Corrections & updates
Every material factual change to a published page is recorded here: what the page said, what it says now, and why. A correction means the page was wrong when it was published. An update means the facts changed afterwards and the page was re-verified. The rules behind this log are in the editorial policy.
9 entries: 7 corrections, 2 updates. Found an error? [email protected].
October 6, 2026 — Correction (1 page)
Corrected /answers/do-claude-console-api-credits-transfer-to-claude-app-2026/
- Was: Contact Anthropic support through the help centre about refunds rather than expecting a transfer.
- Now: Anthropic's help centre states that all credit purchases are non-refundable, so expect neither a refund nor a transfer.
- Why it happened: claim audit
- Source: https://support.claude.com/en/articles/8977456-how-do-i-pay-for-my-claude-api-usage
- Page: /answers/do-claude-console-api-credits-transfer-to-claude-app-2026/
October 6, 2026 — Correction (1 page)
Corrected /answers/best-ocr-pipeline-for-large-scale-data-extraction-2026/
- Was: Datalab reports that Chandra 2 processes about 2 pages per second on one H100 with 96 concurrent requests
- Now: Datalab measured Chandra 2 at 1.44 pages per second on one H100 (vLLM, 96 concurrent sequences) on a hard benchmark set and estimates about 2 pages per second on real-world documents
- Why it happened: claim audit
- Source: https://github.com/datalab-to/chandra
- Page: /answers/best-ocr-pipeline-for-large-scale-data-extraction-2026/
October 5, 2026 — Update (1 page)
Updated /reference/ai-model-pricing/
- Was: Gemini 3.6 Flash $1.50 input / $7.50 output per million tokens
- Now: Gemini 3.6 Flash $0.75 / $3.75 through December 31, 2026 (intro rate), then $1.50 / $7.50 from January 1, 2027
- Why it happened: claim audit
- Source: https://ai.google.dev/gemini-api/docs/pricing
- Page: /reference/ai-model-pricing/
October 5, 2026 — Update (1 page)
Updated /answers/agent-memory-vs-documentation-markdown-workspace-vs-vector-memory-2026/
- Was: Hacker News discussion at 57 points, 42 comments
- Now: 353 points and 237 comments as of October 5, 2026
- Why it happened: claim audit
- Source: https://news.ycombinator.com/item?id=49945933
- Page: /answers/agent-memory-vs-documentation-markdown-workspace-vs-vector-memory-2026/
October 5, 2026 — Correction (1 page)
Corrected /answers/cactus-whistle-vs-whisper-base-vs-moonshine-tiny-on-device-speech-to-text-2026/
- Was: Moonshine Tiny v2 released 2025
- Now: Moonshine Tiny v2 released February 2026
- Why it happened: claim audit
- Source: https://arxiv.org/abs/2602.12241
- Page: /answers/cactus-whistle-vs-whisper-base-vs-moonshine-tiny-on-device-speech-to-text-2026/
October 5, 2026 — Correction (1 page)
Corrected /posts/paperclip-review-open-source-ai-agent-company-orchestration/
- Was: 6.3K open issues
- Now: 2.7K open issues plus 3.6K open pull requests (GitHub's 6.3K figure counts both)
- Why it happened: claim audit
- Source: https://github.com/paperclipai/paperclip
- Page: /posts/paperclip-review-open-source-ai-agent-company-orchestration/
October 5, 2026 — Correction (1 page)
Corrected /answers/meta-muse-spark-solves-open-math-problems-six-papers-explained-october-2026/
- Was: Meta published the six papers on October 3, 2026
- Now: Meta published the six papers on October 2, 2026
- Why it happened: claim audit
- Source: https://research.meta.ai/blog/solving-open-research-problems-together
- Page: /answers/meta-muse-spark-solves-open-math-problems-six-papers-explained-october-2026/
August 6, 2026 — Correction (2 posts)
Removed two duplicate tool reviews (Cognee, published August 2, 2026, and Juggler, published August 6, 2026) that repeated reviews already on the site under near-identical titles.
- Was: Two tools each had two separate review posts.
- Now: One review per tool; the duplicate posts were deleted.
- Why it happened: The topic-selection step had no deterministic duplicate check; a 40-character truncation in the tracker let a repeat through.
- What changed in the process: A mandatory pre-publish dedup script (same GitHub repository = duplicate) now runs before any research starts.
July 22, 2026 — Correction (116 pages)
Corrected Claude API prices across 116 answer pages and posts.
- Was: Claude Opus 4.5–4.8 listed at $15/$75 per million tokens; Claude Fable 5 at $15/$75 on 7 pages; Fast Mode described as a discount on 5 pages.
- Now: Claude Opus 4.5–4.8 at $5/$25 per million tokens (the price since November 2025); Claude Fable 5 at $10/$50; Fast Mode is a 2× premium. Every derived cost-per-task figure on those pages was recomputed and each page re-dated.
- Why it happened: New pages took prices from our own earlier pages instead of the vendor's pricing page, so one stale figure propagated.
- What changed in the process: Prices now come only from a canonical, dated reference checked against vendor pricing pages; our own pages are never accepted as a source for a fact.