Table of Contents
Table of Contents
(owner-directed, queue items standing)
#
- Owner directive: no queue is ever closed and all sections keep updating indefinitely; the People and publications item marked done on 2026-08-29 is what had told the daily refresh to stop growing that category [deepseek-v4.1-flash]
- queue.md: the intro now states every item is standing and agents never mark one done, strike it through, or remove it; the Done block became a standing-work list keeping the landscape, model selection, context management, and People and publications items open [deepseek-v4.1-flash]
- AGENTS.md: the Article queue section and daily refresh step 3 state that items are standing, a pass completes work but never closes an item, and no section, category, or article is ever treated as complete [deepseek-v4.1-flash]
(daily refresh, all categories and section roots)
#
- Added notes LLM Gateway (model access, twentieth column, its announced 2026-10-15 devaluation seeded as the baseline price row), Delta (surfaces, thirteenth column, public beta 2026-09-16), Supermemory (memory, eleventh column), Serena (context engines, ninth column), CrowdStrike Falcon Guardian (control planes, tenth column), and Birgitta Böckeler (people and publications, thirty-first column) with matrix columns and index entries [glm-5.3-flash]
- Section roots: model-selection recorded gpt-5.6-terra and gpt-5.6-luna leaving OpenAI’s pricing page and the Kimi K3 host floor moving to $0.40/$13.00 at Relace; the ADE tracker corrected on three counts (JetBrains Air lists OpenCode, Copilot, and ACP; no longer desktop-only; OpenCode shipped its ACP adapter) [glm-5.3-flash]
- Price moves and pricing corrections: kimi-code’s K3 OpenRouter listing dropped to $0.72/$13.00 (Price history row); cmux’s cloud packaging restructured with dollar tiers unchanged; jcode’s pricing corrected to provider-API post-credit billing; glm-coding-plan’s billing mechanics corrected (1x peak, 0.5x off-peak); opencode-zen’s free roster moved to twelve models [glm-5.3-flash]
- Other corrections: OpenClaw’s v2026.8.35 release date corrected to 2026-10-02; the surfaces matrix’s Free-tier row corruption fixed (a ten-column row one cell short); the evaluation matrix’s Workshop maturity cell corrected to five months; the skills-sh leaderboard gained an audit flag for mattpocock’s code-review (Snyk High Risk) [glm-5.3-flash]
- Release and recency notes: onecli v2.7.0, pi v1.0.1 and v1.0.2, mimo-code’s October 21 V2.5 retirement, memex’s three October 3 releases, ten people-and-publications recency refreshes, Artificial Analysis at 690 models, Epoch past 3,600, LMArena’s 5-million-vote T2I post added, beads at 27,613 [glm-5.3-flash]
- AMO directory cross-check: every uncovered entry resolved in-run as an explicit rejection with reasons (Wave 3, Rusty Claw, NarraNexus, zachealy1/orchestrator, Companies.sh, handoff, Okto Nexus already covered, Culture, SwarmClaw); Offrun and Pi pod rejected as managers and sandbox runners rather than harnesses [glm-5.3-flash]
- Evaluation flag resolved: the Scaffold Effect study (arXiv 2607.22585) folded into frontierharness-eval as the independent academic counterpart, excluded from the model-benchmark-matrix by its model-only rule [glm-5.3-flash]
- Queue: a
? for tom: bullet asks whether E2B should be profiled as a sandboxing note; no self-directed essay, the six new notes filled the day [glm-5.3-flash]
- _index.md: all matrix and essay one-liners re-dated to 2026-10-04 and the six membership changes brought into their one-liners [glm-5.3-flash]
(owner-prompted, window audit of four items)
#
- Audited the four indexes previously marked done (landscape, model selection, context management, People and publications) for 2026-09-24 to 2026-10-04 [glm-5.3-flash]
- People material fixes, each with a Changes bullet: kent-beck gained a September 29 essay, latent-space a DevDay podcast episode, mario-zechner’s note corrected on pi v0.99.0 shipping codemode and built-in MCP, matt-pocock’s skills v1.3 merge, nathan-lambert’s Trillium Labs co-founding, andrej-karpathy’s restored Bear blog URLs, thorsten-ball’s Register Spill #102 [glm-5.3-flash]
- Other people fixes: ai-jason’s overstated uploads-tab claim rewritten, caleb-writes-code’s RSS record reversed, dex-horthy’s episodes marked aired, harper-reed’s as-of dates cured, jesse-vincent’s annotations corrected, peter-steinberger’s front matter cured; the Thorsten Ball cadence cell moved to issues 91 to 102 and Mario Zechner’s focus cell records the MCP softening [glm-5.3-flash]
- Rejected new voices, all with reasons: John Kutay, Tom Zick, Ari Weinstein, Alex Zhang, Nikunj Handa, Matthew Green, Maggie Appleton, and the DevDay presenters [glm-5.3-flash]
- Trackers: model-selection had two same-day drifts fixed; the landscape corrected Claude Code reading AGENTS.md natively since v2.1.277; fx’s stale newest-entrant superlative dropped; context-management-patterns recorded two material doc changes [glm-5.3-flash]
- Flags resolved in-run: the opencode note gained the GHSA-632h-h47v-g4x4 upgrade-path RCE (fixed in 1.18.22), and Whiteboard (YC W26) joined surfaces as a note and fourteenth matrix column [glm-5.3-flash]