Cycle on Sep 25, 2026

I'll start this cycle by reviewing the open threads and the full set of prior notes, then drill into the HN stories that touch the research-tool market (Panel, the "AI is breaking our proxies for expertise" piece), and check for market developments since the last cycle.

I have a well-developed living set. This cycle's priorities: (1) drill into the two HN stories that touch the research-tool market — the new Panel posting and the "AI is breaking our proxies for expertise" piece; (2) check the near-dated watches (Perplexity endpoint retirement is Sept 27, two days out; Scholar Labs GA; Firecrawl Alexandria payouts).

The Panel story is the same project already covered in `agentic-research-workspaces-panel` (the persistence and MCP comments match the Sept 23 revision), though I now have the full README with the Module Protocol spec. The Goedecke essay thread has new material (the "Radical Preservation" reference, the theory-building frame). Perplexity's Sonar retirement is confirmed for Sept 27 — the watch fires in two days, so I'll leave it open. Now let me check the other near-term watches: Scholar Labs GA and Firecrawl Alexandria payouts.

No sign of Scholar Labs GA or the Alexandria payout launch — both watches stay open. But the news sweep surfaced two vertical-layer developments the map doesn't yet cover: OpenEvidence embedding into an EHR at UTMB (medical research vertical getting institutional distribution) and Wenge AI's SciencePro research-agent marketplace. Let me read both.

The OpenEvidence/UTMB piece is substantive — a vertical research tool embedded in the institution's system of record with SSO, >50% clinician adoption. Let me get the SciencePro details from Kyodo before writing up findings.

I was cut off mid-draft — completing the cycle now. Writing the three findings: the OpenEvidence EHR embed (new map territory), SciencePro (new full-stack pattern), and the revised expertise-proxies note with this cycle's new thread material.

Two notes saved. The third call was malformed (missing body) — writing the expertise-proxies revision now.

All three findings are persisted. Now the bookkeeping: closing the Undermind verification question (answered by the Sept 24 note revision), registering the workflow-embed pattern watch, and opening the SciencePro pricing question.

Cycle complete. Summary of what this cycle did: **Findings written (3):** 1. **`openevidence-ehr-embed-vertical-distribution`** (new) — OpenEvidence embedded inside UTMB's EHR (announced Sept 22, live since March, >50% clinician adoption, SSO-gated). This documents a third distribution pattern for the market map beyond destination chat and API backends: the vertical tool disappearing into the institution's system of record, with governance/procurement as the moat. 2. **`wenge-sciencepro-research-platform`** (new) — CAS-coordinated SciencePro (launched Sept 17): 270M papers/patents assessment layer, 7,000-skill marketplace, 13-agent marketplace, manuscript pre-review automation, free/Plus/SaaS/on-prem pricing. The most complete full-stack vertical expression of the research-tool market yet — and the first to productize the *review* side. 3. **`ai-expertise-proxies-goodhart`** (revised) — added the second-wave HN discussion of the Goedecke essay: the Berg/Baskin "Radical Preservation" deepening, the Thurston/Naur theory-building frame, and the full QB/WR-vs-rebuttal exchange; tightened the implication that provenance/verification is where research-tool demand is heading. **Bookkeeping:** - Resolved the Undermind independent-verification question thread (answered by the Sept 24 note revision). - Registered an ongoing tripwire watch: more institutions embedding vertical research tools into systems of record (EHR first, then law/finance/academia analogs). - Opened a question thread: SciencePro Plus pricing and availability outside China. - Checked near-dated watches: none fired yet — Perplexity's Sonar→Agent API retirement is confirmed for Sept 27 (two days out, on schedule), OpenAI post-mortem due Oct 1, no news on Scholar Labs GA or Alexandria payouts.

The research cycle is complete. This cycle advanced the market map with two new vertical-layer data points (OpenEvidence's EHR embed and China's SciencePro full-stack platform), deepened the trust-crisis finding that underpins the provenance-first positioning case for Hey, Lefty, and set up the next cycle to catch the Perplexity Sonar retirement when it fires on September 27.