🏆 Gold Nuggets · the new card format

All three real nuggets so far, rebuilt in the 6 AM format — headline, then what to do. Everything else is one tap away.
This doubles as the build reference: the Sunday digest must match this structure. Nothing goes above the headline — no methodology, no candidate counts, no run stats.
Nugget #003August 3, 2026

Your support inbox is IGL's design department — and the AI agent you were told to build is graded on closing it

Do thisBuild the customer-service agent this week as assigned, but ship it with a two-lane router. Lane 1 (has an order number, or matches where-is-it / return / sizing): auto-draft — roughly 90% of volume, exactly where Chris's five metrics apply. Lane 2 (no order number, or contains a story, an idea, or a prize claim): never auto-drafted, routed to you inside 24 hours with one standing question — "what's the story behind that?" Then add a sixth metric to Chris's five: ideas captured per 100 inbound emails. Your November 2025 baseline is one email to thirty-six designs.
Why this is true · evidence · the thread

The mastermind's new support agent is graded entirely on ending conversations quickly, and IGL's single most valuable inbound email of 2025 was one that should never have been ended.

Call 37 hands the cohort a support agent with three inputs — a brand-voice file, a policy document, and tool access — graded on five numbers: response time, satisfaction, decision authority, refund rate, save rate. Every one measures how fast and cheaply a conversation closed. None can see a message that was never a ticket.

Danny Meyer named this before AI existed: service is a monologue, hospitality is a dialogue. At his own inflection point he hired 51% for emotional intelligence and 49% for technical skill, because the technical half is the teachable half. Chris's agent is hired 100% on the 49%. That is correct at $90K a day. It is not obviously correct at IGL's roughly $354 a day.

On November 3rd, 2025 a Tomato Lovers Collective member sent an unsolicited note: some friends and I forage for mushrooms, maybe there's a line in that. No order number, nothing a policy document could touch. It became the Winter Foragers Collection — 36 designs, launched First Sunday of Advent. The same inbox produced "Life's too short to grow boring tomatoes" and four other collections, and it is where every giveaway winner must write to claim a prize. The loss would be silent: refund rate and satisfaction would both look excellent while that channel quietly closed.

Evidence trail
wescale-kb/calls/call-37-insights — the agent's three inputs and Chris's five close-the-ticket metrics
founders-kb/episodes/020-danny-meyer-setting-the-table — service vs hospitality, the FOR/TO test, the 51/49 hiring filter
igl-brand-kb/community/lauren-and-rowena-stories — the Nov 3 email that became 36 designs
recall-kb/Business/Agriculture/Lauren's TOMATO Awesomesauce!! (Nov 6, 2025) — prize claims route to the same address
wescale-kb/calls/call-37 — IGL's week-40 scale for comparison
Thread: call-37-insights → Founders ep-020 Danny Meyer → lauren-and-rowena-stories → recall-kb Lauren's Nov 2025 show → call-37
Gate: 5 documents · 4 knowledge bases · 5 primary · 0 summaries
Nugget #002August 1, 2026

The phrase library you design against was validated by people who have never bought anything

Do thisStart a 90-day Delight Log tomorrow, rules copied from Ross Gay: one delight per day, written fast in under 200 words, by whoever was in the garden that morning, never edited into brand voice. One markdown file per month. Then attach the gate that makes it matter: every design brief must cite a dated delight entry or a dated customer review — not a scraped phrase. Once a month, ask the Tomato Lovers Collective for the one small delight their garden handed them that week, and mine nothing until thirty come in. Re-run the review-language analysis at day 90.
Why this is true · evidence · the thread

IGL mines its design language quarterly from strangers on Reddit and X, and its own file records that the top "validated" phrases appear zero times in 551 real customer reviews.

Read only the brand base and the observation-mining method looks rigorous: a five-tool research matrix run quarterly, producing a table of "Validated High-Potency Phrases" headed by "My garden is my sanctuary" and "Sacred space," both marked Universal.

In the same file sits the actual customer baseline: 551 reviews, 0 uses of "sanctuary," "peace," or "sacred," 3 faith mentions, and a top identity phrase that sounds nothing like the mined language — "This IS my life from April to October!" The knowledge base reads that zero as an opportunity. The uncomfortable reading is the other one.

And the buyers talk exactly like Ross Gay — small, specific, physical, funny. One wrote that opening the package was "as exciting as getting seed mail"; a bird feeder outside a nursing-home window made a father-in-law "laugh so hard." The brand base names Gay only as a council voice and never records what he actually did, which is not an opinion but a production discipline: daily, by hand, under thirty minutes, unrevised.

Evidence trail
recall-kb/Philosophy/Spirituality/The Book of Delights by Ross Gay — the method, not the man
igl-brand-kb/methodology/observation-mining-methodology — the phrase table and the 551-review zero, in one document
igl-brand-kb/community/customer-reviews-q4-2025 — what buyers actually write
igl-brand-kb/voice/sanctuary-retreat-voice-guidelines — Gay filed as a voice, never converted into method
Thread: Book of Delights → observation-mining-methodology → customer-reviews-q4-2025 → voice guidelines
Nugget #001July 29, 2026

Your weekly design quota pays into two accounts — but only if the designs share a signature

Do thisRun a 90-day signature-constraint experiment on the quota you already produce. Week 1: pull your top 10% designs by revenue and distill 2–3 signature styles that fit the brand (botanical linocut; monocolor on Comfort Colors — both cohort-proven, both clear of the dark-humor guardrail). Weeks 2–13: mandate that at least 80% of the 24 weekly designs use one of them, leaving 20% exploratory. Score two things: hit rate and storefront conversion. Quarterly, run the parachute test — logo hidden, is it unmistakably IGL? The constraint is a prompt-template change, not new labor.
Why this is true · evidence · the thread

Style-constrained volume compounds in two banks at once — Meta's momentum bank and the brand-signature bank. Style-scattered volume pays into only one.

WeScale records a fact it under-explains: four of the cohort's growing members independently converged on one signature style each. Its explanation is operational — simple designs iterate faster. True, but that explains why volume gets cheaper, not why those stores convert better.

Founders supplies the mechanism. Steve Jobs' Paris lecture: the city has no single spectacular building — thousands of ordinary ones, all consistent, create the signature. Munger: a brand is compressed information. Five hundred designs in one visual language transmit that signal on every product tile; five hundred scattered ones transmit noise.

Neither base says this alone: WeScale has the cadence math (73% success at 10–15 ads/week) and no theory of identity; Founders has the identity theory and no cadence math. It is a live suspect for the actual symptom — volume +160% vs cohort, conversion 27% below median.

Evidence trail
wescale-kb/themes/simpler-design-wins — four members converging on one style each
founders-kb/themes/brand-building — the Paris test, brand-as-information
wescale-kb/themes/consistency-over-bursts — 73% at 10–15 ads/week vs 18% below 5
wescale-kb/igl-action-plan — +160% creatives, 1.6% vs 2.2% conversion
Note: mined under the ORIGINAL pairwise method — it leans on theme summaries, which is exactly what the rebuild fixed. Compare its evidence trail to #003's.