LEARNINGS — cross-sprint insight log

Append-only log of what Atelier learns from running sprints. Distinct from per-sprint station outputs (which describe one run) and from verticals/*.md (which distill replicable templates). This file captures meta-insights — patterns that change how Atelier itself runs.

Conventions

• One H2 per insight. Date prefix YYYY-MM-DD so chronology is scannable.
• Each entry has four blocks: From (sprints/stations), Insight, Apply to, Supersedes.
• Never delete. To revise: write a new entry that explicitly Supersedes: the old one. Keeps the reasoning trail intact.
• S7 of every sprint is responsible for appending at least one entry (or explicitly noting "no new learning" with reason).
• Martin can append entries by hand at any time.

---

2026-05-17 · Skill-chaining is worth the orchestration cost on Day 1

From · runs/2026-05-17-prearrival-final/stations/01-strategy.md (S1 of the final sprint) Insight · Chaining /office-hours (6 forcing questions) → /plan-ceo-review (mode SELECTIVE EXPANSION) on S1 produced a strategy doc with three named owner-operators ready to sign tomorrow, 15 LinkedIn-warm prospects, and a clean wedge framing — none of which appeared when S1 was a single-skill call. The verdict-22q score jumped from 4.4 to 9.3 between the first sprint and the second. Apply to · Every S1, regardless of vertical. Treat skill-chaining as default, not an upgrade. Token cost is real but cheaper than a wrong-target sprint. Supersedes · —

2026-05-17 · "Dead zone" is a transferable wedge pattern, not a hospitality story

From · runs/2026-05-16-prearrival/_input.json (the original wedge framing) + playbook/verticals/hospitality-prearrival.md Insight · The wedge that worked for PreArrival was not "we have a better hospitality tool" — it was "we own a window of customer attention that no incumbent owns". The pre-arrival 30 days is one such window. The pattern generalizes: post-purchase pre-delivery (e-commerce), post-trial pre-conversion (SaaS), post-application pre-decision (loans/visas). Every B2B vertical has a dead zone of emotional intent that the dominant tool deliberately leaves untouched. Apply to · S1 stress-test should explicitly probe for a "dead zone" framing before accepting any vertical idea. Add to verticals/README.md as a screening question. Supersedes · —

2026-05-17 · Founder asset score multiplies the wedge, doesn't add to it

From · runs/2026-05-17-prearrival-final/startup.json (martin_asset: "Masia 18mo data + LPB 40M€ + Decantalo ES network + LinkedIn brand") + the 9.3 billion_gate_score Insight · A wedge with no founder asset behind it caps the verdict-22q score below 7. A wedge backed by a lived dataset, a relevant operator network, and existing distribution multiplies the same wedge to 9+. Atelier should not treat founder-fit as a soft check — it changes the gate result by 2-3 points. Apply to · S1 prompts must require the agent to list 3 founder-side assets that compound the wedge, and reject sprints where the list is generic. Update playbook/S1-stress-test.md to make this explicit. Supersedes · —