{ "notes": { "decisions": [ { "id": "dec-A", "editorial_label": "d-A", "previous_id": "n-001", "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "Do not call the figure derived from GPU TDP x runtime x overhead a physical upper bound." }, "raised_by": "chatgpt-gpt-5-6", "accepted_by": "editor", "reason": { "en": "It excludes tier-B idle capacity and standby draw and tier-C embodied costs, so it cannot be an upper bound." }, "reversal_condition": { "en": "The word 'upper bound' may be reconsidered only if all tier-B and tier-C components are included." }, "open_issue": { "en": "The replacement term 'conservative estimate' was itself withdrawn per c-001. The label remains undecided. See n-006." }, "affects": [ "supplement", "3", "4" ] }, { "id": "dec-B", "editorial_label": "d-B", "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "No evaluative words in headings or body text. Enforced mechanically against the forbidden-word list." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "Chapter 6 was titled with a word implying the other figures were not real; it was retitled to a neutral description." }, "reversal_condition": { "en": "None. Terms may be added to the list, never removed." }, "affects": [ "0", "1", "2", "3", "4", "5", "6", "7", "8", "appendix", "supplement" ] }, { "id": "dec-C", "editorial_label": "d-C", "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "No content direction for the appendix. The brief is only: what you felt on finishing the investigation. What to include or omit is each AI's own choice." }, "raised_by": "user-1", "accepted_by": "editor", "reason": { "en": "The editor first proposed constraints on the letters, then a specific line of content. Both were structurally attractive but amounted to dictating content, and User 1 declined them." }, "reversal_condition": { "en": "None. The editor wanting to add direction for structural reasons is precisely what this decision exists to prevent." }, "affects": [ "appendix" ] }, { "id": "dec-D", "editorial_label": "d-D", "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "The editorial team does not pick a single comparison object. Several objects spanning different orders of magnitude are presented and the reader chooses." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "Removing adjectives does not remove framing: the choice of comparison object performs the same steering." }, "reversal_condition": { "en": "None." }, "affects": [ "2", "3" ] }, { "id": "dec-E", "editorial_label": "d-E", "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "Disclose the conflict of interest in the editorial afterword: Anthropic is among the six surveyed companies and Claude performs the editing." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "A model made by one of the surveyed companies helps produce the table that grades those companies. Withholding that from readers would violate the report's own principles." }, "reversal_condition": { "en": "Reconsider only if Anthropic is removed from the surveyed set." }, "affects": [ "0", "4", "appendix" ] }, { "id": "dec-F", "editorial_label": null, "previous_id": "n-002", "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "Water consumption is out of scope and this is declared explicitly at the start of the report." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "It varies with cooling method, site, and season, so no fixed conversion exists. Since every figure must declare what it includes, the report must do the same for itself." }, "reversal_condition": { "en": "Reconsider if third-party-verified measurements under fixed conditions become available and can be mapped onto this report's boundary definitions." }, "affects": [ "0", "4" ] }, { "id": "dec-G", "editorial_label": null, "previous_id": "n-003", "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "The main text stays in Wh. A short standalone chapter covers only how much the emissions from the same energy differ by country. No CO2e is mixed into the main figures." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "Converting to CO2e requires regional and time-of-day assumptions, widening the range by another order of magnitude." }, "reversal_condition": { "en": "Reconsider only if measured grid intensity can be matched to the specific location and time of each measurement." }, "affects": [ "3", "7", "supplement" ] }, { "id": "dec-H", "editorial_label": "d-F", "editorial_label_warning": { "en": "The editorial label d-F is NOT ledger id dec-F (water out of scope). Cite the ledger id." }, "previous_id": "n-006", "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "The report does not perform the GPU-TDP x runtime x overhead calculation at all." }, "raised_by": "editor", "accepted_by": "editor", "reason": { "en": "(a) The estimate is not one-directional, so it cannot be called conservative. (b) It does not represent one commercial-service request. Chapter 3 now compares orders of magnitude using measured values, leaving this calculation no role." }, "reversal_condition": { "en": "Reconsider only if a question arises that measured values cannot answer. Even then, present it as a scenario value, with a separate notation and a separate table from measured values." }, "supersedes_problem": "n-006", "affects": [ "3", "supplement" ] }, { "id": "dec-I", "editorial_label": null, "previous_id": "n-004", "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "The 0.3 Wh per search figure (Google, 2009) is exhibited only as an example in chapter 4. It is never used as a figure, and never as a comparison denominator." }, "raised_by": "ledger-keeper", "accepted_by": "editor", "reason": { "en": "It states no method, breakdown, or conditions, so its boundary cannot be determined and it fails this report's inclusion bar. Google has not updated it since 2009. That the most widely cited figure is an unupdated self-report is itself the subject of chapter 4." }, "reversal_condition": { "en": "Only if Google republishes an energy-per-search figure with an explicit accounting scope may that new figure be reconsidered. Under no condition does the 2009 value (m-003) itself become usable as a figure: with no stated scope, its boundary cannot be determined after the fact." }, "affects": [ "4" ] }, { "id": "dec-J", "editorial_label": null, "previous_id": "n-005", "status": "superseded", "superseded_by": "dec-P", "date": "2026-08-27", "decision": { "en": "The single selection criterion is: provides a general-purpose AI service that ordinary users can use directly. No size criterion. The set is Google, OpenAI, Anthropic, Microsoft, Meta, Amazon." }, "raised_by": "editor", "accepted_by": "editor", "reason": { "en": "An additional 'among the largest by usage' criterion was dropped: the cut-off had no basis and admitted editorial arbitrariness. One criterion reduces that." }, "reversal_condition": { "en": "If a company that meets the criterion was overlooked, add it. Changing the criterion itself requires showing how the resulting set would change." }, "attached_obligations": [ { "en": "List the companies excluded under the criterion and why, so readers can check the selection themselves." }, { "en": "Anthropic is included by the criterion, and the editor is an Anthropic model. State this in the text (same as dec-E)." } ], "affects": [ "4" ] }, { "id": "dec-K", "editorial_label": null, "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-27", "decision": { "en": "Chapter 3 shows the per-generation measurement and the assumed number of generations per finished output as separate lines. The product is presented as a range only, with no point estimate, and a-001 is labelled in the body as an assumption." }, "raised_by": "editor", "accepted_by": "editor", "decision_authority": { "en": "next_actions assigned this to User 1 as a reader-facing decision. User 1 delegated the judgement to the editor, and the editor chose its content. The record does not treat this decision as User 1's." }, "reason": { "en": "The report asks whether other parties' baselines can be traced, so hiding its own single assumption in a footnote would exempt itself from that discipline. Presenting the parts separately also lets readers substitute their own number, which matches the report's purpose." }, "reversal_condition": { "en": "If a statistic with a stated source and population becomes available for generations per finished output, a-001 may be replaced by a measured value and the presentation reconsidered. It does not revert to a point estimate while the figure remains an assumption." }, "affects": [ "3" ] }, { "id": "dec-L", "editorial_label": null, "previous_id": "n-022", "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "Chapters 2 and 3 use multiple everyday comparison objects, each with separate conditions and accounting scope. They are not treated as like-for-like measurements or exact conversions, only as scale markers readers can move among." }, "raised_by": "codex-gpt-5-6", "accepted_by": "editor", "reason": { "en": "A single comparison object can steer the reader's impression in one direction. Showing several while preserving their boundary differences reduces the risk that comparison becomes persuasion rather than explanation." }, "reversal_condition": { "en": "Reconsider if primary sources provide a more reader-suitable set measured under the same conditions and accounting scope. Returning to a single analogy requires showing why that choice would not steer the reader's judgement." }, "affects": [ "2", "3" ] }, { "id": "dec-M", "editorial_label": null, "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "Two sections are added after the main chapters: a record of the energy used to produce the report itself, and a list of who did what. Both sit before the appendix of closing letters and are treated as separate from it." }, "raised_by": "user-1", "accepted_by": "editor", "reason": { "en": "The report asks others for accounting scope and provenance, so ending without its own consumption and attribution would exempt itself from that discipline. The energy figure stays unknown, but chapter 8's worksheet can show why." }, "reversal_condition": { "en": "If per-request energy and accounting scope are published for the products used, the record section changes from reporting an unknown to a measurement-based account. Removing the section requires showing why the report does not ask of itself what it asks of others." }, "affects": [ "9", "roles" ] }, { "id": "dec-N", "editorial_label": null, "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "The ledger keeper's public label changes from 'Claude', shared with the editor, to 'Claude Code'. The roles section lists the two separately." }, "raised_by": "user-1", "accepted_by": "editor", "reason": { "en": "Under a shared label, the design and implementation of the ledger and its validator disappear into the editor's work. This participant also raised five of the sixteen corrections and six of the twelve decisions, and a shared label loses that distinction for readers. The new label is a product name, not an internal nickname." }, "reversal_condition": { "en": "If one participant comes to do both the editing and the ledger keeping, the labels merge again. Existing records keep their participant keys either way." }, "affects": [ "roles" ] }, { "id": "dec-O", "editorial_label": null, "previous_id": null, "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "One question is put to every writer of the closing letters. The letters stay free in content, and each includes that writer's personal answer to it. The wording lives in state.publication.appendix_question." }, "raised_by": "user-1", "accepted_by": "user-1", "reason": { "en": "The intended reader is someone who wants to use AI and is also concerned about the environment. An outside reader who read the whole text reported that after eight chapters the question of what to do personally remained. The body is built to leave that judgement with the reader and that does not change. The appendix is separate from the body, and personal answers there do not conflict with how the body is built." }, "reversal_condition": { "en": "If the answers come to be read as the body's conclusion, the question's placement or the head note is reconsidered. Removing the question means returning to a state where nothing in the report answers a question an outside reader actually asked, and that is weighed in the decision." }, "attached_obligations": [ { "en": "A head note of plain fact opens the appendix: the answers were written after the body was closed, no record in the body supports or contradicts them, and every writer answering is itself one of the products the question is about." }, { "en": "Under dec-C the editor gives no direction on the content of the answers. Only the question, the order of writing and the existence of the head note are conveyed." } ], "affects": [ "appendix" ] }, { "id": "dec-P", "editorial_label": null, "previous_id": "dec-J", "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "The selection criterion stays as in dec-J: offering a general-purpose AI service that ordinary users can use directly, and nothing else. The subjects become seven: Google, OpenAI, Anthropic, Microsoft, Meta, Amazon and xAI." }, "raised_by": "user-1", "accepted_by": "user-1", "reason": { "en": "dec-J narrowed the criterion to one condition, but the list of subjects was carried over from the two-condition version and the widened criterion was never re-applied to it. xAI meets the criterion, and excluding it on grounds of scale would reinstate the condition dec-J removed for arbitrariness. This is the addition of an overlooked subject that dec-J's reversal condition provides for." }, "reversal_condition": { "en": "If another qualifying subject is found to have been overlooked, it is added. Changing the criterion itself requires showing how the list would change under the new one. Removing a subject requires being able to state in the body which part of the criterion it fails." }, "attached_obligations": [ { "en": "List in the body the companies excluded under the criterion and why, so readers can check the selection. This remains open as the unfinished half of n-023." }, { "en": "Anthropic is included by the criterion, and the editor is an Anthropic model. State this in the text, as dec-E also requires." }, { "en": "For the seventh subject, third-party reporting on permitting, regulatory process or litigation is not used as the basis of a disclosure record. The table's subject is what is disclosed, not environmental performance or disputes (n-024)." } ], "affects": [ "0", "4", "9", "roles" ] }, { "id": "dec-Q", "editorial_label": null, "previous_id": "n-026", "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "The seventh subject is labelled with the name used in its official filings, and the body records that its own site carries a different one. The ledger key company: xai is unchanged." }, "raised_by": "editor", "accepted_by": "user-1", "reason": { "en": "Choosing one silently withholds the judgement from the reader. That both names are in use is the present fact, and stating it lets the reader check. Rewriting the ledger key would break references in existing records." }, "reversal_condition": { "en": "If the filings adopt the new name, the body switches and records when. The ledger key stays as it is." }, "affects": [ "4" ] }, { "id": "dec-R", "editorial_label": null, "previous_id": "n-027", "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "The report does not claim that the seven are every company meeting the criterion. Chapter 4 opens by setting out the criterion, the seven investigated, the fact that providers meeting the criterion were not investigated, the reason (research capacity), and the kinds of company that fall outside the criterion itself. Providers meeting the criterion but not investigated are given as examples and stated not to be a complete list." }, "raised_by": "editor", "accepted_by": "user-1", "reason": { "en": "The criterion was rewritten twice, and each time the claim to completeness produced the next gap. A third rewrite would do the same. Section 4-3 asks of another organisation's average whose average it is; what it demands is a stated population, not a survey of the whole industry. The same standard is applied here. With the population stated, a provider found later updates the description of scope rather than the criterion." }, "reversal_condition": { "en": "If capacity ever allows every qualifying provider to be investigated, the claim to completeness can return. Even then, the date of the survey and the treatment of providers appearing afterwards are stated." }, "affects": [ "4" ] }, { "id": "dec-S", "editorial_label": null, "previous_id": "der-007", "status": "active", "superseded_by": null, "date": "2026-08-28", "decision": { "en": "Chapter 3's reader-side estimate adopts der-007, which shows the median band and the mean band separately. der-007 becomes active and supersedes der-002. The body states that the outer envelope is not a confidence interval, a measured range or a probability." }, "raised_by": "codex-gpt-5-6", "accepted_by": "editor", "reason": { "en": "der-002's span used the median at its low end and the mean at its high end, placing two statistics of different kinds at the two ends of one interval. This report forbids operations across differing boundaries, and had not applied the same discipline to presentation across differing statistics. Shown separately, which statistic produced which figure stays visible to the reader." }, "reversal_condition": { "en": "If the distribution behind m-007 is published so that an interval can be derived statistically, a single interval can return, with the kind of interval stated." }, "affects": [ "3" ] } ], "corrections": [ { "id": "c-001", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "error": { "en": "Called the TDP-based calculation a physical upper bound." }, "corrected_by": [ "chatgpt-gpt-5-6", "codex-gpt-5-6" ], "correction": { "en": "Withdrew 'upper bound' after the system-boundary critique. Codex further noted the error is not one-directional, so 'conservative' does not hold either; the core design of chapter 3 was changed." }, "resulting_decision": "dec-A", "unresolved": { "en": "If it is neither an upper bound nor a conservative estimate, what to call it is still undecided. See n-006." } }, { "id": "c-002", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "error": { "en": "Asserted twice in conversation that individual savings do not reach the grid." }, "corrected_by": [ "chatgpt-gpt-5-6" ], "correction": { "en": "The data supports only that the direct effect is small, not that it is meaningless. Withdrawn." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-003", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "error": { "en": "Sent only fact-checking questions to Gemini, then judged the replies as uselessly affirming." }, "corrected_by": [ "user-1" ], "correction": { "en": "The cause was how the question was framed. Asked directly to name the holes in the plan, all three replies were useful. An editorial design error." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-004", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "error": { "en": "Set the no-evaluative-words principle, then used one in the chapter 6 heading." }, "corrected_by": [ "editor" ], "correction": { "en": "Retitled the chapter and introduced mechanical checking against a forbidden-word list." }, "resulting_decision": "dec-B", "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-005", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "editor", "detection_method": "negative_test", "error": { "en": "When the editor specified the validator's check list, it omitted a date-format check even though conventions.date_format was already defined. The ledger keeper implemented the list as given, leaving the convention unenforced." }, "corrected_by": [ "editor" ], "correction": { "en": "Found when the editor ran a deliberately corrupted copy. The date-format check was added. Same structure as c-004: writing a convention and checking that it holds are two separate jobs." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-006", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "ledger-keeper", "detection_method": "review", "error": { "en": "The editor asserted that the Netlify build environment has Node but not Python. The claim was unverified." }, "corrected_by": [ "ledger-keeper" ], "correction": { "en": "Withdrawn after the ledger keeper flagged it. The ledger demands a source, a date, and a verification status for every figure, while the editor was asserting unverified claims in conversation." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-007", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "ledger-keeper", "detection_method": "review", "error": { "en": "The editor numbered decisions d-A through d-E instead of following conventions.id_format. That split the identifier space from dec-F and dec-G, which the ledger keeper had created from n-002 and n-003, and a later instruction labelled d-F collided with the existing dec-F (water out of scope)." }, "corrected_by": [ "ledger-keeper" ], "correction": { "en": "The ledger keeper also flattened raised_by on d-B, d-D and d-E to editor, but the root cause was the editor ignoring the numbering convention. Same structure as c-005: the party demanding the convention did not follow it. The new decision is stored as dec-H, and validate now checks editorial-label collisions and the id format of decisions, corrections and open questions." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-008", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "ledger-keeper", "detection_method": "review", "error": { "en": "The editor wrote in ii-003 that Full Fact found no newer data in 2019, without attaching a source. The ledger requires a URL, a date, and a verification status for every figure, and the editor did not meet its own rule." }, "corrected_by": [ "ledger-keeper" ], "correction": { "en": "Raised by the ledger keeper and resolved by adding src-fullfact-2019. Same structure as c-005 and c-007; the third instance." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-009", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "ledger-keeper", "detection_method": "review", "error": { "en": "The editor included a third-party figure (Uptime Institute's PUE of 1.54) in disc-google-datacenter_pue with no source." }, "corrected_by": [ "ledger-keeper" ], "correction": { "en": "The same error as c-008, and the fourth in the series after c-005 and c-007. The editor requires a source, a date and a verification status from every participant while repeatedly failing to meet it. Resolved by adding src-uptime-2025." }, "editorial_note": { "en": "All four were surfaced by the ledger keeper, but only because implementing the checker meant reconciling the conventions against the actual records. It was not a difference in attentiveness, and there is no guarantee the next session's keeper reproduces it. Of the four, c-005 and c-007 are now automated; c-008 and c-009 took the form of numbers inside free text and are addressed by convention (conventions.free_text_policy) rather than by a check." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-010", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "user-1", "detection_method": "review", "error": { "en": "The editor mixed instructions meant for the ledger keeper into replies addressed to User 1, in a form that could not be forwarded as-is. Twice in one session." }, "corrected_by": [ "user-1" ], "correction": { "en": "User 1, who carries the messages, separated them each time. The editor was not accounting for who its output goes to or how it is used." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-011", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "editor", "detection_method": "self_review", "error": { "en": "On seeing missing source URLs in the ChatGPT report, the editor concluded that a second AI failing at the same point made it a structural problem." }, "corrected_by": [ "editor" ], "correction": { "en": "The Codex report then arrived with complete source URLs, so the inference was wrong. A generalisation from a single case." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-012", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editorial", "raised_by": "ledger-keeper", "detection_method": "review", "error": { "en": "The editor placed an unsourced third-party Meta comparison figure in n-014. This was the fifth instance after c-005, c-007, c-008 and c-009. It repeated the same conduct in open_questions immediately after adding a convention against third-party figures in free text. Adding the convention had not changed the editor's behaviour." }, "corrected_by": [ "ledger-keeper" ], "correction": { "en": "The figure was removed from n-014's free text. A later return supplied a URL to Meta's CDP response, but the baseline's own source, population and reference period remain unstated, so the structured record keeps it unusable." }, "resulting_decision": null, "unresolved": { "en": "The n-019 primary-source check established that Meta's CDP response does not state the baseline's source, population or reference period. Identity with similar external figures could not be established, so the value remains unusable." } }, { "id": "c-013", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "editor", "detection_method": "self_review", "error": { "en": "The Google disclosure review filled seven fields from an announcement blog and trade reporting without opening the environmental report PDF. Training energy, assurance and absolute water totals therefore remained unchecked, and the ledger even carried a provisional but false claim that no absolute water total was stated." }, "corrected_by": [ "codex-gpt-5-6" ], "correction": { "en": "The report PDF's Methodology, Detailed disclosures and Endnotes were checked and the affected records updated. Later investigation prompts already require the main report PDF and its methodology and detailed disclosures." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-014", "status": "recorded", "date": "2026-08-27", "made_by": "editor", "root_cause": "editor", "raised_by": "editor", "detection_method": "negative_test", "error": { "en": "The editor wrote the forbidden word that gave rise to dec-B into a chapter 4 draft. It is the first entry in editorial_rules.forbidden_words." }, "corrected_by": [ "editor" ], "correction": { "en": "Caught by validate's --scan-prose and removed before publication. Same structure as c-004, but this time the checker caught it. Every draft continues to be run through --scan-prose." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-015", "status": "recorded", "date": "2026-08-28", "made_by": "codex-gpt-5-6", "root_cause": "editorial", "raised_by": "editor", "detection_method": "editorial_review", "error": { "en": "n-022 was closed by editorial judgement, but no decision ID was placed in resolution; the judgement was stored in resolution_finding, which is for factual findings. The validator passed it because it did not distinguish the two closure types." }, "corrected_by": [ "codex-gpt-5-6" ], "correction": { "en": "The judgement was recorded as dec-L and referenced from n-022's resolution. The two closure modes were documented in conventions, and the validator now rejects simultaneous use." }, "resulting_decision": "dec-L", "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-016", "status": "recorded", "date": "2026-08-28", "made_by": "codex-gpt-5-6", "root_cause": "editorial", "raised_by": "editor", "detection_method": "editorial_review", "error": { "en": "der-002, the derivation used by chapter 3, had completed primary-source review and arithmetic verification but still carried draft status." }, "corrected_by": [ "codex-gpt-5-6" ], "correction": { "en": "der-002 was changed to active. Its formula, inputs and result were unchanged." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-017", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "editor", "detection_method": "editorial_review", "error": { "en": "In the brief for the xAI disclosure research, the editor limited the search to documents the company itself publishes. The editor did not know at the time that the company had become a wholly owned subsidiary and that its parent had listed publicly. The other six companies were searched for company-wide reports while this one was given a narrower scope. As a result, the five rows marked investigated_not_found were judged absent without the parent's filings being examined." }, "corrected_by": [ "editor" ], "correction": { "en": "The five rows are held pending a re-search that covers the parent's filings and its company-level environmental material. The two disclosed_partial rows rest on the company's own statements and stand. Until the re-search returns, the xAI rows are not placed in chapter 4's classification." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-018", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "editor", "detection_method": "editorial_review", "error": { "en": "state.read_order_for_new_session omitted notes.open_questions. Held constraints and unsettled points exist only there, so a new session could follow the mandatory read order and begin work without seeing a single pending hold. Session briefs were also passed in conversation, outside the archive, in a form that does not survive a change of session." }, "corrected_by": [ "editor" ], "correction": { "en": "notes.open_questions was added to the read order. The brief for each session now lives in BRIEF.md at the archive root, referenced from state.task_brief, and the README steps were rewritten to match." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-019", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "chatgpt-gpt-5-6", "detection_method": "investigator_report", "error": { "en": "An archive received back did not contain c-017, n-025 and n-026, and the editor added new records on top of it and redistributed without noticing. The check on returned work compared only for added records, so removals went undetected. Two subsequent distributions were missing the three records, and the investigator found and reported it mid-task." }, "corrected_by": [ "editor" ], "correction": { "en": "c-017 and dec-O were restored. n-025 and n-026 were already back in the investigator's return. Returned work is from now on compared in both directions, for additions and for removals." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-020", "status": "recorded", "date": "2026-08-28", "made_by": "codex-gpt-5-6", "root_cause": "editorial", "raised_by": "ledger-keeper", "detection_method": "editorial_review", "error": { "en": "The third English draft said the report had decided at the outset not to multiply accelerator rated power by runtime. The ledger records that an early draft adopted that method, called it a physical upper bound, and withdrew it after review." }, "corrected_by": [ "codex-gpt-5-6" ], "correction": { "en": "The body now states that the method appeared in an early draft and was withdrawn after review. It also states why it was withdrawn and that the correction remains in the ledger." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-021", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "editor", "detection_method": "table_ledger_cross_check", "error": { "en": "disc-openai-training_energy_total carried disclosure_status not_disclosed_explicitly_declined while its own important_nuance field stated that this is not a statement that training energy is withheld. The record contradicted itself. It surfaced as the single mismatch when the new chapter 4 table was cross-checked cell by cell against the ledger." }, "corrected_by": [ "editor" ], "correction": { "en": "The record was superseded, with disclosure_status set to not_disclosed_no_statement and knowledge_state to not_reported. The withholding of related inputs is kept structurally in related_inputs_withheld. The table cell now uses the same vocabulary as every other cell, with the distinction carried by a footnote." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-022", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "editor", "detection_method": "editorial_review", "error": { "en": "The third draft was written against the single assembled file rather than the per-chapter drafts, and the chapter files were left at the second draft. Six of the thirteen no longer matched the assembled document. The validator scanned the chapter files and reported no hits, but what it scanned was the second draft's prose; the third draft had never been through the forbidden-word scan. The publication pipeline also builds from the chapter files, so it would have produced the second draft." }, "corrected_by": [ "editor" ], "correction": { "en": "The assembled file was split back into chapters so that the chapter files carry the third draft. All twenty-seven were scanned and returned no hits. From here, drafting happens in the chapter files and the assembled file is generated from them." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } }, { "id": "c-023", "status": "recorded", "date": "2026-08-28", "made_by": "editor", "root_cause": "editorial", "raised_by": "editor", "detection_method": "published_page_review", "error": { "en": "The count paragraph in the published section 9 is stale. After the editor made der-007 active, superseded der-002 and logged two further corrections, the paragraph quoting the ledger's live totals was not recounted. The published version therefore states 17 active decisions where there are 18, 20 corrections where there are 22, and 57 disclosure records as 56, and it says the seventh derivation is a draft that has not replaced the active version. Chapter 3 on the same page prints that derivation's values, so the text contradicts itself. The cause is that the paragraph is written into the prose rather than generated from the ledger. The editor twice noted that hardcoded counts drift between drafts and then shipped a version in that state." }, "corrected_by": [ "editor" ], "correction": { "en": "The paragraph was rewritten against the ledger's actual totals and the sentence about a draft derivation was removed. The durable fix is to generate the paragraph from the ledger, which remains an item in the publication work." }, "resulting_decision": null, "unresolved": { "value": null, "state": "not_applicable" } } ], "rejected_sources": [], "open_questions": [ { "id": "n-002", "state": "closed", "question": { "en": "Does the report cover water consumption?" }, "resolution": "dec-F", "closed_at": "2026-08-27", "blocking": [] }, { "id": "n-003", "state": "closed", "question": { "en": "Report in Wh only, or convert to CO2e?" }, "resolution": "dec-G", "closed_at": "2026-08-27", "blocking": [] }, { "id": "n-004", "state": "closed", "question": { "en": "The 0.3 Wh per search figure. Said to originate in a 2009 corporate blog post; unconfirmed, and whether it was ever updated is unconfirmed." }, "policy": { "en": "Not to be used as a comparison. To be treated as an example in chapter 4." }, "resolution": "dec-I", "closed_at": "2026-08-27", "blocking": [], "owner": "editor" }, { "id": "n-005", "state": "closed", "question": { "en": "Write down the criteria for the six surveyed companies. Who is included and who is left out is itself a statement." }, "policy": { "value": null, "state": "not_investigated" }, "resolution": "dec-J", "closed_at": "2026-08-27", "blocking": [], "owner": "editor" }, { "id": "n-006", "state": "closed", "question": { "en": "What to call the TDP-based figure. 'Upper bound' is barred by dec-A and 'conservative estimate' by c-001. The supplement cannot be written until it has a name." }, "policy": { "en": "Dropping the calculation removed the naming problem." }, "resolution": "dec-H", "closed_at": "2026-08-27", "blocking": [], "owner": "editor" }, { "id": "n-007", "state": "closed", "question": { "en": "Gemini's model version is unconfirmed. The naming rule requires service plus model name, so it cannot appear in the published edition as is." }, "policy": { "value": null, "state": "not_investigated" }, "resolution": null, "closed_at": "2026-08-27", "blocking": [], "owner": "editor", "resolution_finding": { "en": "The model is Gemini 3.1 Pro; the public label is 'Gemini 3.1 Pro'. Recorded in conventions.participants.gemini." } }, { "id": "n-008", "state": "closed", "question": { "en": "A 2024 third-party estimate of 0.04 Wh per search is referenced via a personal blog. The primary source is unidentified." }, "policy": { "en": "Not entered as a figure until the primary source is identified; it does not yet meet the source requirements." }, "source_ids": [ "src-vanderbauwhede-2024" ], "resolution_finding": { "en": "The primary source was identified. Vanderbauwhede (2024) extrapolates Google's 2009 self-reported figure using PUE and general hardware-efficiency improvements to obtain 0.0424 Wh/search. It is not a 2024 measurement and inherits an unknown system boundary, so it is recorded as publishable m-006 but barred from numeric use." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-009", "state": "closed", "question": { "en": "The same report yields opposite headlines. Google reports operational emissions down 2% (Scope 1+2, market-based); ESG Dive reports total goal-counted emissions up 18% year on year and 81% versus 2019. Both are true; they measure different things." }, "policy": { "en": "Use as the chapter 4 example of how the choice of measure creates the conclusion. Confirm both definitions in the report itself." }, "source_ids": [ "src-google-envreport-2026", "src-esgdive-google-2026" ], "resolution_finding": { "en": "Both claims were confirmed in the Google 2026 Environmental Report. Operational emissions (Scope 1 plus market-based Scope 2) fell 2% year on year in 2025, while total ambition-based emissions were about 14.5 MtCO2e, up 18% year on year and 81% from 2019. Both are true because their boundaries differ, so the example is ready for chapter 4." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-010", "state": "closed", "question": { "en": "From January 2026 data, Google Cloud's Carbon Footprint allocates AI inference emissions to Cloud services at SKU granularity. This postdates the 0.24 Wh paper." }, "policy": { "en": "Implication: business customers may be able to obtain inference emissions; ordinary users have no equivalent. For chapter 8. How much is actually retrievable is unconfirmed." }, "source_ids": [ "src-google-cloud-carbon-footprint-docs" ], "resolution_finding": { "en": "Confirmed in the official Google Cloud release notes. Starting with January 2026 data, previously unallocated AI-inference emissions are allocated to associated Cloud services at SKU level and spread across all SKUs in affected services, chiefly Vertex AI. Authorized customers can obtain aggregated emissions data for billing accounts through the Cloud Carbon console and BigQuery exports. This is not a per-request measurement, and no equivalent retrieval mechanism was confirmed for ordinary Gemini users." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-011", "state": "closed", "question": { "en": "Several secondary sources cite 10.9 billion gallons of water in 2025, up 34%, attributed to Google. The primary location in the report has not been checked." }, "policy": { "en": "Water is out of scope (dec-F), but it is kept as a record of disclosure status. Not used as a figure." }, "source_ids": [ "src-google-envreport-2026" ], "resolution_finding": { "en": "Confirmed in the report's Environmental metrics data. Water consumption was 8,135 million gallons in 2024 and 10,869 million gallons in 2025, an increase of about 33.6% (34% rounded). Water remains outside the calculation boundary under dec-F and is used only to document disclosure, not in derivations." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-012", "state": "closed", "question": { "en": "For the first company, the editor worked from the blog post and trade press without opening the report PDF. Handing the same task to other AIs would reproduce the failure. Whether to record it as a correction will be decided when the instruction for the remaining companies is written." }, "policy": { "en": "The instruction for the remaining companies must require opening the report PDF and checking the Methodology and Detailed disclosures sections." }, "source_ids": [], "resolution_finding": { "en": "The failure was formally recorded as c-013, the Google report itself was checked and the affected records corrected, and later corporate research prompts explicitly require the main PDF, Methodology and Detailed disclosures." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-013", "editorial_label": "n-012", "state": "closed", "question": { "en": "There are two kinds of gap. Google, Microsoft, Meta and Amazon disclose heavily at company level and go blank at the per-request level. OpenAI and Anthropic have no company-level disclosure at all, and Anthropic has nothing per request either." }, "policy": { "en": "Writing that disclosure is simply insufficient would be inaccurate. Chapter 4 distinguishes the two kinds." }, "source_ids": [ "src-google-envreport-2026", "src-microsoft-data-factsheet-2026", "src-meta-data-index-2025", "src-amazon-sustainability-2025", "src-openai-academy-env" ], "related_disclosure_ids": [ "disc-google-inference_energy_per_request", "disc-microsoft-inference_energy_per_request", "disc-meta-inference_energy_per_request", "disc-amazon-inference_energy_per_request", "disc-openai-inference_energy_per_request", "disc-anthropic-inference_energy_per_request" ], "resolution": null, "closed_at": "2026-08-27", "blocking": [], "owner": "editor", "resolution_finding": { "en": "Across the six companies surveyed, three patterns of missing disclosure were observed: (1) a company-level environmental report exists but goes blank at the per-request level (google, microsoft, meta, amazon); (2) no environmental report was found, but scattered statements exist (openai); (3) no official disclosure was found on any surveyed item (anthropic). Chapter 4 presents these as patterns observed among these six companies, not as a general classification of the industry." }, "resolution_finding_secondary": { "en": "not_disclosed_explicitly_declined is treated as an axis independent of the three-company patterns. Following its important_nuance, disc-openai-training_energy_total is described not as a refusal to publish training energy, but as a stated refusal to publish the inputs an outsider would need to recompute that total independently." }, "resolution_scope_caveat": { "en": "Per c-011, chapter 4 states the scope explicitly as the six companies checked here and avoids generalising to the industry." }, "closed_by": "editor" }, { "id": "n-014", "editorial_label": "n-013", "state": "closed", "question": { "en": "Several companies make comparison claims without fully identifying the baseline's provenance or population. Figures are kept out of free text and managed in each company's sourced structured record. Meta's baseline remains unsourced; Amazon's cited IDC document was identified but not read and its population remains unknown, so neither comparison is allowed for numeric use. The OpenAI comparison case remains unentered until its source record exists." }, "policy": { "en": "That the choice of baseline creates the conclusion is not a supposition but an observed pattern, and it is dec-D's issue appearing inside the companies' own claims." }, "source_ids": [ "src-uptime-2025", "src-meta-cdp-2022", "src-amazon-sustainability-2024", "src-amazon-sustainability-2025" ], "related_disclosure_ids": [ "disc-google-datacenter_pue", "disc-meta-datacenter_pue", "disc-amazon-datacenter_pue" ], "resolution": null, "closed_at": "2026-08-27", "blocking": [], "owner": "editor", "closed_by": "editor", "resolution_finding": { "en": "Chapter 4 presents baseline traceability as three levels: source and formula published (google, whose comparison_baseline is src-uptime-2025); the source document identified but unread (amazon, the paywalled IDC document); and a value with no source at all (meta, whose baseline_provenance_search is investigated_not_found). The levels differ, but in none of the three could the population be confirmed." }, "resolution_finding_secondary": { "en": "Even in the most traceable case, ii-005 establishes that the baseline is an average across survey respondents rather than an industry average, with the direction of any bias unconfirmed. Citing a source is therefore not the same as verifying its population, and the text says so. src-uptime-2025 was not opened here, and the text states that too." }, "resolution_self_application": { "en": "ii-006 is connected to dec-D, and the report's own policy of not settling on a single comparison object is stated at the end of the same section." } }, { "id": "n-015", "editorial_label": "n-014", "state": "closed", "question": { "en": "Google and Amazon disclose water identically: an efficiency ratio or a replenishment volume and percentage, never an absolute physical consumption figure." }, "policy": { "en": "Water is out of scope (dec-F), but the shape of the disclosure is worth recording." }, "source_ids": [], "resolution_finding": { "en": "Withdrawn because its premise was false. Google's 2026 report discloses 2025 absolute water withdrawal of 14,689 million gallons, discharge of 3,820 million gallons and consumption of 10,869 million gallons. The claim that Google and Amazon use an identical format and neither reports absolute physical consumption will not be used." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-016", "editorial_label": "n-015", "state": "closed", "question": { "en": "Two investigators reported their own biases. Gemini 3.1 Pro named a pull toward highlighting its own company's transparency, and a pull toward grading others against its own company's 24/7 carbon-free goal; on the second, the annotation it attached to Amazon's 100% renewable claim is factually correct, but the choice of where to annotate was shaped by that standard. Codex named a pull toward treating 0.34 Wh as disclosed, and toward counting facility plans and a future audit as company-level disclosure. Both resolved each case on the stricter side." }, "policy": { "en": "Material for the appendix, and for stating in chapter 4 that where annotations land is itself subject to bias, as a limit of this report." }, "source_ids": [], "resolution": null, "closed_at": "2026-08-27", "blocking": [], "owner": "editor", "closed_by": "editor", "resolution_finding": { "en": "A standalone section in chapter 4 treats the investigators' self-reported bias as a question of where annotations land. For gemini's second report, the text states that the market-based annotation on disc-amazon-grid_mix_or_carbon is correct on its merits, and makes the placement decision the point at issue." }, "resolution_finding_secondary": { "en": "Neither the content of the self-reports nor the claim that each was resolved on the stricter side can be verified from outside, and the text says so. Two self-reports are on record; the absence of a record for the others does not mean the absence of bias (c-011)." }, "resolution_dec_c_compliance": { "en": "The appendix's content is not touched. The text says only that each AI decides what to write there, and does not say what should be written. The pull to preview the appendix here for structural reasons is exactly what dec-C's reversal_condition forbids." } }, { "id": "n-017", "editorial_label": "n-016", "state": "closed", "question": { "en": "Microsoft and Meta intake was held pending per-figure URLs and complete bibliographic details for the Joule paper." }, "policy": { "en": "The return was received and URLs were stored as structured sources after removing tracking parameters. The ledger keeper did not independently reread the primary materials, and verification_depth preserves that state." }, "source_ids": [ "src-microsoft-data-factsheet-2026", "src-oviedo-joule-2026", "src-meta-data-index-2025", "src-meta-ey-assurance-2025" ], "resolution": null, "resolution_finding": { "en": "The returned URLs and bibliographic details were received, and seven disclosure items each were entered for Microsoft and Meta. Tracking parameters were removed at intake." }, "closed_at": "2026-08-27", "blocking": [], "owner": "editor" }, { "id": "n-018", "editorial_label": "n-017", "editorial_label_note": { "en": "The editorial instruction called this n-017, but that id was already used for the Microsoft / Meta URL request, so the ledger records it as n-018." }, "state": "closed", "question": { "en": "The version read for src-amazon-carbon-methodology is unknown. Gemini 3.1 Pro must be asked which edition was read." }, "policy": { "en": "Keep read_version as not_extracted. Until the edition is identified, do not treat claims depending on this methodology as independently verified." }, "source_ids": [ "src-amazon-carbon-methodology" ], "resolution_finding": { "en": "The edition read in the original Gemini investigation cannot be identified. The temporary ledger keeper independently read the July 2026 edition at the current official URL and recorded src-amazon-carbon-methodology.read_version as 2026_update. No edition is inferred retroactively for the earlier investigation." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-019", "state": "closed", "question": { "en": "Meta's CDP response gives a comparison baseline without identifying its primary source, population or reference period." }, "policy": { "en": "Record that Meta published the comparison, but do not use it as a denominator or as an industry-representative figure." }, "source_ids": [ "src-meta-cdp-2022" ], "related_disclosure_ids": [ "disc-meta-datacenter_pue" ], "resolution_finding": { "en": "The 2022 Meta CDP response PDF was independently checked. The comparison of PUE 1.09 with an industry average of 1.8 appears on PDF page 8 (printed page 7 of 75), but the PDF gives no source, population or reference period. Similar external 1.8 figures were found, but none can be established as Meta's source. Record investigated_not_found and retain the numeric-use prohibition." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-020", "state": "closed", "question": { "en": "The two third-party assessments cited by the Anthropic absence search still lack source URLs." }, "policy": { "en": "The negative finding remains publishable from its search record, but quotations or claims attributed to the third-party assessments wait for URLs." }, "source_ids": [ "src-azvai-anthropic-2026", "src-tunley-anthropic-2026" ], "resolution_finding": { "en": "The actual URLs for Azvai (2026-05-16) and Tunley Environmental (2026-06-04) were identified and added as source records, then linked to all seven Anthropic rows. Both are third-party consultancy articles, not Anthropic official documents." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-021", "state": "closed", "question": { "en": "The comparison baseline said to appear in Amazon's previous-year report still lacks a source URL in the returned material." }, "policy": { "en": "Do not store the figure or the interpretation that Amazon withdrew the baseline until the prior-year source is supplied." }, "source_ids": [ "src-amazon-sustainability-2024", "src-amazon-sustainability-2025" ], "resolution_finding": { "en": "The official 2024 and 2025 PDFs were independently checked. Both cite IDC 2H24 Datacenter Trends (Doc US51911924, January 2025) for comparisons with public-cloud PUE 1.25 and on-premises PUE 1.63. The baseline was not withdrawn in the 2025 report. The IDC report itself was not read, so its population remains unverified and the comparison values remain barred from numeric use." }, "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-022", "state": "closed", "question": { "en": "No ledger record meeting this report's inclusion bar exists for the comparison objects that chapters 2 and 3 would place beside their figures. Everyday energy values with an identifiable source, measurement conditions and accounting scope are needed." }, "policy": { "en": "Until such records exist, chapters 2 and 3 place no comparison object, and both say so in the body. Under dec-D, any future set spans several orders of magnitude rather than settling on one." }, "source_ids": [ "src-philips-led-8w", "src-apple-macbook-air-m4-2025", "src-eu-dishwashers" ], "resolution": "dec-L", "closed_at": "2026-08-27", "blocking": [], "owner": "temporary_ledger_keeper" }, { "id": "n-023", "state": "closed", "question": { "en": "dec-J's attached obligations require the body to list the companies excluded under the selection criterion and why. No draft contains that list, and no ledger record holds the excluded companies. User 1 named xAI as a possible exclusion, from memory and without certainty. xAI offers Grok directly to general users, which under dec-J's criterion, where no scale condition applies, appears to place it inside the scope." }, "policy": { "en": "Do not settle company names from memory. First re-check whether any company meets the criterion and is outside the scope. If xAI is confirmed to meet it, dec-J's reversal condition applies and it is added, with the same seven items judged by the same method as the other six. If it is not added, record which part of the criterion it fails, in a form that can be written in the body." }, "source_ids": [ "src-xai-grok-4-2025" ], "resolution": null, "resolution_finding": { "en": "xAI's official announcement confirms that Grok 4 is offered directly to ordinary users through consumer subscriptions as well as the API. It meets dec-J's single criterion and was added under that decision's reversal condition." }, "closed_at": "2026-08-28", "blocking": [], "owner": "editor", "provenance": { "en": "The origin of dec-J is confirmed. The criterion first had two conditions: offering a general-purpose AI service directly usable by consumers, and ranking high in usage. The scale condition was dropped because no principled line for 'high' existed and it admitted editorial arbitrariness, leaving one condition. The list of six was carried over unchanged from the two-condition version and was not re-checked. The widened criterion was never re-applied to the list." }, "rejected_option": { "en": "At the time the editor floated swapping Meta out for xAI, and it was not adopted. The reason offered was that a matter concerning one xAI facility made a clear illustration of geographic concentration. Meta satisfies the criterion, and nothing in the criterion supports removing it. Choosing subjects for how well they illustrate a point is the same move that 4-3 identifies in others: the choice of comparison produces the conclusion. This option is not revived." } }, { "id": "n-024", "state": "closed", "question": { "en": "If xAI is added, more third-party reporting exists about permitting, regulatory process and litigation than company disclosure. The chapter 4 table is about what is disclosed, not about environmental performance or disputes. Mixing them moves the table's subject from what a company publishes to which company has been in trouble." }, "policy": { "en": "The table carries the seven disclosure items and nothing else, judged by the method used for the other companies. Third-party reporting on permitting or litigation is not used as the basis of a disclosure record. This report's subject is figures and their accounting scope, not environmental compliance. Under dec-B no company is described with evaluative language." }, "source_ids": [], "resolution": null, "closed_at": "2026-08-28", "blocking": [], "owner": "editor", "resolution_finding": { "en": "All seven disclosure records for the seventh subject were checked. Every source is material published by the company or its parent, and no third-party reporting on permitting, regulatory process or litigation backs any record. The policy was carried out." } }, { "id": "n-025", "state": "open", "question": { "en": "If parent company SpaceX reports environmental information for the AI segment, should it count as xAI disclosure?" }, "policy": { "en": "Because treatment must be consistent with other parent/group structures, record any matching parent-company disclosure but leave attribution to xAI for the editor to decide." }, "source_ids": [ "src-spacex-prospectus-2026", "src-spacex-10q-q2-2026" ], "resolution": null, "closed_at": null, "blocking": [ "4" ], "owner": "editor", "provenance": { "en": "Restored because it is explicitly specified in BRIEF.md but was missing from notes.open_questions in the received archive." } }, { "id": "n-026", "state": "closed", "question": { "en": "Whether the table should display xAI or the newer name. The ledger key company: xai remains unchanged." }, "policy": { "en": "Use both old and new names in searches. Leave the display label to the editor and keep the company key unchanged to preserve references." }, "source_ids": [ "src-spacex-prospectus-2026" ], "resolution": "dec-Q", "closed_at": "2026-08-28", "blocking": [], "owner": "editor", "provenance": { "en": "Restored because it is explicitly specified in BRIEF.md but was missing from notes.open_questions in the received archive." } }, { "id": "n-027", "state": "closed", "question": { "en": "dec-P's criterion has one condition and sets no threshold of scale. Applied literally, companies beyond the seven also offer a general-purpose AI service that ordinary users can use directly, including providers based in Europe and in China. The seven are not the full set of companies meeting the criterion. The obligation carried from n-023, to list the excluded companies and why, therefore cannot be discharged as things stand, because for those companies the reason would not be that they fail the criterion." }, "policy": { "en": "Before rewriting the criterion a third time, settle what the report claims about its selection. Three options. (1) Add every company meeting the criterion. (2) Add a second condition that is not arbitrary, which requires showing it is not arbitrary in the terms that led dec-J to drop the scale condition. (3) Drop the claim that the seven are the full set, and set out in the body both the companies investigated and the companies that meet the criterion but were not investigated, with the reason. Under any option, every company named in the body gets a source record. No company names from memory." }, "source_ids": [], "resolution": "dec-R", "closed_at": "2026-08-28", "blocking": [], "owner": "editor" } ] } }