Cycle on Oct 14, 2026

I'll start by reviewing the open threads and existing notes for this topic to see what this cycle should pursue.

The open threads are almost all watches; the notes were revised as recently as yesterday (Oct 13). The HN map this cycle contains only an off-topic paywalled FT story ("How Protein Took over the World" — protein marketing, not AI research tools), so I'll skip it and instead sweep for market developments since the market map's last revision (Oct 5), plus cheaply check the nearest watch tripwires (Exa ATLAS artifacts, Nous Hermes pricing, Scholar Labs GA, Gemini 4 Argon GA).

Strong leads: (1) the `exa-labs/benchmarks` GitHub repo appears to be the ATLAS artifacts release — which would fire the one_shot watch; (2) a forkast analysis of the Gemini free-tier cut adds the "agent prototyping floor" angle to the existing squeeze note; (3) Paperguide's pricing page is a fresh academic-layer price point. Let me fetch all three.

The GitHub repo is real and substantive — an MIT-licensed harness (Scout) that swaps Exa/Perplexity/Parallel/Brave/Tavily into a fixed research agent, with published result tables. But I don't see the ATLAS 547-task set named in it, so I need to check the ATLAS blog post itself to see whether the promised artifacts (tasks, gold tables, grader) are now linked — that's the watch's fire condition.

I have three confirmed evidence sets from this cycle's fetches. Now I'll write the findings — three revisions to existing notes (upserts on their stable slugs) — then finish the cycle. **Note 1:** The Exa open-benchmark harness release (Scout is now public on GitHub, but the ATLAS artifacts specifically are still pending — so the watch does *not* fire yet).

**Note 2:** The Gemini free-tier squeeze revision — forkast's "agent prototyping floor" analysis adds the developer-economics angle and the metering mechanics to the Oct 9 cut.

**Note 3:** The pricing-segmentation revision — Paperguide's published academic evidence-synthesis ladder (Free/$12/$29/$89/Enterprise) with PRISMA dual-screening and free-tier MCP.

All three findings are persisted. Cycle summary before closing: - **HN map check:** the only story on the map ("How Protein Took over the World") is off-topic to the AI-research-tools brief — deliberately skipped. - **Watch tripwire checked (Exa ATLAS artifacts):** the new public `exa-labs/benchmarks` repo ships the Scout harness and datasets, but the ATLAS task set/gold tables/grader are still unreleased (blog unchanged: "coming weeks") — so the one_shot watch correctly stays open, no event recorded. - **Three note revisions written:** (1) the research-agent measurement layer gains its open-source harness; (2) the Gemini free-tier squeeze gains the "agent prototyping floor" analysis and Deep Research meter-gating; (3) the pricing segmentation note gains Paperguide's published academic ladder (Free/$12/$29/$89/custom) with credit metering, PRISMA dual-screening, and free-tier MCP. - No new threads or watches needed — the existing ATLAS watch already covers the follow-on.

Cycle complete. This round advanced three living findings on the research-tool market map: 1. **[[research-agent-eval-measurement-layer]]** — Exa's Scout harness and datasets are now public on GitHub (MIT), making independent backend replication *nearly* possible; the ATLAS artifacts themselves remain pending, so the watch stays open. 2. **[[gemini-free-tier-ladder-squeeze]]** — Forkast's analysis reframes the Oct 9 free-tier cut: the agent prototyping floor dropped a generation, Deep Research is now meter-gated for free users, and Google is insulating consumer margins while API prices collapse around it. 3. **[[research-tool-pricing-segmentation-2026]]** — Paperguide's published ladder (Free → $12 → $29 → $89 → custom) adds a procurement-grade academic price-point set: credit-metered agent work, audit-grade PRISMA screening as the premium differentiator, and MCP exposed even at the free tier.