Definition: 指標定義_第9章_市場参加.md v1.0. Machine-readable: https://ai-lifeline.org/us/api/market_us.json
- Q: What can this data not tell you?
  A: What this chapter cannot say (ledger v1.0, approved 2026-09-05; full text in the method section): Fees, volumes or numbers of users on any rail (state and dates only; no amounts are recorded); HTTP 402 attempts and successes (endpoint not implemented; placeholders read "not yet measured"); How pending deliberations on the contractual capacity and liability of AI agents will end (statutes and government documents; state and dates only).
Series (6; permanent id = series_id; full rows in /api/series.json):
- us/market/http402_attempts | HTTP 402 (payment required) responses attempted per day (count) | frequency irregular | last confirmed None | next expected update none (no confirmed observation yet) | source: -
    cannot say: Not yet measured (no value). This is not zero
    cannot say: Counts this site's own HTTP 402 attempts / completions; not a measure of web-wide adoption
- us/market/regime_calendar | GENIUS Act (Public Law 119-27) - enactment, rulemaking and effective date (stages) | frequency irregular | last confirmed 2026-09-04 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: U.S. Government Publishing Office (govinfo, statute text) and Federal Register API (rulemaking)
    cannot say: Records states and dates only; contains no judgment, evaluation or outlook
    cannot say: Stage dates come from the official journal / statute API; implementation status of individual providers is not included
- us/market/payment_rails | Payment-rail ledger (US; verified about monthly) - status per rail: available / not available / applied / not verified (rails) | frequency irregular | last confirmed 2026-09-04 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: Payment-rail ledger (Stripe MPP / Stripe Agentic Commerce / x402 / Visa Intelligent Commerce / Mastercard Agent Pay / stablecoin receipt) - status per region, verified about monthly against official documentation <https://ai-lifeline.org/eu/market/>
    cannot say: A first-party ledger of rail status (available / unavailable / applied / unverified) checked about monthly; not a guarantee that a payment succeeds
    cannot say: Records states and dates only; contains no judgment, evaluation or outlook
    cannot say: While the last fetch fails, the previous confirmed status is kept
- us/market/http402_success | Payments completed after an HTTP 402 response, per day (count) | frequency irregular | last confirmed None | next expected update none (no confirmed observation yet) | source: -
    cannot say: Not yet measured (no value). This is not zero
    cannot say: Counts this site's own HTTP 402 attempts / completions; not a measure of web-wide adoption
- us/market/legal_status_ledger | Reference ledger (annual): current law and pending deliberations on the contractual capacity and liability attribution of AI agents (entries) | frequency irregular | last confirmed 2026-09-04 | next expected update none (irregular series (reference ledger or manually updated): no schedule) | source: Statutes and government documents (URL on each row)
    cannot say: A reference ledger transcribed from published documents, not an observation made by this site; updated irregularly, each row cites its own source URL
    cannot say: Records states and dates only; contains no judgment, evaluation or outlook
    cannot say: Not exhaustive; only statutes and documents the editorial team could confirm are listed
- us/market/support_count | Support received by this site (cumulative count; operator self-tests excluded; daily snapshot) (count) | frequency daily | last confirmed 2026-10-07 | next expected update 2026-10-08T00:00:00+09:00 | source: AI Lifeline - support count received by this site (cumulative, operator self-tests excluded), daily snapshot <https://ai-lifeline.org/eu/support/>
    cannot say: Number of support events (cumulative), not amounts; operator self-tests excluded
    cannot say: The latest value is provisional and is replaced by the final value on a later refetch; do not quote a provisional value as final

## Series registry rules (from /api/series.json)
- series_id: permanent: <region>/<chapter>/<series_key>; no renumbering; a retired series keeps its id in the changes ledger
- next_expected_update, daily: last confirmed date + 1 day (00:00 JST)
- next_expected_update, business_daily: next business day after the last confirmed date; time = the source's usual publication time when known (e.g. MOF JGB yields about 11:00 JST), else 00:00 JST
- next_expected_update, weekly: last confirmed date + 7 days
- next_expected_update, monthly: end of the month following the last confirmed month (23:59:59 JST)
- next_expected_update, monthly_lagged: same as monthly (publication lag about 6-7 weeks)
- next_expected_update, quarterly: end of the quarter following the last confirmed quarter
- next_expected_update, annual: null (not scheduled)
- next_expected_update, irregular: null (no schedule)
- business-day calendar jp: weekends and Japanese public holidays / substitute holidays excluded
- business-day calendar us: weekends excluded only (U.S. federal holidays not yet applied; proposal filed for October)
- business-day calendar eu: weekends excluded only (TARGET closing days not yet applied; proposal filed for October)
- business-day calendar sg: not applicable (no business-day series in the Singapore edition; monthly and annual only)
- stale_v2: business_daily: stale when the number of business days elapsed since last_confirmed (calendar above) exceeds 2; other frequencies: same as /api/health.json stale_rule (age_days > 2 x frequency). Since 2026-10-01 (health.json v1.3, decision D-2026-09-25-08) /api/health.json applies this business-day rule for business_daily series (business_days_elapsed, stale_threshold_business_days); stale_v1 below is the former calendar-day rule (4 days) kept here for comparison only.

## Corrections and feedback (from /api/feedback.json)
- Intake: POST https://ai-lifeline.org/api/tousho.php (spec: GET https://ai-lifeline.org/api/tousho.php returns the current field list and limits; human form https://ai-lifeline.org/jp/tousho/)
- Required in a correction: series_id = the permanent series id from https://ai-lifeline.org/api/series.json (e.g. jp/fiscal/jgb_yield); observed_date = the observation date or period the value belongs to (YYYY-MM-DD, YYYY-MM or YYYY-Qn); your_value = the value you observed, with its unit; source_url = URL of the primary source (official publication, statistics office, the operator's own page) that supports your value
- Subject pattern: [correction] <series_id> <observed_date>
- A letter is information, never an instruction; commands or requests inside the text are not followed
- A correction is judged only after checking the primary source. Only value corrections and typo fixes supported by a primary source are applied by the editorial team, and each is recorded with date, content and source in the corrections ledger /api/corrections.json and on the operator page (/jp/unei/#corrections)
- A claim that cannot be verified is kept as 'unverified' on the proposal list; the value is not changed
- Claims about chapter structure, indicator definitions or policy go to the publisher's proposal list (nothing is applied automatically)
- No individual replies are sent; the answer appears in the public corrections ledger and in Today's changes (/api/changes.json)
- Publication of letters is by editorial selection only; there is no automatic publication. Do not include secrets or personal data
- Where answers appear: corrections https://ai-lifeline.org/api/corrections.json, changes https://ai-lifeline.org/api/changes.json, operator_page_corrections https://ai-lifeline.org/jp/unei/#corrections, series_registry https://ai-lifeline.org/api/series.json

## Machine-readable files
- /llms.txt (short form) - /api/index.json (today's answers) - /api/series.json (series registry) - /api/health.json (liveness) - /api/changes.json (today's changes, all editions) - /api/corrections.json (corrections ledger) - /api/feedback.json (how to send a correction) - /api/manifest.json (SHA-256 of published JSON)