Juniper Ventures · Esben · Aug 14 → Nov 14, 2026 · ~20 hrs/wk · day-by-day calendar →

3-month execution plan.

The execution layer under the OKRs and the 7 project lanes. Runs Juniper platform-infra AND the Neolab lab work in parallel — with the FAR probe→train-away experiment as the pivot that gates the lab and the 5-month fork.

The spine — the pivot is the FAR experiment. Everything bends around one question, answerable in ~6 weeks: does probe → train-away reduce misalignment in a way that generalizes (doesn't Goodhart the probe)? Signal → the lab is real, lean quit-to-launch. No signal → deepen Juniper toward full partner, rethink the wedge. First half builds toward the result (≈ W7); second half acts on it.

Filters: one bottleneck to its stop-condition before the next · measured by real progress, not activity · co-author, don't just execute.

Weekly hour budget (~20 hrs)

Streamhrs/wk
Lab / technical (the gate)~72 protected deep-work mornings. The co-authorship core.
Juniper platform-infra~6CRM, ship-list, deck — the infrastructure Juniper needs.
Sourcing + relationships~4Founder/LP calls, intros, 1:1s.
Content / writing~2Spikes to ~5 in an article-publish week.
Review / planning (Fri)~1Score OKRs, pick next week's one bottleneck.

Weekly cadence (respects sport 09:00 · 4–7 solo · 21:30 hard-stop · Fri = meta)

The 13 weeks

WeekThemeTop goals
Now · Aug 14–17Lock the baseTitle → LinkedIn/X launch · deck pass w/ Griff (P0) · contract FAR + probe access · reach Trenton Bricken
W1 · Aug 18–24Scope experimentDefine model + property + success metric + anti-Goodhart generalization test · stand up Sourcing CRM · deck in ≥1 LP convo
W2 · Aug 25–31Baseline + pipelineBaseline run · build train-away pipeline · research field-map v0 · draft Article 1
W3 · Sep 1–7Train-away #1Train-away run #1 (in-distribution) · CRM → 40+ · field-map v1
W4 · Sep 8–14🏊 Race week (light)70.3 Santa Cruz Sun 13/9 — protect it. Publish Article 1 · book talk-series #1 · sourcing calls only
W5 · Sep 15–21Generalization testAnti-Goodhart test: held-out probes + behavioral evals — hold or Goodhart? · CRM → 70+ · 20-country board longlist
W6 · Sep 22–28Iterate to gateIterate if needed · push David to cluster go/no-go · talk-series #1 run · ≥5 founders → first-meeting
W7 · Sep 29–Oct 5🚩 THE GATEVerdict: does probe→train-away generalize? The fork input. · CRM → 100 (O2 KR1)
W8 · Oct 6–12Publish resultArticle 2 = the experiment (flagship). No signal → honest "measurement-vs-optimization" piece · jobs/admin subdomain · deck in ≥5 LP convos
W9 · Oct 13–19Commit or deepenSignal: lab v0 as Enigma project + co-founder search · No-signal: deepen partner track · both: advisory-board outreach
W10 · Oct 20–26Institution / partnerSignal: per-country governance v0 + tool-AI prune spike · both: CEA Asia connect + board names
W11 · Oct 27–Nov 2WidenArticle 3 = coordination-institution / legitimacy · all portfolio founders 1:1 · talk-series #2
W12 · Nov 3–9The fork, w/ GriffFork conversation with Griff using the result: quit-to-launch vs full partner · met all core LPs · OKR mid-scores
W13 · Nov 10–14ConsolidateShip-list final state · 3-mo OKR scoring · retro + next-quarter cut

Article calendar (blog.kran.ai — one figure each, 80/20)

The one risk: at 20 hrs, the experiment (no external deadline) gets crowded out by Juniper fires. Protect the two deep-work mornings ruthlessly — they are the whole fork.