◉AIAT 2.0 full draft written — 16 sourced figures + data + interactive graphs (Capitalist Case) 2026-08-19
○Genverify flagged capital numbers before distribution 2026-08-29↳ needs: AIAT 2.0 full draft written — 16 sourced figures + data + interactive graphs (Capitalist Case)
○Analyst review pass (Nick/Finn) on capital numbers 2026-09-05↳ needs: Genverify flagged capital numbers before distribution
○Fold 2.0 into aiat.report Docusaurus stack + wire interactive figures 2026-09-26↳ needs: Analyst review pass (Nick/Finn) on capital numbers
2026-08-20 12:06 — Full reconcile to current status. **P07 10→15%**: network build started — Meaning Dinner with the public-goods-funding circle + Tehama drop-ins; new person-notes filed (Renaissance Philanthropy/Tom Kalil, Funding the Commons, Protocol Labs/Brad Holden, Goodfire/Max Loeffler). juniper.xyz now proxied through Cloudflare with /deck + /dataroom 301s live. All other projects verified current (P01 72, P02 55, P03 0, P04 60, P05 35, P06 15, P08 70).
2026-08-20 (P02) — Reconciled AIAT 2.0: it's a **near-complete working draft** ("The Capitalist Case for AI Assurance"), not an outline — abstract + Prologue/Part I–III, 16 sourced figures, data JSON, interactive graphs (kran.ai/a/aiat-2-graphs), 7-move summary, 6-category 2.0 taxonomy, caveats. Remaining = genverify flagged capital numbers → analyst review (Nick/Finn) → fold into aiat.report Docusaurus stack → publish. P02 28→55%.
2026-08-20 — Week-1 catch-up: **juniper.xyz** rebuilt in Tehama editorial style + **cutover to Cloudflare Pages** (DNS GoDaddy→CF, live, Google email intact) → P02/P08. **Juniper Fellowship site** shipped (juniper-fellows.pages.dev / ideas.html "What needs building": 49 costed briefs + 258 directions, PDL-enriched career matcher, "why they care" field, Foundation-framing) → feeds P01 talent + P02 thought-leadership. **AIAT 2.0 graphs** (kran.ai/a/aiat-2-graphs) → P02. **David datacenter pitch** built (kran.ai/a/juniper-house-cluster + cluster-tiers, Blackwell B200 budget) + sent via Nick; walk-the-building Thu → P05 (ask finalized). **LP intelligence** deepened: METR-funder cross-ref + Eric-Ries network list → +7 net-new LP candidates written to CRM Deals (Astralis Foundation = top LP is a METR funder) → P04. S2 + PDL keys live → P01.
2026-08-16 — Engine v3: **founder-DNA lookalike scorer**. Learned the pre-founding signature of the founders behind Juniper's ~41 on-thesis assurance companies (dna.py + founders.json, 88 founders). KEY FINDING: 47/88 are **operators** (repeat/industry founders with NO papers) vs 20 researchers → OpenAlex-only was blind to over half the pattern. Added PDL career layer (pdl.py, key-gated): enrich (frontier-lab/repeat-founder — works in prod, enriched=14) + ex-frontier-lab operator sourcing (validated locally: ex-OpenAI/DeepMind founders dna 92). New fields: DNA score/archetype/Resembles; table now ranked by DNA (researchers correctly matched — Sharkey→Goodfire, Andriushchenko→Gray Swan). OPEN: PDL Person Search returns 0 in the Action (operators_sourced=0) — likely PDL search-credit/plan limit (separate quota from enrich); enrich is fine. Both S2 + PDL keys set as secrets + in the workflow yml.
2026-08-14 19:45 — Engine v2: added the **Founder-ripe** signal (Esben's ask: founders who've been too long at a company) — OpenAlex affiliation year-ranges → current employer + company tenure; flags company/lab tenure ≥4y. Added alignment labs (Anthropic/OpenAI/DeepMind/Redwood/Apollo/EleutherAI/…) to the corpus so it covers where safety talent works now, not just universities. Semantic Scholar corpus wired but key-gated (S2_API_KEY, added later) — widens discovery, resolves to OpenAlex for unified enrichment. +4 Airtable fields. Verified fields populate (Toronto/Vector tenure). Founder-ripe correctly 0 for the current all-academic set; fires for company/lab-tenured people as the corpus widens. Note: OpenReview is login-walled (403) and MATS /alumni is testimonials not a roster — reliable "where they work now" comes from OpenAlex last-known-institution + the tenure signal, not a scraped alumni join.
2026-08-14 19:10 — Engine v1 SHIPPED: repo Juniper-Ventures/sourcing-engine + weekly GitHub Actions cron + AIRTABLE_TOKEN secret + new Airtable "Talent Radar (engine)" table (tblsxo8b…). v1 = OpenAlex disambiguation (real author IDs, not names), co-author clusters, tenure + publication-cadence, pre-exit signals (productization / founder-timing / publication-slowdown / tight-cluster). HONEST finding: free bibliometric APIs cap signal — arXiv is hard rate-limited (429 on my IP AND GitHub runners); OpenAlex keyword = noisy; OpenAlex institution-filtered = clean but low-volume (sparse institution metadata). Table seeded w/ 6 high-signal top-institution candidates (Yaodong Yang, Xuehai Pan = real alignment researchers). Deliberately did NOT dump noise. v2 = better source (OpenReview API / Semantic Scholar key / arXiv bulk mirror).
2026-08-14 17:40 — LIVE Airtable access (PAT saved in ~/.zshrc). Found the base already has populated Sourcing (26f) + Watchlist (14f) tables + an active curated workflow — so the schema task was already done, not to-build. Proved engine→CRM wiring: deduped + wrote one tagged test candidate (David Demitri Africa, rec…PW65) into Sourcing. NOT bulk-importing the 184 raw candidates into the curated table — cadence/ownership to confirm with Griff/Nick.
2026-08-14 17:22 — Sourcing engine v0 SHIPPED: runnable sourcing_engine.py pulls 633 recent AI-safety papers from arXiv → 184 ranked author-candidates (first-authorship weighted) → /sourcing subpage + candidates.json. First working slice of the engine. (v0: name-only dedup; v1 = affiliation disambiguation + Apart/LessWrong/OpenReview feeds.)
2026-08-14 17:14 — Autonomous batch: reviewed Nick's CRM (Airtable base + crm-updater agent) → crm-review subpage w/ watchlist schema spec (unblocks P01 schema task); built LP-intelligence list (31 ranked LPs + pull method) → P04; mapped AI-safety talent programs → P01 sourcing feeds; built swag playbook ($5K BOM, Seldon→Juniper) → P06. GoDaddy + DoorDash done (Esben). LP-list validation done (Preqin/PitchBook cover raw data, none score AI-safety fit → thin scraper + scoring is the build). 4 subpages linked.
2026-08-14 16:56 — GitHub invite done. Research build → in progress (research.juniperventures.xyz still NXDOMAIN — DNS pending, domain landed today). Software task dates pulled in (AI/software leverage); LP-list fast-tracked to Sep 5. GPU provisioning + talks unchanged.