Record reviewed readings of the week-40, week-41, and week-42 briefs.

Each publication keeps only claims a ledger capture supports. Launch stories, index scores, and Frontier, Efficient, and Budget labels stay excluded.

Assistant: grok
Assistant-Session: 01a1925d-3496-7011-a33e-6bfb99f4f4d3

SCOPE.md

@@ -10,32 +10,59 @@
 - Empirical outcome recording with unique runs, exact-task/profile Wilson intervals,
   provisional sufficiency thresholds and explicit missing evidence.
 - Frozen recommendations, deterministic historical replay and Markdown reports.
+- Reviewed publication records. An included claim must name a capture already in
+  the ledger. A generated draft is not that record. Reviewed publications now
+  cover the week-40, week-41, and week-42 briefs.
 - Configured Europe/Berlin daily/weekly job runner, retries, locking, health receipts,
   and user systemd timer units for the local workstation.
-- Retention of unarchived ChatGPT and Grok automation briefs from the local drop folder.
-  Website accounts are not polled; those runs have no public export.
+- Retention of unarchived ChatGPT and Grok automation briefs from the local drop folder,
+  and of Grok mail routed through inbound-library. Website accounts are not polled.
+  Mail seen so far is a teaser, not the full brief.
 - Behavior tests and isolated ledger/advisor demonstrations. `make check` is the current count.
-- Coulomb workplans and Statehub integration.
+- Coulomb workplans and Statehub integration. The hub holds work projections, not research evidence.
 
-## Current evidence boundary
+## Evidence boundary (2026-10-31)
 
-Four primary source pages have been captured. A reviewed two-profile price comparison
-is retained, but requester access and exact benchmark results are not evidenced.
-The initial real comparison therefore withholds Frontier/Efficient/Budget winners.
-Other imported claims remain unverified research leads. Real task outcomes are absent;
-full ranking and calibration behavior are demonstrated with isolated synthetic fixtures.
+Five catalog sources are captured through 30 October. SWE-bench has not changed.
+Reviewed readings, status still reported, cover the 30 October Anthropic pricing
+page and OpenRouter catalog membership and per-token fields from 22–30 October.
+The pricing page presents Max at $100 a month and Ultra at $200. No Sonnet, Opus,
+or Haiku token-price change was found in the visible text. The Sonnet page and the
+Sonnet 5 launch note changed a footer sentence only. `frontier/ledger-2026-09.md`
+remains the September import. Week 40 publication
+`b7a2614df227142c610e1402f2e0f7d84c62ef48bbb1a503af1d072159149d0a` keeps Sonnet 5.5
+prices and OpenRouter rows for Sonnet 5.5 and GPT-6.1 Sol. Week 41 publication
+`ae08625ddf94184bbbd716be73d5e2debac28948a6fe6817bbb302fc278942c7` keeps the
+7 October Haiku 5.5 rates, the Sonnet 5.5 cache-read change from $0.20 to $0.10,
+and the OpenRouter rows for Haiku 5.5 and Mistral Large 4. Week 42 publication
+`db9e3a7068329a605daee94cc0429706d0aedfaeb42c4f167b88d21483783339` keeps the
+OpenRouter rows for Muse Wonder, Gemini 4 Argon, Muse Spark Open, and DeepSeek
+V4.1 Flash 1017, and the 18 October Haiku 5.5 and Sonnet 5.5 rates. The rest of
+each Grok brief stays excluded. Price readings are in `frontier/readings-2026-W40.md`,
+`frontier/readings-2026-W41.md`, and `frontier/readings-2026-W42.md`. A reviewed two-profile comparison is retained, and
+it withholds Frontier, Efficient, and Budget winners because requester access and
+exact benchmark results are not evidenced.
+
+Grok daily briefs for 1–30 October and weekly briefs for weeks 40–43 are retained
+as unverified external digests. The 31 October daily file is empty and still in
+the drop folder. ChatGPT has no mail path. Pulling website runs without an operator
+remains IR-WP-0009-T01 (wait). A week of full briefs with no operator step remains
+IR-WP-0010-T05 (todo). The October files were placed in the drop folder.
+
+The key check in IR-WP-0005 is done. Paid field runs still wait on IR-WP-0006-T03:
+the USD 0.023712 reservation has no authoritative provider billing artifact.
+IR-WP-0007-T01 (comparable profiles) is todo. T02 and T04 stay wait until that
+reconciliation and matched outcomes exist. Real task outcomes are absent. Full
+ranking and calibration behavior are demonstrated with isolated synthetic fixtures.
 
 The implementation is a local CLI and curator workflow, not a hosted service or UI.
 A budgeted OpenRouter work-sample runner and portable recommendation bundles are
-implemented; real execution is waiting on valid credentials. It does not learn
-cross-task calibration, extract trustworthy
-facts automatically, optimize decomposition, or estimate empirical cost/runtime
-quantiles. Published costs are explicit workload scenarios. Catalog expansion and
-actual outcome collection continue in IR-WP-0007.
-
-Statehub holds work projections, not research evidence. Artifact hashes prove local
+implemented. It does not learn cross-task calibration, extract trustworthy facts
+automatically, optimize decomposition, or estimate empirical cost/runtime quantiles.
+Published costs are explicit workload scenarios. Artifact hashes prove local
 integrity, not truth. Scheduler operation depends on this workstation and its user
-service manager; publication still requires editorial synthesis/review.
+service manager. A draft or an external brief becomes a belief, a price, or a
+recommendation only after editorial review.
 
 ## Generated work-record index
 

data/publications/08b39af20eef61d36593b4ce4e4a83235184fd689f8ce31c27a48b31fe0c0abf.json

@@ -0,0 +1 @@
+{"claims":[{"capture_id":"50eb50a66a6c4770b065c379ccd56b96d1f68a7b98e6ed28874780d7cc506421","disposition":"included","statement":"The 6 October capture of https://claude.com/pricing lists Sonnet 5.5 prompt-cache read at $0.20 per million tokens and does not contain Haiku 5.5."},{"capture_id":"16d63ca029d7b392c26fb6f73c5c6dbfe26f65edbceb5c5503caf3f72e0d58d9","disposition":"included","statement":"The 7 October capture of https://claude.com/pricing lists Haiku 5.5. For prompts of at most 100,000 tokens the row is input $0.10, output $0.50, cache read $0.01, and cache write $0.125 per million tokens. For longer prompts it is input $0.50, output $2.50, cache read $0.05, and cache write $0.625. Haiku 4.5 remains on the legacy table at input $1, output $5, and cache read $0.10. Sonnet 5.5 cache read on this page is $0.10 per million tokens, with cache write $2.50, input $2, and output $10. The page does not state a 75% or 90% run-cost reduction, effort levels, or OSWorld or Terminal-Bench scores."},{"capture_id":"6f7047207dcfa2c384603d0e71ec6d666789d20ee6d944288e4e74dbb44dbb7e","disposition":"included","statement":"The 11 October capture of the pricing page still lists Haiku 5.5 at those tiered rates and Sonnet 5.5 cache read at $0.10 per million tokens."},{"capture_id":"c413e6a4edb7040e76e6459bb498cc7e3fe9bd95a47ab0d3763eef29466a9d64","disposition":"included","statement":"The 5 October OpenRouter models capture lists neither mistralai/mistral-large-4-0 nor anthropic/claude-haiku-5.5."},{"capture_id":"662d5a928eeafc1742f4d0d0b5a586af0735e56a59db7881000cb1f1aebb721d","disposition":"included","statement":"The 6 October OpenRouter models capture lists mistralai/mistral-large-4-0 as Mistral: Mistral Large 4, context 524288, prompt 0.00000068 and completion 0.00000209 per token. It does not list anthropic/claude-haiku-5.5. The row does not state a parameter count, a weights-release date, or an index score."},{"capture_id":"9c1d72cba28a9e34f85bc8cdeb3c8993588c35e4c631f126ab6219f370583e77","disposition":"included","statement":"The 7 October OpenRouter models capture lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, with an override from 100000 prompt tokens at prompt 0.0000005 and completion 0.0000025. Mistral Large 4 is still at context 524288."},{"capture_id":"f437a81287dd115f97de66b7058a31e19fe0b20a18e38909c0c2e5f13377d49e","disposition":"included","statement":"The 8 October OpenRouter models capture lists mistralai/mistral-large-4-0 at context 1048576, with the same prompt 0.00000068 and completion 0.00000209 per token."},{"capture_id":"1f957c1af24ec80bdae7adedba3020dfbf28ac2ed11a75ed1ad09800df1bc5e5","disposition":"included","statement":"The 11 October OpenRouter models capture still lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, and mistralai/mistral-large-4-0 at context 1048576 with prompt 0.00000068 and completion 0.00000209."},{"disposition":"excluded","exclusion_reason":"None of those primaries are captured in the ledger.","statement":"Anthropic's report of unintended Claude actions and the live-internet cutoff for internal evaluations; OpenAI's mathematics manuscript release; the GPT-6 ChatGPT rollout; Nadella's insider-risk essay; the OpenAI revenue figures; and the Super Intelligence Force."},{"disposition":"excluded","exclusion_reason":"The pricing page lists tiered token rates and does not state those percentages, effort levels, or benchmark scores. No Artificial Analysis capture is in the ledger.","statement":"The brief's claim that Haiku 5.5 cuts average run cost by about 75% versus Haiku 4.5 and by 90% on short prompts, that it is the first Haiku with effort levels and adaptive thinking, and the stated OSWorld, Terminal-Bench, HLE, and Artificial Analysis scores."},{"disposition":"excluded","exclusion_reason":"No Mistral announcement is captured. The OpenRouter row is a separate included claim and does not establish those statements.","statement":"Mistral Large 4's parameter count, the lab's multimodal and context claims, the end-of-October weights promise, and the Artificial Analysis score of about 38."},{"disposition":"excluded","exclusion_reason":"No Epoch or Artificial Analysis capture is in the ledger.","statement":"Epoch Capabilities Index figures and the Artificial Analysis Index v4.3.2 ordering for Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-6 Astra, Gemini 4 Argon, and Haiku 5.5."},{"disposition":"excluded","exclusion_reason":"Those labels need a task, five instruments, and comparable profiles. The compendium proposals depend on the uncaptured Anthropic report and the Nadella text.","statement":"The brief's Frontier, Efficient, and Budget labels, and the compendium proposals on agent circumvention and external emergency brakes."}],"reviewed_at":"2026-10-31T17:07:51+01:00","reviewer":"grok","schema_version":1,"title":"Week 41 brief against ledger captures"}

data/publications/4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8.json

@@ -0,0 +1 @@
+{"claims":[{"capture_id":"14cfc543613953d0ca53fb12a0c1fbb6089d9a0b6ddf8199a61ba0876cc9984a","disposition":"included","statement":"The 28 September capture of https://www.anthropic.com/claude/sonnet contains Claude Sonnet 5.5. The 27 September capture of that URL does not. The 28 September page states a price of $2 per million input tokens and $10 per million output tokens, names the API id claude-sonnet-5-5, and states US-only inference at 1.1x for input and output tokens."},{"capture_id":"1c69271364dc66cd25db0cee0992f01fc7fc674499f4ad0094618c997e029517","disposition":"included","statement":"The 4 October capture of the same Sonnet page still presents Sonnet 5.5 as available to chat on Claude.ai (web, iOS, and Android) and on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. It calls Sonnet 5.5 a faster, lower-cost complement to Opus 5.5 and says it needs far fewer tokens than Sonnet 5. It says API users have fine-grained control over thinking effort. It does not say there are five effort levels."},{"capture_id":"e154561c0dbc3fb6ec81c5ef3bfe485951d7d299541fd2852610f9b6c9454f04","disposition":"included","statement":"The 4 October capture of https://claude.com/pricing lists Sonnet 5.5 at $2 input and $10 output per million tokens, with prompt-cache read $0.20 and write $2.50. The same table lists Sonnet 5 at $2 input and $10 output, Opus 5.5 at $4 input and $20 output with cache read $0.20 and write $5, Fable 5.1 at $10 input and $50 output with cache read $0.25 and write $12.50, and Haiku 4.5 at $1 input and $5 output. It states US-only inference at 1.1x. It does not list a Haiku 5.5 row."},{"capture_id":"1c85432915fef9dbdbca4fc15b395b46e870d1ca5a95aa3f8ac19cbbf038c8a0","disposition":"included","statement":"The 29 September OpenRouter models capture is the first in this ledger that lists openai/gpt-6.1-sol. That listing is a catalog row. It is not an OpenAI launch post, a ChatGPT product surface, or a cached-input price."},{"capture_id":"888d6bb7b47f4a07de9eaa66b536cd58209e687cd8b890b804fae5ceea5f0765","disposition":"included","statement":"The 4 October OpenRouter models capture lists anthropic/claude-sonnet-5.5 at context 1000000, prompt 0.000002 and completion 0.00001 per token; openai/gpt-6.1-sol and openai/gpt-6.1-sol-pro at context 1050000 with those same prompt and completion fields; openai/gpt-6-luna at prompt 0.0000001 and completion 0.0000005; and openai/gpt-6-astra at prompt 0.00001 and completion 0.00005. It lists no id containing gpt-6.1-astra and no id containing argon."},{"disposition":"excluded","exclusion_reason":"No Artificial Analysis artifact is in the ledger. The token-heavy cost claim also runs against the Sonnet page, which says Sonnet 5.5 needs far fewer tokens than Sonnet 5.","statement":"Artificial Analysis scores in the week-40 brief, including Sonnet 5.5 at 56, Opus 5.5 at 58, Terminal-Bench 4.0 at 64%, about $7.60 per Index task, the Pareto placement, and a pre-release structured-output bug."},{"disposition":"excluded","exclusion_reason":"No OpenAI or TechCrunch capture is in the ledger. The OpenRouter row for gpt-6.1-sol is a separate included claim and does not establish those product or journalism statements.","statement":"OpenAI's GPT-6.1 Sol launch post, availability in ChatGPT Work and Codex, cached input at $0.10, a higher rate above 272k prompt tokens, the DeepSWE lab claim, and the report that GPT-6.1 Astra was held back."},{"disposition":"excluded","exclusion_reason":"No Google capture is in the ledger. Absence of an argon id on the 4 October OpenRouter document does not establish the announcement.","statement":"The Gemini 4 Argon announcement, Fairwind access, the stated list prices, and the lab benchmark figures."},{"disposition":"excluded","exclusion_reason":"No Epoch capture is in the ledger.","statement":"Epoch Capabilities Index figures for 4 October, including Opus 5.5 at 167 and Sonnet 5.5 at about 165."},{"disposition":"excluded","exclusion_reason":"None of those primaries are captured in the ledger.","statement":"The White House Accord on Super Intelligence, the ARC Prize GLM 5.3 Flash percentages, AA-AgentPerf-Local, the Solar Mini 4 Index score, Paper2Agent, and the Gothenburg robot-scientist note."},{"disposition":"excluded","exclusion_reason":"Those labels need a task, five instruments, and comparable profiles. The output-token proposal depends on the uncaptured Artificial Analysis cost figure.","statement":"The brief's Frontier, Efficient, and Budget labels, and the compendium proposal that a mid-tier matches flagship scores by spending output tokens."}],"reviewed_at":"2026-10-31T16:58:10+01:00","reviewer":"grok","schema_version":1,"title":"Week 40 brief against ledger captures"}

data/publications/b7a2614df227142c610e1402f2e0f7d84c62ef48bbb1a503af1d072159149d0a.json

@@ -0,0 +1 @@
+{"claims":[{"capture_id":"68d873eed4c424fd3683220a0d5c34a8c48f196305160de937ba2131ad4cd668","disposition":"included","statement":"The 27 September capture of https://www.anthropic.com/claude/sonnet does not contain Sonnet 5.5."},{"capture_id":"14cfc543613953d0ca53fb12a0c1fbb6089d9a0b6ddf8199a61ba0876cc9984a","disposition":"included","statement":"The 28 September capture of https://www.anthropic.com/claude/sonnet contains Claude Sonnet 5.5. It states a price of $2 per million input tokens and $10 per million output tokens, names the API id claude-sonnet-5-5, and states US-only inference at 1.1x for input and output tokens."},{"capture_id":"1c69271364dc66cd25db0cee0992f01fc7fc674499f4ad0094618c997e029517","disposition":"included","statement":"The 4 October capture of the same Sonnet page still presents Sonnet 5.5 as available to chat on Claude.ai (web, iOS, and Android) and on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. It calls Sonnet 5.5 a faster, lower-cost complement to Opus 5.5 and says it needs far fewer tokens than Sonnet 5. It says API users have fine-grained control over thinking effort. It does not say there are five effort levels."},{"capture_id":"e154561c0dbc3fb6ec81c5ef3bfe485951d7d299541fd2852610f9b6c9454f04","disposition":"included","statement":"The 4 October capture of https://claude.com/pricing lists Sonnet 5.5 at $2 input and $10 output per million tokens, with prompt-cache read $0.20 and write $2.50. The same table lists Sonnet 5 at $2 input and $10 output, Opus 5.5 at $4 input and $20 output with cache read $0.20 and write $5, Fable 5.1 at $10 input and $50 output with cache read $0.25 and write $12.50, and Haiku 4.5 at $1 input and $5 output. It states US-only inference at 1.1x. It does not list a Haiku 5.5 row."},{"capture_id":"1c85432915fef9dbdbca4fc15b395b46e870d1ca5a95aa3f8ac19cbbf038c8a0","disposition":"included","statement":"The 29 September OpenRouter models capture is the first in this ledger that lists openai/gpt-6.1-sol. That listing is a catalog row. It is not an OpenAI launch post, a ChatGPT product surface, or a cached-input price."},{"capture_id":"888d6bb7b47f4a07de9eaa66b536cd58209e687cd8b890b804fae5ceea5f0765","disposition":"included","statement":"The 4 October OpenRouter models capture lists anthropic/claude-sonnet-5.5 at context 1000000, prompt 0.000002 and completion 0.00001 per token; openai/gpt-6.1-sol and openai/gpt-6.1-sol-pro at context 1050000 with those same prompt and completion fields; openai/gpt-6-luna at prompt 0.0000001 and completion 0.0000005; and openai/gpt-6-astra at prompt 0.00001 and completion 0.00005. It lists no id containing gpt-6.1-astra and no id containing argon."},{"disposition":"excluded","exclusion_reason":"No Artificial Analysis artifact is in the ledger. The token-heavy cost claim also runs against the Sonnet page, which says Sonnet 5.5 needs far fewer tokens than Sonnet 5.","statement":"Artificial Analysis scores in the week-40 brief, including Sonnet 5.5 at 56, Opus 5.5 at 58, Terminal-Bench 4.0 at 64%, about $7.60 per Index task, the Pareto placement, and a pre-release structured-output bug."},{"disposition":"excluded","exclusion_reason":"No OpenAI or TechCrunch capture is in the ledger. The OpenRouter row for gpt-6.1-sol is a separate included claim and does not establish those product or journalism statements.","statement":"OpenAI's GPT-6.1 Sol launch post, availability in ChatGPT Work and Codex, cached input at $0.10, a higher rate above 272k prompt tokens, the DeepSWE lab claim, and the report that GPT-6.1 Astra was held back."},{"disposition":"excluded","exclusion_reason":"No Google capture is in the ledger. Absence of an argon id on the 4 October OpenRouter document does not establish the announcement.","statement":"The Gemini 4 Argon announcement, Fairwind access, the stated list prices, and the lab benchmark figures."},{"disposition":"excluded","exclusion_reason":"No Epoch capture is in the ledger.","statement":"Epoch Capabilities Index figures for 4 October, including Opus 5.5 at 167 and Sonnet 5.5 at about 165."},{"disposition":"excluded","exclusion_reason":"None of those primaries are captured in the ledger.","statement":"The White House Accord on Super Intelligence, the ARC Prize GLM 5.3 Flash percentages, AA-AgentPerf-Local, the Solar Mini 4 Index score, Paper2Agent, and the Gothenburg robot-scientist note."},{"disposition":"excluded","exclusion_reason":"Those labels need a task, five instruments, and comparable profiles. The output-token proposal depends on the uncaptured Artificial Analysis cost figure.","statement":"The brief's Frontier, Efficient, and Budget labels, and the compendium proposal that a mid-tier matches flagship scores by spending output tokens."}],"reviewed_at":"2026-10-31T16:58:30+01:00","reviewer":"grok","schema_version":1,"supersedes":"4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8","title":"Week 40 brief against ledger captures"}

digests/reviewed/08b39af20eef61d36593b4ce4e4a83235184fd689f8ce31c27a48b31fe0c0abf.md

@@ -0,0 +1,37 @@
+# Reviewed publication
+
+This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
+
+- Revision: `08b39af20eef61d36593b4ce4e4a83235184fd689f8ce31c27a48b31fe0c0abf`
+- Title: Week 41 brief against ledger captures
+- Reviewer: grok
+- Reviewed: 2026-10-31T17:07:51+01:00
+
+## Claims
+
+- included: The 6 October capture of https://claude.com/pricing lists Sonnet 5.5 prompt-cache read at $0.20 per million tokens and does not contain Haiku 5.5.
+  - Capture: `50eb50a66a6c4770b065c379ccd56b96d1f68a7b98e6ed28874780d7cc506421`
+- included: The 7 October capture of https://claude.com/pricing lists Haiku 5.5. For prompts of at most 100,000 tokens the row is input $0.10, output $0.50, cache read $0.01, and cache write $0.125 per million tokens. For longer prompts it is input $0.50, output $2.50, cache read $0.05, and cache write $0.625. Haiku 4.5 remains on the legacy table at input $1, output $5, and cache read $0.10. Sonnet 5.5 cache read on this page is $0.10 per million tokens, with cache write $2.50, input $2, and output $10. The page does not state a 75% or 90% run-cost reduction, effort levels, or OSWorld or Terminal-Bench scores.
+  - Capture: `16d63ca029d7b392c26fb6f73c5c6dbfe26f65edbceb5c5503caf3f72e0d58d9`
+- included: The 11 October capture of the pricing page still lists Haiku 5.5 at those tiered rates and Sonnet 5.5 cache read at $0.10 per million tokens.
+  - Capture: `6f7047207dcfa2c384603d0e71ec6d666789d20ee6d944288e4e74dbb44dbb7e`
+- included: The 5 October OpenRouter models capture lists neither mistralai/mistral-large-4-0 nor anthropic/claude-haiku-5.5.
+  - Capture: `c413e6a4edb7040e76e6459bb498cc7e3fe9bd95a47ab0d3763eef29466a9d64`
+- included: The 6 October OpenRouter models capture lists mistralai/mistral-large-4-0 as Mistral: Mistral Large 4, context 524288, prompt 0.00000068 and completion 0.00000209 per token. It does not list anthropic/claude-haiku-5.5. The row does not state a parameter count, a weights-release date, or an index score.
+  - Capture: `662d5a928eeafc1742f4d0d0b5a586af0735e56a59db7881000cb1f1aebb721d`
+- included: The 7 October OpenRouter models capture lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, with an override from 100000 prompt tokens at prompt 0.0000005 and completion 0.0000025. Mistral Large 4 is still at context 524288.
+  - Capture: `9c1d72cba28a9e34f85bc8cdeb3c8993588c35e4c631f126ab6219f370583e77`
+- included: The 8 October OpenRouter models capture lists mistralai/mistral-large-4-0 at context 1048576, with the same prompt 0.00000068 and completion 0.00000209 per token.
+  - Capture: `f437a81287dd115f97de66b7058a31e19fe0b20a18e38909c0c2e5f13377d49e`
+- included: The 11 October OpenRouter models capture still lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, and mistralai/mistral-large-4-0 at context 1048576 with prompt 0.00000068 and completion 0.00000209.
+  - Capture: `1f957c1af24ec80bdae7adedba3020dfbf28ac2ed11a75ed1ad09800df1bc5e5`
+- excluded: Anthropic's report of unintended Claude actions and the live-internet cutoff for internal evaluations; OpenAI's mathematics manuscript release; the GPT-6 ChatGPT rollout; Nadella's insider-risk essay; the OpenAI revenue figures; and the Super Intelligence Force.
+  - Excluded because: None of those primaries are captured in the ledger.
+- excluded: The brief's claim that Haiku 5.5 cuts average run cost by about 75% versus Haiku 4.5 and by 90% on short prompts, that it is the first Haiku with effort levels and adaptive thinking, and the stated OSWorld, Terminal-Bench, HLE, and Artificial Analysis scores.
+  - Excluded because: The pricing page lists tiered token rates and does not state those percentages, effort levels, or benchmark scores. No Artificial Analysis capture is in the ledger.
+- excluded: Mistral Large 4's parameter count, the lab's multimodal and context claims, the end-of-October weights promise, and the Artificial Analysis score of about 38.
+  - Excluded because: No Mistral announcement is captured. The OpenRouter row is a separate included claim and does not establish those statements.
+- excluded: Epoch Capabilities Index figures and the Artificial Analysis Index v4.3.2 ordering for Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-6 Astra, Gemini 4 Argon, and Haiku 5.5.
+  - Excluded because: No Epoch or Artificial Analysis capture is in the ledger.
+- excluded: The brief's Frontier, Efficient, and Budget labels, and the compendium proposals on agent circumvention and external emergency brakes.
+  - Excluded because: Those labels need a task, five instruments, and comparable profiles. The compendium proposals depend on the uncaptured Anthropic report and the Nadella text.

digests/reviewed/4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8.md

@@ -0,0 +1,33 @@
+# Reviewed publication
+
+This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
+
+- Revision: `4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8`
+- Title: Week 40 brief against ledger captures
+- Reviewer: grok
+- Reviewed: 2026-10-31T16:58:10+01:00
+
+## Claims
+
+- included: The 28 September capture of https://www.anthropic.com/claude/sonnet contains Claude Sonnet 5.5. The 27 September capture of that URL does not. The 28 September page states a price of $2 per million input tokens and $10 per million output tokens, names the API id claude-sonnet-5-5, and states US-only inference at 1.1x for input and output tokens.
+  - Capture: `14cfc543613953d0ca53fb12a0c1fbb6089d9a0b6ddf8199a61ba0876cc9984a`
+- included: The 4 October capture of the same Sonnet page still presents Sonnet 5.5 as available to chat on Claude.ai (web, iOS, and Android) and on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. It calls Sonnet 5.5 a faster, lower-cost complement to Opus 5.5 and says it needs far fewer tokens than Sonnet 5. It says API users have fine-grained control over thinking effort. It does not say there are five effort levels.
+  - Capture: `1c69271364dc66cd25db0cee0992f01fc7fc674499f4ad0094618c997e029517`
+- included: The 4 October capture of https://claude.com/pricing lists Sonnet 5.5 at $2 input and $10 output per million tokens, with prompt-cache read $0.20 and write $2.50. The same table lists Sonnet 5 at $2 input and $10 output, Opus 5.5 at $4 input and $20 output with cache read $0.20 and write $5, Fable 5.1 at $10 input and $50 output with cache read $0.25 and write $12.50, and Haiku 4.5 at $1 input and $5 output. It states US-only inference at 1.1x. It does not list a Haiku 5.5 row.
+  - Capture: `e154561c0dbc3fb6ec81c5ef3bfe485951d7d299541fd2852610f9b6c9454f04`
+- included: The 29 September OpenRouter models capture is the first in this ledger that lists openai/gpt-6.1-sol. That listing is a catalog row. It is not an OpenAI launch post, a ChatGPT product surface, or a cached-input price.
+  - Capture: `1c85432915fef9dbdbca4fc15b395b46e870d1ca5a95aa3f8ac19cbbf038c8a0`
+- included: The 4 October OpenRouter models capture lists anthropic/claude-sonnet-5.5 at context 1000000, prompt 0.000002 and completion 0.00001 per token; openai/gpt-6.1-sol and openai/gpt-6.1-sol-pro at context 1050000 with those same prompt and completion fields; openai/gpt-6-luna at prompt 0.0000001 and completion 0.0000005; and openai/gpt-6-astra at prompt 0.00001 and completion 0.00005. It lists no id containing gpt-6.1-astra and no id containing argon.
+  - Capture: `888d6bb7b47f4a07de9eaa66b536cd58209e687cd8b890b804fae5ceea5f0765`
+- excluded: Artificial Analysis scores in the week-40 brief, including Sonnet 5.5 at 56, Opus 5.5 at 58, Terminal-Bench 4.0 at 64%, about $7.60 per Index task, the Pareto placement, and a pre-release structured-output bug.
+  - Excluded because: No Artificial Analysis artifact is in the ledger. The token-heavy cost claim also runs against the Sonnet page, which says Sonnet 5.5 needs far fewer tokens than Sonnet 5.
+- excluded: OpenAI's GPT-6.1 Sol launch post, availability in ChatGPT Work and Codex, cached input at $0.10, a higher rate above 272k prompt tokens, the DeepSWE lab claim, and the report that GPT-6.1 Astra was held back.
+  - Excluded because: No OpenAI or TechCrunch capture is in the ledger. The OpenRouter row for gpt-6.1-sol is a separate included claim and does not establish those product or journalism statements.
+- excluded: The Gemini 4 Argon announcement, Fairwind access, the stated list prices, and the lab benchmark figures.
+  - Excluded because: No Google capture is in the ledger. Absence of an argon id on the 4 October OpenRouter document does not establish the announcement.
+- excluded: Epoch Capabilities Index figures for 4 October, including Opus 5.5 at 167 and Sonnet 5.5 at about 165.
+  - Excluded because: No Epoch capture is in the ledger.
+- excluded: The White House Accord on Super Intelligence, the ARC Prize GLM 5.3 Flash percentages, AA-AgentPerf-Local, the Solar Mini 4 Index score, Paper2Agent, and the Gothenburg robot-scientist note.
+  - Excluded because: None of those primaries are captured in the ledger.
+- excluded: The brief's Frontier, Efficient, and Budget labels, and the compendium proposal that a mid-tier matches flagship scores by spending output tokens.
+  - Excluded because: Those labels need a task, five instruments, and comparable profiles. The output-token proposal depends on the uncaptured Artificial Analysis cost figure.

digests/reviewed/ae08625ddf94184bbbd716be73d5e2debac28948a6fe6817bbb302fc278942c7.md

@@ -0,0 +1,38 @@
+# Reviewed publication
+
+This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
+
+- Revision: `ae08625ddf94184bbbd716be73d5e2debac28948a6fe6817bbb302fc278942c7`
+- Title: Week 41 brief against ledger captures
+- Reviewer: grok
+- Reviewed: 2026-10-31T17:08:25+01:00
+- Supersedes: `08b39af20eef61d36593b4ce4e4a83235184fd689f8ce31c27a48b31fe0c0abf`
+
+## Claims
+
+- included: The 6 October capture of https://claude.com/pricing lists Sonnet 5.5 prompt-cache read at $0.20 per million tokens and does not contain Haiku 5.5.
+  - Capture: `50eb50a66a6c4770b065c379ccd56b96d1f68a7b98e6ed28874780d7cc506421`
+- included: The 7 October capture of https://claude.com/pricing lists Haiku 5.5. For prompts of at most 100,000 tokens the row is input $0.10, output $0.50, cache read $0.01, and cache write $0.125 per million tokens. For longer prompts it is input $0.50, output $2.50, cache read $0.05, and cache write $0.625. Haiku 4.5 remains on the legacy table at input $1, output $5, and cache read $0.10. Sonnet 5.5 cache read on this page is $0.10 per million tokens, with cache write $2.50, input $2, and output $10. The page does not state a 75% or 90% run-cost reduction, effort levels, or OSWorld or Terminal-Bench scores.
+  - Capture: `16d63ca029d7b392c26fb6f73c5c6dbfe26f65edbceb5c5503caf3f72e0d58d9`
+- included: The 11 October capture of https://claude.com/pricing still lists Haiku 5.5 at input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 for longer prompts, with cache read $0.01 / $0.05 and cache write $0.125 / $0.625 on those two tiers. Sonnet 5.5 cache read is still $0.10 per million tokens.
+  - Capture: `6f7047207dcfa2c384603d0e71ec6d666789d20ee6d944288e4e74dbb44dbb7e`
+- included: The 5 October OpenRouter models capture lists neither mistralai/mistral-large-4-0 nor anthropic/claude-haiku-5.5.
+  - Capture: `c413e6a4edb7040e76e6459bb498cc7e3fe9bd95a47ab0d3763eef29466a9d64`
+- included: The 6 October OpenRouter models capture lists mistralai/mistral-large-4-0 as Mistral: Mistral Large 4, context 524288, prompt 0.00000068 and completion 0.00000209 per token. It does not list anthropic/claude-haiku-5.5. The row does not state a parameter count, a weights-release date, or an index score.
+  - Capture: `662d5a928eeafc1742f4d0d0b5a586af0735e56a59db7881000cb1f1aebb721d`
+- included: The 7 October OpenRouter models capture lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, with an override from 100000 prompt tokens at prompt 0.0000005 and completion 0.0000025. Mistral Large 4 is still at context 524288.
+  - Capture: `9c1d72cba28a9e34f85bc8cdeb3c8993588c35e4c631f126ab6219f370583e77`
+- included: The 8 October OpenRouter models capture lists mistralai/mistral-large-4-0 at context 1048576, with the same prompt 0.00000068 and completion 0.00000209 per token.
+  - Capture: `f437a81287dd115f97de66b7058a31e19fe0b20a18e38909c0c2e5f13377d49e`
+- included: The 11 October OpenRouter models capture still lists anthropic/claude-haiku-5.5 at context 1000000, prompt 0.0000001 and completion 0.0000005 per token, and mistralai/mistral-large-4-0 at context 1048576 with prompt 0.00000068 and completion 0.00000209.
+  - Capture: `1f957c1af24ec80bdae7adedba3020dfbf28ac2ed11a75ed1ad09800df1bc5e5`
+- excluded: Anthropic's report of unintended Claude actions and the live-internet cutoff for internal evaluations; OpenAI's mathematics manuscript release; the GPT-6 ChatGPT rollout; Nadella's insider-risk essay; the OpenAI revenue figures; and the Super Intelligence Force.
+  - Excluded because: None of those primaries are captured in the ledger.
+- excluded: The brief's claim that Haiku 5.5 cuts average run cost by about 75% versus Haiku 4.5 and by 90% on short prompts, that it is the first Haiku with effort levels and adaptive thinking, and the stated OSWorld, Terminal-Bench, HLE, and Artificial Analysis scores.
+  - Excluded because: The pricing page lists tiered token rates and does not state those percentages, effort levels, or benchmark scores. No Artificial Analysis capture is in the ledger.
+- excluded: Mistral Large 4's parameter count, the lab's multimodal and context claims, the end-of-October weights promise, and the Artificial Analysis score of about 38.
+  - Excluded because: No Mistral announcement is captured. The OpenRouter row is a separate included claim and does not establish those statements.
+- excluded: Epoch Capabilities Index figures and the Artificial Analysis Index v4.3.2 ordering for Opus 5.5, Sonnet 5.5, Fable 5.1, GPT-6 Astra, Gemini 4 Argon, and Haiku 5.5.
+  - Excluded because: No Epoch or Artificial Analysis capture is in the ledger.
+- excluded: The brief's Frontier, Efficient, and Budget labels, and the compendium proposals on agent circumvention and external emergency brakes.
+  - Excluded because: Those labels need a task, five instruments, and comparable profiles. The compendium proposals depend on the uncaptured Anthropic report and the Nadella text.

digests/reviewed/b7a2614df227142c610e1402f2e0f7d84c62ef48bbb1a503af1d072159149d0a.md

@@ -0,0 +1,36 @@
+# Reviewed publication
+
+This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
+
+- Revision: `b7a2614df227142c610e1402f2e0f7d84c62ef48bbb1a503af1d072159149d0a`
+- Title: Week 40 brief against ledger captures
+- Reviewer: grok
+- Reviewed: 2026-10-31T16:58:30+01:00
+- Supersedes: `4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8`
+
+## Claims
+
+- included: The 27 September capture of https://www.anthropic.com/claude/sonnet does not contain Sonnet 5.5.
+  - Capture: `68d873eed4c424fd3683220a0d5c34a8c48f196305160de937ba2131ad4cd668`
+- included: The 28 September capture of https://www.anthropic.com/claude/sonnet contains Claude Sonnet 5.5. It states a price of $2 per million input tokens and $10 per million output tokens, names the API id claude-sonnet-5-5, and states US-only inference at 1.1x for input and output tokens.
+  - Capture: `14cfc543613953d0ca53fb12a0c1fbb6089d9a0b6ddf8199a61ba0876cc9984a`
+- included: The 4 October capture of the same Sonnet page still presents Sonnet 5.5 as available to chat on Claude.ai (web, iOS, and Android) and on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Foundry. It calls Sonnet 5.5 a faster, lower-cost complement to Opus 5.5 and says it needs far fewer tokens than Sonnet 5. It says API users have fine-grained control over thinking effort. It does not say there are five effort levels.
+  - Capture: `1c69271364dc66cd25db0cee0992f01fc7fc674499f4ad0094618c997e029517`
+- included: The 4 October capture of https://claude.com/pricing lists Sonnet 5.5 at $2 input and $10 output per million tokens, with prompt-cache read $0.20 and write $2.50. The same table lists Sonnet 5 at $2 input and $10 output, Opus 5.5 at $4 input and $20 output with cache read $0.20 and write $5, Fable 5.1 at $10 input and $50 output with cache read $0.25 and write $12.50, and Haiku 4.5 at $1 input and $5 output. It states US-only inference at 1.1x. It does not list a Haiku 5.5 row.
+  - Capture: `e154561c0dbc3fb6ec81c5ef3bfe485951d7d299541fd2852610f9b6c9454f04`
+- included: The 29 September OpenRouter models capture is the first in this ledger that lists openai/gpt-6.1-sol. That listing is a catalog row. It is not an OpenAI launch post, a ChatGPT product surface, or a cached-input price.
+  - Capture: `1c85432915fef9dbdbca4fc15b395b46e870d1ca5a95aa3f8ac19cbbf038c8a0`
+- included: The 4 October OpenRouter models capture lists anthropic/claude-sonnet-5.5 at context 1000000, prompt 0.000002 and completion 0.00001 per token; openai/gpt-6.1-sol and openai/gpt-6.1-sol-pro at context 1050000 with those same prompt and completion fields; openai/gpt-6-luna at prompt 0.0000001 and completion 0.0000005; and openai/gpt-6-astra at prompt 0.00001 and completion 0.00005. It lists no id containing gpt-6.1-astra and no id containing argon.
+  - Capture: `888d6bb7b47f4a07de9eaa66b536cd58209e687cd8b890b804fae5ceea5f0765`
+- excluded: Artificial Analysis scores in the week-40 brief, including Sonnet 5.5 at 56, Opus 5.5 at 58, Terminal-Bench 4.0 at 64%, about $7.60 per Index task, the Pareto placement, and a pre-release structured-output bug.
+  - Excluded because: No Artificial Analysis artifact is in the ledger. The token-heavy cost claim also runs against the Sonnet page, which says Sonnet 5.5 needs far fewer tokens than Sonnet 5.
+- excluded: OpenAI's GPT-6.1 Sol launch post, availability in ChatGPT Work and Codex, cached input at $0.10, a higher rate above 272k prompt tokens, the DeepSWE lab claim, and the report that GPT-6.1 Astra was held back.
+  - Excluded because: No OpenAI or TechCrunch capture is in the ledger. The OpenRouter row for gpt-6.1-sol is a separate included claim and does not establish those product or journalism statements.
+- excluded: The Gemini 4 Argon announcement, Fairwind access, the stated list prices, and the lab benchmark figures.
+  - Excluded because: No Google capture is in the ledger. Absence of an argon id on the 4 October OpenRouter document does not establish the announcement.
+- excluded: Epoch Capabilities Index figures for 4 October, including Opus 5.5 at 167 and Sonnet 5.5 at about 165.
+  - Excluded because: No Epoch capture is in the ledger.
+- excluded: The White House Accord on Super Intelligence, the ARC Prize GLM 5.3 Flash percentages, AA-AgentPerf-Local, the Solar Mini 4 Index score, Paper2Agent, and the Gothenburg robot-scientist note.
+  - Excluded because: None of those primaries are captured in the ledger.
+- excluded: The brief's Frontier, Efficient, and Budget labels, and the compendium proposal that a mid-tier matches flagship scores by spending output tokens.
+  - Excluded because: Those labels need a task, five instruments, and comparable profiles. The output-token proposal depends on the uncaptured Artificial Analysis cost figure.

digests/reviewed/db9e3a7068329a605daee94cc0429706d0aedfaeb42c4f167b88d21483783339.md

@@ -0,0 +1,41 @@
+# Reviewed publication
+
+This file is a reviewed publication. It is not a generated daily, weekly, or external draft.
+
+- Revision: `db9e3a7068329a605daee94cc0429706d0aedfaeb42c4f167b88d21483783339`
+- Title: Week 42 brief against ledger captures
+- Reviewer: grok
+- Reviewed: 2026-10-31T17:18:01+01:00
+
+## Claims
+
+- included: The 11 October OpenRouter models capture lists neither meta/muse-wonder, nor google/gemini-4-argon, nor meta/muse-spark-open. No model id in that capture contains wonder.
+  - Capture: `1f957c1af24ec80bdae7adedba3020dfbf28ac2ed11a75ed1ad09800df1bc5e5`
+- included: The 12 October OpenRouter models capture lists meta/muse-wonder as Meta: Muse Wonder, context 1048576, prompt 0.000004 and completion 0.00002 per token. It does not list google/gemini-4-argon or meta/muse-spark-open. The only model id containing wonder is meta/muse-wonder. The row does not state Active Memory, an index score, or a subscription product.
+  - Capture: `5253f5d4ed3ca3fa28eeb3cd604042e8e8dcad5dae50271b3092a0d662eb31fb`
+- included: The 13 October OpenRouter models capture lists google/gemini-4-argon as Google: Gemini 4 Argon, context 1048576, prompt 0.000002 and completion 0.00001 per token, and google/gemini-4-argon:batch at prompt 0.000001 and completion 0.000005 per token. The rows do not state an introductory expiry date, a later rate, an app or cloud surface, or a benchmark score.
+  - Capture: `34ead8146ca85f71cf27d55517e32e69df885203ac59f2610ad4547886532981`
+- included: The 14 October OpenRouter models capture does not list meta/muse-spark-open.
+  - Capture: `8aa8cddc47ee1e901ba9fe9d5347c6486acdf95729921a3154864c7252b8d35e`
+- included: The 15 October OpenRouter models capture lists meta/muse-spark-open as Meta: Muse Spark Open, context 1048576, prompt 0.00000125 and completion 0.00000425 per token. The row does not state a parameter count, a hardware requirement, or an index score.
+  - Capture: `c2babd6a7580610b1b3f625440a39b98fd678a2faa6a8be6a3bc62aef5d91dfb`
+- included: The 16 October OpenRouter models capture does not list deepseek/deepseek-v4.1-flash-1017.
+  - Capture: `1310d0ed25b4a6c39a102b735a3eb8e68b614893066c6b1cc734e09eaea815ce`
+- included: The 17 October OpenRouter models capture lists deepseek/deepseek-v4.1-flash-1017 as DeepSeek: DeepSeek V4.1 Flash 1017, context 1048576, prompt 0.00000015 and completion 0.0000006 per token. The row includes UTC-window overrides, and some weekday windows are prompt 0.0000003 and completion 0.0000012. The row does not state an index score.
+  - Capture: `99f7dc881b11fe2f9f4136f6797bf884cea16cae53ae1e03a598035587c3b670`
+- included: The 18 October OpenRouter models capture lists meta/muse-wonder at context 1048576, prompt 0.000004 and completion 0.00002 per token; google/gemini-4-argon at context 1048576, prompt 0.000002 and completion 0.00001 per token; google/gemini-4-argon:batch at prompt 0.000001 and completion 0.000005 per token; meta/muse-spark-open at context 1048576, prompt 0.00000125 and completion 0.00000425 per token; and deepseek/deepseek-v4.1-flash-1017 at context 1048576, prompt 0.00000015 and completion 0.0000006 per token. The only model id containing wonder is meta/muse-wonder.
+  - Capture: `48a9163cfeafb75b72c8132a9c9cc866e3d7ee7bb22fd02f6640839e68bb6fb4`
+- included: The 18 October capture of https://claude.com/pricing lists Sonnet 5.5 at input $2 and output $10 per million tokens, with prompt-cache read $0.10 and cache write $2.50. It lists Haiku 5.5 at input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 for longer prompts, with cache read $0.01 and $0.05 and cache write $0.125 and $0.625 on those two tiers. Haiku 4.5 is on the legacy table at input $1 and output $5, cache read $0.10 and cache write $1.25. Opus 5.5 is input $4 and output $20, cache read $0.20 and cache write $5. The visible text does not contain Argon, Muse, Wonder, Gemini, or DeepSeek.
+  - Capture: `1bfb5ccc51900ff8bbf06aba43093d56c5da63705910a436d6d4fa17fc2b85fe`
+- excluded: Meta's Muse Wonder and Muse Wonder Full launch, Active Memory and Focus, the Artificial Analysis scores 55.8, 57.6, and 56, the frontier-class description, and the claim that Wonder Full is subscription-only with no public API.
+  - Excluded because: The Meta news page is not captured. The OpenRouter row is a separate included claim and does not establish the launch story, the index scores, or a subscription-only product. Absence of a second wonder id is not that product claim.
+- excluded: Gemini 4 Argon broadly shipping after Fairwind, introductory pricing until 30 November and a later $4 and $20 rate, availability in the Gemini app, Search, and Cloud, the ARC-AGI scores, the Artificial Analysis score near 52.6, and the Vals Index ranking.
+  - Excluded because: The Google blog and the ARC Prize posts are not captured. The OpenRouter token fields are a separate included claim. Those fields do not establish an introductory expiry or a later $4 and $20 rate.
+- excluded: Muse Spark Open as an 816B mixture-of-experts model with 42B active parameters, the claim that it runs on two Mac Studios, and the Artificial Analysis score near 39.
+  - Excluded because: The research.meta.ai post is not captured. The OpenRouter row is a separate included claim and does not establish a parameter count, a hardware requirement, or an index score.
+- excluded: Anthropic's METR employee-like access, China's draft export controls, Anthropic's workforce report, the Super Intelligence Force incident statement, Anthropic's Q3 revenue figure, and the court, attorney-general, FTC, and CERT-In items.
+  - Excluded because: None of those primaries are captured in the ledger.
+- excluded: The ARC Prize results for Gemini 4 Argon and the earlier GPT-6 Astra score of 99.9% on ARC-AGI-3.
+  - Excluded because: No ARC Prize capture is in the ledger.
+- excluded: The brief's Frontier, Efficient, and Budget labels, its compendium accept, watch, and reject proposals, the economics narrative, the task-router atlas notes, and the look-ahead items.
+  - Excluded because: Those labels need a task, five instruments, and comparable profiles. The compendium proposals, the economics narrative, and the task-router notes depend on uncaptured lab posts and index scores. The look-ahead dates are outside this week's captures.

digests/weekly/2026-W40.md

@@ -211,7 +211,14 @@ Source content was captured or changed. Extracted claims require editorial revie
 - Observed: 2026-09-28T19:41:58.299313+00:00
 - Observation: e869e72b634ba36c8f1726899a1306d136a055e7c06c74bea6427a734f65b31a
 
+## Reviewed reading
+
+The Grok week-40 brief is retained at `digests/external/weekly/2026-W40-grok.md` and stays unverified as a whole. The pass against ledger captures is publication `b7a2614df227142c610e1402f2e0f7d84c62ef48bbb1a503af1d072159149d0a`, which supersedes `4f36132f5fb9bc60aaaf44bbf8ed675a8f8048c01cd2f16107ba1419c26a85b8`.
+
+Included from captures: Sonnet 5.5 is absent on the 27 September Sonnet page and present on 28 September at $2 / $10 per million tokens, with API id `claude-sonnet-5-5` and US-only inference at 1.1×. The 4 October Sonnet page still shows it on Claude.ai and the named clouds, and says it needs far fewer tokens than Sonnet 5. The 4 October pricing page agrees on $2 / $10, lists cache read $0.20, and lists Opus 5.5 at $4 / $20, Fable 5.1 at $10 / $50, and Haiku 4.5 rather than Haiku 5.5. OpenRouter first lists `openai/gpt-6.1-sol` on 29 September. On 4 October it lists Sonnet 5.5 and GPT-6.1 Sol at the same per-token fields, and it lists neither `gpt-6.1-astra` nor an argon id.
+
+Left out: Artificial Analysis scores and the $7.60 task cost, the OpenAI launch post and the Astra-hold report, Gemini 4 Argon, Epoch's index, the White House accord, ARC Prize scores, and the compendium proposals. Frontier, Efficient, and Budget are not assigned. `frontier/ledger-2026-09.md` is unchanged. Price readings from the pricing page are in `frontier/readings-2026-W40.md`.
+
 ## Editorial follow-up
 
-Review source claims, contradictions, frontier/price impacts, compendium edits,
-benchmark health, and evidence gaps. Draft generation is not fact verification.
+Source-update sections above remain unextracted change notices. This pass does not make the external brief a belief.

digests/weekly/2026-W41.md

@@ -202,7 +202,14 @@ Source content was captured or changed. Extracted claims require editorial revie
 - Observed: 2026-10-07T19:44:35.676090+00:00
 - Observation: f5469fa1d0b1015497afbfcab81641a4fe3fa5868a0e8786dadca7360f53ac5b
 
+## Reviewed reading
+
+The Grok week-41 brief is retained at `digests/external/weekly/2026-W41-grok.md` and stays unverified as a whole. The pass against ledger captures is publication `ae08625ddf94184bbbd716be73d5e2debac28948a6fe6817bbb302fc278942c7`, which supersedes `08b39af20eef61d36593b4ce4e4a83235184fd689f8ce31c27a48b31fe0c0abf`.
+
+Included from captures: the 6 October pricing page has no Haiku 5.5 and lists Sonnet 5.5 cache read at $0.20 per million tokens. The 7 October page lists Haiku 5.5 at $0.10 / $0.50 per million for prompts up to 100,000 tokens and $0.50 / $2.50 above that, and it lists Sonnet 5.5 cache read at $0.10. Haiku 4.5 stays on the legacy table at $1 / $5. Those Haiku 5.5 rates are still on the 11 October page. OpenRouter has no Mistral Large 4 or Haiku 5.5 row on 5 October, lists Mistral Large 4 on 6 October at context 524288, lists Haiku 5.5 on 7 October at context 1000000, and shows Mistral Large 4 at context 1048576 from 8 October through 11 October. Prompt and completion fields for Mistral Large 4 stay `0.00000068` and `0.00000209`.
+
+Left out: the unintended-actions report, the mathematics manuscripts, the ChatGPT rollout, the Nadella essay, the revenue figures, the Super Intelligence Force, the 75% and 90% run-cost sentences, the lab benchmark percentages, Epoch, and Artificial Analysis. Frontier, Efficient, and Budget are not assigned. Price readings are in `frontier/readings-2026-W41.md`.
+
 ## Editorial follow-up
 
-Review source claims, contradictions, frontier/price impacts, compendium edits,
-benchmark health, and evidence gaps. Draft generation is not fact verification.
+Source-update sections above remain unextracted change notices. This pass does not make the external brief a belief.

digests/weekly/2026-W42.md

@@ -193,7 +193,14 @@ Source content was captured or changed. Extracted claims require editorial revie
 - Observed: 2026-10-12T20:59:57.926222+00:00
 - Observation: fd3ec2ab228f8706f4faba0d1df8141285ef447246e4e2f5bf6aa2c1868d2ae3
 
+## Reviewed reading
+
+The Grok week-42 brief is retained at `digests/external/weekly/2026-W42-grok.md` and stays unverified as a whole. The pass against ledger captures is publication `db9e3a7068329a605daee94cc0429706d0aedfaeb42c4f167b88d21483783339`.
+
+Included from captures: the 11 October OpenRouter capture lists neither `meta/muse-wonder`, nor `google/gemini-4-argon`, nor `meta/muse-spark-open`. The 12 October capture lists `meta/muse-wonder` at context 1048576, prompt `0.000004` and completion `0.00002` per token, and that file's only wonder id is `meta/muse-wonder`. The 13 October capture lists `google/gemini-4-argon` at prompt `0.000002` and completion `0.00001`, and the batch id at prompt `0.000001` and completion `0.000005`. The 15 October capture lists `meta/muse-spark-open` at context 1048576, prompt `0.00000125` and completion `0.00000425`; the 14 October capture does not list that id. The 17 October capture lists `deepseek/deepseek-v4.1-flash-1017` at prompt `0.00000015` and completion `0.0000006`, with weekday windows at prompt `0.0000003` and completion `0.0000012`; the 16 October capture does not list that id. The 18 October OpenRouter capture lists those same prompt and completion fields. The 18 October pricing capture lists Haiku 5.5 at $0.10 / $0.50 and $0.50 / $2.50 per million tokens and Sonnet 5.5 at $2 / $10 with cache read $0.10. Its visible text does not contain Argon, Muse, or DeepSeek. The Sonnet page text gained the line "Claude for Google Workspace". That line is outside the brief.
+
+Left out: the Muse Wonder Full launch story and the index scores, the Argon shipping story and the later $4 / $20 rate, METR access, the China draft, the workforce report, the Super Intelligence Force, the ARC Prize scores, the Spark Open parameter and hardware claims, the revenue figure, and the court and regulator items. Frontier, Efficient, and Budget are not assigned. Price readings are in `frontier/readings-2026-W42.md`.
+
 ## Editorial follow-up
 
-Review source claims, contradictions, frontier/price impacts, compendium edits,
-benchmark health, and evidence gaps. Draft generation is not fact verification.
+Source-update sections above remain unextracted change notices. This pass does not make the external brief a belief.

frontier/readings-2026-W40.md

@@ -0,0 +1,17 @@
+# Price readings — 2026-W40
+
+Read on 31 October 2026 from captures already in the ledger. These lines do not assign Frontier, Efficient, or Budget. They do not replace [ledger-2026-09.md](ledger-2026-09.md).
+
+Figures are USD per million tokens as shown on the 4 October capture of https://claude.com/pricing, capture `e154561c0dbc3fb6ec81c5ef3bfe485951d7d299541fd2852610f9b6c9454f04`.
+
+| Row on that page | Input | Output | Cache read | Cache write |
+| --- | --- | --- | --- | --- |
+| Fable 5.1 | $10 | $50 | $0.25 | $12.50 |
+| Opus 5.5 | $4 | $20 | $0.20 | $5 |
+| Sonnet 5.5 | $2 | $10 | $0.20 | $2.50 |
+| Sonnet 5 | $2 | $10 |  |  |
+| Haiku 4.5 | $1 | $5 | $0.10 | $1.25 |
+
+The same page states US-only inference at 1.1× for input and output tokens. It does not list Haiku 5.5.
+
+The 4 October OpenRouter capture `888d6bb7b47f4a07de9eaa66b536cd58209e687cd8b890b804fae5ceea5f0765` lists `openai/gpt-6-astra` at prompt `0.00001` and completion `0.00005` per token, and `openai/gpt-6-luna` at prompt `0.0000001` and completion `0.0000005`. Those are catalog fields, not an OpenAI price card.

frontier/readings-2026-W41.md

@@ -0,0 +1,18 @@
+# Price readings — 2026-W41
+
+Read on 31 October 2026 from captures already in the ledger. These lines do not assign Frontier, Efficient, or Budget. They do not replace [ledger-2026-09.md](ledger-2026-09.md) or [readings-2026-W40.md](readings-2026-W40.md).
+
+On 6 October, capture `50eb50a66a6c4770b065c379ccd56b96d1f68a7b98e6ed28874780d7cc506421` of https://claude.com/pricing lists Sonnet 5.5 prompt-cache read at $0.20 per million tokens and does not list Haiku 5.5.
+
+On 7 October, capture `16d63ca029d7b392c26fb6f73c5c6dbfe26f65edbceb5c5503caf3f72e0d58d9` lists the following USD per million tokens. The 11 October capture `6f7047207dcfa2c384603d0e71ec6d666789d20ee6d944288e4e74dbb44dbb7e` still shows the Haiku 5.5 rows and the Sonnet 5.5 cache-read rate.
+
+| Row | Input | Output | Cache read | Cache write |
+| --- | --- | --- | --- | --- |
+| Haiku 5.5, prompts ≤ 100K tokens | $0.10 | $0.50 | $0.01 | $0.125 |
+| Haiku 5.5, prompts > 100K tokens | $0.50 | $2.50 | $0.05 | $0.625 |
+| Haiku 4.5, legacy table | $1 | $5 | $0.10 | $1.25 |
+| Sonnet 5.5 | $2 | $10 | $0.10 | $2.50 |
+
+The page does not state a percentage run-cost reduction. The short-prompt Haiku 5.5 input and output rates are one tenth of the Haiku 4.5 rates on the same page.
+
+OpenRouter lists `anthropic/claude-haiku-5.5` from 7 October at prompt `0.0000001` and completion `0.0000005` per token, context 1000000, with a higher override from 100000 prompt tokens. That is a catalog field, not the Anthropic price card. The card is the pricing-page capture above.

frontier/readings-2026-W42.md

@@ -0,0 +1,15 @@
+# Price readings — 2026-W42
+
+Read on 31 October 2026 from captures already in the ledger. These lines do not assign Frontier, Efficient, or Budget. They do not replace [ledger-2026-09.md](ledger-2026-09.md), [readings-2026-W40.md](readings-2026-W40.md), or [readings-2026-W41.md](readings-2026-W41.md).
+
+OpenRouter prices below are catalog fields in USD per token. The per-million equivalents are that field times one million. They are not a vendor price card.
+
+On 12 October, capture `5253f5d4ed3ca3fa28eeb3cd604042e8e8dcad5dae50271b3092a0d662eb31fb` lists `meta/muse-wonder` as Meta: Muse Wonder, context 1048576, prompt `0.000004` and completion `0.00002` per token. That is $4 and $20 per million tokens. The 11 October capture `1f957c1af24ec80bdae7adedba3020dfbf28ac2ed11a75ed1ad09800df1bc5e5` does not list that id. The 18 October capture `48a9163cfeafb75b72c8132a9c9cc866e3d7ee7bb22fd02f6640839e68bb6fb4` still lists those two fields, and its only model id containing wonder is `meta/muse-wonder`.
+
+On 13 October, capture `34ead8146ca85f71cf27d55517e32e69df885203ac59f2610ad4547886532981` lists `google/gemini-4-argon` at context 1048576, prompt `0.000002` and completion `0.00001` per token ($2 and $10 per million), and `google/gemini-4-argon:batch` at prompt `0.000001` and completion `0.000005`. The 12 October capture does not list `google/gemini-4-argon`. The 18 October capture still lists those fields. The catalog row does not state an expiry date or a later rate.
+
+On 15 October, capture `c2babd6a7580610b1b3f625440a39b98fd678a2faa6a8be6a3bc62aef5d91dfb` lists `meta/muse-spark-open` at context 1048576, prompt `0.00000125` and completion `0.00000425` per token ($1.25 and $4.25 per million). The 14 October capture `8aa8cddc47ee1e901ba9fe9d5347c6486acdf95729921a3154864c7252b8d35e` does not list that id. The 18 October capture still lists those fields.
+
+On 17 October, capture `99f7dc881b11fe2f9f4136f6797bf884cea16cae53ae1e03a598035587c3b670` lists `deepseek/deepseek-v4.1-flash-1017` at context 1048576, prompt `0.00000015` and completion `0.0000006` per token ($0.15 and $0.60 per million). Some weekday UTC windows on that row are prompt `0.0000003` and completion `0.0000012`. The 16 October capture `1310d0ed25b4a6c39a102b735a3eb8e68b614893066c6b1cc734e09eaea815ce` does not list that id. The 18 October capture still lists the top-level prompt and completion fields.
+
+On 18 October, capture `1bfb5ccc51900ff8bbf06aba43093d56c5da63705910a436d6d4fa17fc2b85fe` of https://claude.com/pricing lists the Haiku 5.5 tiers and the Sonnet 5.5 rates already recorded for 11 October in [readings-2026-W41.md](readings-2026-W41.md): Haiku 5.5 input $0.10 and output $0.50 per million tokens for prompts of at most 100,000 tokens, and input $0.50 and output $2.50 above that, with cache read $0.01 and $0.05. Sonnet 5.5 is input $2, output $10, and cache read $0.10. The visible text of this capture does not contain Argon, Muse, or DeepSeek.