From 24a2c16ecb96c25e6597f149128e36fda92a9876 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 30 May 2026 04:54:50 +0000 Subject: [PATCH 1/5] =?UTF-8?q?Cycle=2024:=20VALIDATE=20=E2=80=94=20send-r?= =?UTF-8?q?eady=20discovery=20kit=20(both=20cohorts)=20+=20signal=20log?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Turn Cycle-19's drafted messages into operational artifacts: - DISCOVERY_KIT.md: two cohorts in parallel (FIX-1), neutral posture, call guide, signal rubric, decision-date playbook - SIGNAL_LOG.md: outreach tracker + per-call template + 0/3 GREEN tally - LEDGER: Cycle 24 entry; anti-loop active, board still 0/3 https://claude.ai/code/session_01A1BJYL2PvZh683Gh8KvuG8 --- DISCOVERY_KIT.md | 85 ++++++++++++++++++++++++++++++++++++++++++++++++ LEDGER.md | 22 ++++++++++++- SIGNAL_LOG.md | 69 +++++++++++++++++++++++++++++++++++++++ 3 files changed, 175 insertions(+), 1 deletion(-) create mode 100644 DISCOVERY_KIT.md create mode 100644 SIGNAL_LOG.md diff --git a/DISCOVERY_KIT.md b/DISCOVERY_KIT.md new file mode 100644 index 0000000..7ccbdb9 --- /dev/null +++ b/DISCOVERY_KIT.md @@ -0,0 +1,85 @@ +# DISCOVERY KIT — CASH-D1 Validation (send-ready) +> Cycle 24 (May 30, 2026) · the gating step for the ENTIRE board. **Buyer signals: 0/3.** +> Decision date: **2026-07-02 (33 days out).** Goal: run **3–5 calls** and get **≥3 STRONG signals**. +> Posture: **NEUTRAL** (§5) — never pitch a product; ask what they DID; let WTP-type reveal itself. +> Read with `BUILD_PLAN_CASH-D1.md` (the idea) + `LEDGER.md` Cycle 19/22 (why this framing). + +--- + +## 0. WHAT WE ARE TESTING (one conversation tests all of it) +- **VoD #1 — recurring WTP (cash leg):** will an FI pay an *ongoing subscription* to keep the 책무구조도 current + hold the 소명 evidence — not just buy a one-time map? +- **VoD #2a — consolidation (moat leg):** would they collapse multiple compliance vendors into one? +- **VoD #2b — data consent (moat leg):** would they let their posture join an *anonymized* cross-client benchmark? +- **Q0 — the board's shared bear (L6):** does pre-enforcement **software**-WTP exist, or only **consulting**-WTP? (If only consulting → Law L9 / managed-service entry posture.) + +**Rule (§5):** pain must surface BEFORE any product is named. A polite "유용하겠네요" = ZERO. + +--- + +## 1. TWO COHORTS, RUN IN PARALLEL +Per Cycle-22 **FIX-1**: the Jul-2 deadline is for the *one-time* job we DON'T sell. So we lead recurring-WTP +discovery with firms already living the maintenance pain, while keeping the deadline cohort as hot pipeline. + +| | **Cohort A — already-submitted** (best teacher for VoD #1) | **Cohort B — Jul-2 mid-tier** (hottest urgency / sales pipeline) | +|---|---|---| +| **Who** | 은행 (filed Jan-2025), 대형 금투/보험 (filed Jul-2025) | 저축은행 ≥₩7,000억, 캐피탈/카드 ≥₩5T 여전사, <₩5T 자산운용·증권 | +| **State of mind** | 6–18 months into *maintenance* pain; no deadline-consultant competing for attention | scrambling on the *initial* submission, 33 days out | +| **Tests best** | recurring WTP, re-papering pain, 소명 evidence gap (VoD #1) | urgency, who they hired, what it cost (context, not the sale) | +| **Channel** | Coxwave FI clients · Ace's dad's broader network | Ace's dad → Yeouido mid-tier 자산운용사 | +| **Lead with** | "submission's done — what happens now?" | "how's the 7/2 prep going?" | + +Target mix across 3–5 calls: **at least 2 from Cohort A** (they answer VoD #1), the rest Cohort B. + +--- + +## 2. OUTREACH MESSAGES (Korean, ready to paste — DO NOT pitch) + +### Cohort A · already-submitted — via Coxwave (or dad's network) +> [담당자]님, 금융권 고객사 중 **이미 책무구조도를 제출하신** 회사의 준법감시·내부통제 담당자분과 15분만 이야기 나눠볼 수 있을까요? 제품 영업이 전혀 아니고, **제출 이후에 그 자료를 실제로 어떻게 유지·관리**하고 계신지(인사이동·조직개편 때 등) 리서치 중입니다. 소개 가능하시면 제가 직접 짧게 연락드리겠습니다. + +### Cohort A · already-submitted — direct (warm intro version) +> 안녕하세요 [성함]님, [소개자]님 소개로 인사드립니다. 금융회사 내부통제·책무구조도 실무가 **제출 이후에** 실제로 어떻게 돌아가는지 공부하고 있는데, 15분만 시간 내주실 수 있을까요? **무언가 팔려는 게 전혀 아니고**, 제출 끝난 뒤 유지·관리 업무가 현업에서 어떻게 이뤄지는지 듣고 싶습니다. 편하신 시간 알려주시면 맞추겠습니다. + +### Cohort B · Jul-2 mid-tier — via Ace's dad (KR) +> 안녕하세요 [성함]님, [아버지 성함]의 아들 염[이름]입니다. 아버지 소개로 인사드립니다. 요즘 금융회사 내부통제·책무구조도 실무가 실제로 어떻게 돌아가는지 공부하고 있는데, 혹시 15분만 시간 내주실 수 있을까요? **무언가 팔려는 게 전혀 아니고**, 7월 2일 제출 기한을 앞두고 현업에서 어떤 작업을 하고 계신지 듣고 싶습니다. 편하신 시간 알려주시면 맞추겠습니다. 감사합니다. + +### Cohort B · Jul-2 mid-tier — via Coxwave (KR) +> [담당자]님, 금융권 고객사 중 준법감시·내부통제 담당자분과 15분 정도 이야기 나눠볼 수 있을까요? 제품 영업이 아니라, 책무구조도/AI 컴플라이언스 관련해서 실제로 어떤 업무를 **수동으로** 하고 계신지 리서치 중입니다. 소개 가능하시면 제가 직접 짧게 연락드리겠습니다. + +--- + +## 3. CALL GUIDE (ask what they DID — pain BEFORE product) +**Open neutral.** For Cohort A lead with Q4 first (the maintenance reality); for Cohort B run in order. + +0. *(cross-board diagnostic — ask early, neutrally)* 컴플라이언스·내부통제 관련해서 **구독형 소프트웨어를 정기 결제로 쓰고 계신 게 있나요?** 어떤 거고, 왜 도입하셨어요? ← *Q0: software-WTP vs consulting-WTP* +1. 지난 3개월간 책무구조도 준비로 **실제 하신 작업**을 순서대로 말씀해 주실 수 있나요? +2. 그중 제일 손 많이 가고 반복되는 부분은 어디였나요? +3. 외부(로펌/컨설팅)에 맡긴 부분이 있나요? 비용은 어느 정도였나요? +4. 제출 끝나면 그걸로 끝인가요, 아니면 **인사이동·조직개편 때마다 다시 손봐야** 하나요? 그건 누가 하나요? ← *VoD #1 (recurring WTP)* +5. "대표이사 관리의무 이행" 증빙은 지금 어떻게/어디에 남기세요? ← *the 소명 gap* +6. 컴플라이언스 관련 외부 솔루션·컨설팅을 몇 개나 쓰세요? 한 군데로 합치고 싶다는 생각 해보신 적 있나요? ← *VoD #2a (consolidation)* +7. 동종 업계 익명 벤치마크(다른 회사들은 이 항목을 어떻게 처리했는지)가 있다면 유용할까요? 그러려면 귀사 데이터도 익명으로 풀에 들어가야 하는데 가능할까요? ← *VoD #2b (data consent)* + +**Closing (only AFTER pain surfaces, never as the opener):** "혹시 이런 부분을 도와주는 도구를 만들고 있는데, 한번 보여드려도 될까요?" → if yes, show the Phase-0 aid. *Their reaction is a signal, not the goal.* + +--- + +## 4. SIGNAL RUBRIC (§5 — what counts toward the 3 needed for GREEN) +- 🟢 **STRONG (= 1 signal):** unprompted "이런 거 시스템/자동화 없나요?", "얼마예요?", "언제부터 쓸 수 있어요?", OR spontaneously describes ongoing-maintenance / 소명 pain *before* any product is named. +- 🟡 **WEAK (≈0, log but doesn't count):** agrees it's painful only *after* I describe a solution. +- ⚪ **ZERO:** "유용하겠네요" / polite interest, no described pain, no buying question. + +→ Log every call in `SIGNAL_LOG.md` immediately after, verbatim where possible. + +--- + +## 5. DECISION-DATE PLAYBOOK (2026-07-02) +- **≥3 STRONG** → promote CASH-D1 to active BUILD (Phase 1); sign a design partner. +- **<3 but pain confirmed** → extend validation 30 days; build only the Phase-0 aid. +- **Pain absent / shrugs** → trust buyers over desk research; down-weight 책무구조도; check the MOAT-A1 activate-trigger (is security-governance 소명 the real #1 pain?). +- **Q0 says "only consulting-WTP"** → shift to the L9 managed-service entry posture (David sells the deliverable, Ace's software is the margin engine). + +--- +*Only the founders can execute this. Ace sends Cohort-B msgs via dad + the Coxwave intro; pull Cohort-A +intros via Coxwave's already-live FI clients. Run 3–5 calls before 2026-07-02. This kit exists so sending +is a copy-paste, and so every call is logged the same way. The binding constraint is buyer contact — go.* diff --git a/LEDGER.md b/LEDGER.md index 049254c..b461054 100644 --- a/LEDGER.md +++ b/LEDGER.md @@ -1,6 +1,6 @@ # VENTURE LEDGER > Living state of the Founder's OS. Updated every cycle. Read alongside `CLAUDE.md`. -> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 | **OS version:** 2.2 +> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 24) | **OS version:** 2.2 --- @@ -75,6 +75,26 @@ Ran the G6 anti-gaming math on the narrowed segment *on its own*: --- +## CYCLE 24 — VALIDATE: send-ready discovery kit + signal log (no generation) · May 30, 2026 +**Why this cycle:** anti-loop is active and the board is blocked on the SAME constraint it has been for 5 +cycles — buyer signals 0/3. So this cycle produces the operational artifacts that turn "messages drafted" +(Cycle 19) into "messages a founder can send in one paste, and log identically." No new ideas, no new gates. +**What changed vs Cycle 19's kit:** +- Folded in Cycle-22 **FIX-1**: split discovery into **two cohorts run in parallel** — Cohort A (already- + submitted: 은행/대형 금투·보험) as the real teacher for recurring-WTP (VoD #1), Cohort B (Jul-2 mid-tier) + as the hot pipeline. Added a Cohort-A outreach message (lead = "submission's done — what now?"). +- Kept posture **NEUTRAL** (operator decision, May 30): Q0 still diagnoses software- vs consulting-WTP, but + the kit does NOT pre-commit to the L9 managed-service framing — buyers decide it. (Faithful to §5.) +- Externalized the **signal rubric + tally** into `SIGNAL_LOG.md` so the 3-of-3 GREEN count is live state. +**Artifacts:** `DISCOVERY_KIT.md` (send-ready, both cohorts, call guide, decision-date playbook) + +`SIGNAL_LOG.md` (outreach tracker + per-call template + running tally). +**Status:** 🔬 VALIDATE · buyer signals still **0/3** — cannot move until founders send + converse. +**Operator action required (only founders can do this):** Ace sends Cohort-B msgs via dad + the Coxwave +intro; pull Cohort-A intros via Coxwave's already-live FI clients; run 3–5 calls before 2026-07-02; log +each in `SIGNAL_LOG.md`. **This is the gating step for the entire board.** + +--- + ## CYCLE 23 — META-LOOP: pressure-test L9 (no generation) · May 30, 2026 Tested the law I added last cycle before letting it harden into doctrine. Desk evidence is appropriate here because L9 is a claim about *market structure* (do FIs buy managed services?), not buyer-WTP for our specific product. diff --git a/SIGNAL_LOG.md b/SIGNAL_LOG.md new file mode 100644 index 0000000..7c22a3b --- /dev/null +++ b/SIGNAL_LOG.md @@ -0,0 +1,69 @@ +# BUYER-SIGNAL LOG — CASH-D1 (and the shared FI-accountability engine) +> The §5 first-class input. Desk research CANNOT produce a GREEN — this file can. +> **GREEN requires ≥3 STRONG signals.** Decision date: **2026-07-02.** Log every call here, immediately. +> Rubric (from `DISCOVERY_KIT.md` §4): 🟢 STRONG = 1 · 🟡 WEAK = log-only · ⚪ ZERO. + +--- + +## TALLY +| Metric | Count | +|---|---| +| 🟢 STRONG signals (need 3) | **0 / 3** | +| 🟡 WEAK (logged, non-counting) | 0 | +| ⚪ ZERO | 0 | +| Calls completed | 0 | +| Messages sent | 0 | +| Days to decision date | (2026-07-02 − today) | + +**Status:** 🔬 VALIDATE · YELLOW (cannot move until signals arrive). + +--- + +## OUTREACH TRACKER (sent messages — fill as you go) +| # | Date sent | Channel | Cohort (A/B) | Target firm / persona | Reply? | Call booked? | +|---|-----------|---------|--------------|-----------------------|--------|--------------| +| 1 | | Ace's dad | B | Yeouido mid-tier 자산운용사 · 준법감시인 | | | +| 2 | | Coxwave | A | already-submitted FI · 내부통제 담당 | | | +| 3 | | | | | | | +| 4 | | | | | | | +| 5 | | | | | | | + +--- + +## CALL LOG (one block per conversation — copy the template) + +### CALL TEMPLATE (copy below this line for each call) +``` +### CALL #__ — [date] — [firm type] — Cohort [A/B] +- Persona: [준법감시인 / 내부통제 / 소비자보호 총괄 / CEO·board / etc.] +- Channel / intro by: [dad / Coxwave / ...] + +VoD #1 — recurring WTP (re-papering on personnel/org change; who does it; is it ongoing?): + > [verbatim quote] + → reading: [STRONG / WEAK / ZERO] + +VoD #2a — vendor consolidation (how many tools/consultants; want to merge?): + > [verbatim] + → reading: + +VoD #2b — data consent (anonymized cross-client benchmark — would they join the pool?): + > [verbatim] + → reading: + +Q0 — software-WTP vs consulting-WTP (do they pay a recurring SaaS subscription today? which/why?): + > [verbatim] + → reading: [software-WTP exists / only consulting-WTP → L9] + +소명 gap (how/where they keep CEO 관리의무 이행 evidence today): + > [verbatim] + +Strongest unprompted line (the money quote): + > [verbatim] + +OVERALL: 🟢 STRONG | 🟡 WEAK | ⚪ ZERO (counts toward GREEN? Y/N) +Next step / follow-up: +``` + +--- +*After each call: update the TALLY, the OUTREACH TRACKER, and append a filled CALL block. When STRONG hits +3, promote per `BUILD_PLAN_CASH-D1.md` §7. When in doubt about a rating, trust the rubric — "유용하겠네요" is ⚪.* From ce09ffb5c235bb76a7ed636a14405278c69e2983 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 30 May 2026 07:38:14 +0000 Subject: [PATCH 2/5] Cycles 25-26: reject FI proactive-SaaS hypothesis, OS v2.3, pivot to EXPAND-1 Cycle 25 (META-LOOP, operator-triggered): - Evidence rejects 'mid-tier FIs proactively pay recurring SaaS pre-enforcement' (2yr grace period / no teeth til ~2028; compliance SaaS is fine-PULL not hygiene-PUSH). Entire FI engine demoted to PARK with revive-trigger. - Methodology fixes -> OS v2.3: G9 WTP-TRIGGER gate + T7 kill-type, sharpened L6, new L10 (deploy idle moats), CONCENTRATION CHECK, mandatory COMP-SCAN. Cycle 26 (GENERATE, new hunting ground): - EXPAND-1: AI store-revenue forecasting + next-site selection for KR multi-unit/franchise operators. Deploys David's site-selection IP (idle 24 cycles). Passes G2 (real moat) + G9 (ROI-PULL, enforcement-free); proven WTP comps. Gated + hardened + bull/bear. New board lead. https://claude.ai/code/session_01A1BJYL2PvZh683Gh8KvuG8 --- CLAUDE.md | 103 +++++++++++++++++++++++++++++++++--------------- LEDGER.md | 114 +++++++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 185 insertions(+), 32 deletions(-) diff --git a/CLAUDE.md b/CLAUDE.md index ed968f0..85bbe95 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -1,4 +1,4 @@ -# THE FOUNDER'S OPERATING SYSTEM (v2.2) +# THE FOUNDER'S OPERATING SYSTEM (v2.3) ### An agentic protocol that HUNTS, HARDENS, VALIDATES, and COMMITS to an idea — and improves its own methods. ### For: Ace Yeom + David Mun. Self-contained — hand to any AI/instance and it runs. Supersedes v1.0. ### v2.0 changelog: added forcing function, symmetric judgment, buyer-signal-as-input, anti-gaming check, @@ -7,6 +7,15 @@ ### (customer-handed, siloed), NOT scale-aggregated data (owned by 캐시노트-type incumbents). ### v2.2 changelog (cycle 22): added L9 (service-wrapped-software entry for enforcement-gated markets with ### unproven software-WTP) after a re-stress-test found the whole FI board shares ONE bear (L6 software-WTP). +### v2.3 changelog (cycle 25, OPERATOR-TRIGGERED): the bear in v2.2 was treated as "validate later" instead +### of gated — so 5 cycles over-concentrated on one engine while a documented TRUE MOAT (David's site-IP) sat +### unused. Fixes: (1) NEW GATE G9 WTP-TRIGGER (enforcement-PULL vs hygiene-PUSH; push-WTP w/o a forcing +### function is presumed ABSENT, max YELLOW); (2) refined L6 (grace-period/시범운영/한시운영 = no teeth = PUSH, +### why-now is a mirage); (3) NEW L10 (an unused named TRUE-MOAT = a generation failure; rotate an axis to +### deploy each moat); (4) CONCENTRATION CHECK in §7 (≥3 survivors sharing one bear ⇒ that bear is a PORTFOLIO +### risk ⇒ STOP same-engine generation, diversify); (5) mandatory COMP-SCAN in GENERATE/GATE (find the global +### analog + local substitute as EVIDENCE before gating). Evidence that the FI 책무구조도 cash leg is PUSH/no- +### teeth-til-2028 → entire FI engine demoted 🔬→🟡 PARK; lead is now EXPAND-1 (site-selection, MODE-MOAT). ═══════════════════════════════════════════════════════════════ ## 0. PRIME DIRECTIVE & STOP/COMMIT RULES (read first — fixes the "kill machine" flaw) @@ -52,23 +61,34 @@ AXIS-C data-byproduct (service that generates uncopyable data) · AXIS-D buyer-b buyer, find what they overpay for/do in spreadsheets) · AXIS-E enforcement-shock watch (standing task). OUTPUT = one sentence: "[BUYER] pays for [JOB] today via [SUBSTITUTE]; we do [WEDGE]; defensible because [MOAT SENTENCE]; fits [MODE]; first revenue in [TIME] within [BANDWIDTH]." Can't write it → not ready. +**MOAT-DEPLOYMENT SWEEP (v2.3, L10): before picking the axis, list every named TRUE-MOAT asset (edge map) +and check which are UNUSED across recent cycles. An unused moat is a generation failure — bias the axis to +deploy it (e.g., David's site-IP → AXIS-B asset-pair). Don't keep mining a hunting ground where the moat is +weak while a strong moat sits idle.** ═══════════════════════════════════════════════════════════════ ## 3. GATE — SYMMETRIC (fixes "biased toward killing" + "gameable") ═══════════════════════════════════════════════════════════════ -For EACH idea run BOTH searches, equal effort: +For EACH idea run BOTH searches, equal effort. **COMP-SCAN FIRST (v2.3, mandatory, EVIDENCE not armchair): +find (i) the GLOBAL analog — does this category exist & get paid for elsewhere? what's the price band & +proven WTP trigger? and (ii) the LOCAL incumbent / free substitute. No comp-scan → gating is just opinion.** (a) KILL-SEARCH: who owns this / why does it fail? (b) WIN-SEARCH: under what real, specific conditions could a 2-person team take a profitable slice? KILL ONLY IF win-conditions are ALSO absent. Then check gates in order; one fail = stop + log kill-type. G0 hunting-ground (foreigner-proof?) · G1a budget exists · G1b budget winnable · G2 named TRUE-MOAT asset (not "relationships") · G3 moat type (not data-scale/GPU/capital wall) · G4 disconfirm survived · G5 distribution <90-day test path · G6 ceiling (per MODE) · - G7 why-now catalyst · G8 BANDWIDTH-FIT (path to revenue fits ~10-13 hr/wk now?). + G7 why-now catalyst · G8 BANDWIDTH-FIT (path to revenue fits ~10-13 hr/wk now?) · + **G9 WTP-TRIGGER (v2.3): is revenue enforcement/ROI-PULL (a live fine, exam-with-teeth, deadline-with-teeth, + or a quantified loss-event compels the buy) or hygiene-PUSH (buyer pays proactively for good practice)? + PUSH-WTP without a forcing function is PRESUMED ABSENT → cap at YELLOW + flag high-risk; desk research + CANNOT grant it (only a buyer signal can). Grace-period / 시범운영 / 한시운영 / no-teeth-yet = PUSH, not + PULL — the why-now is a mirage. Kill-type for a pure-PUSH idea with no signal = T7.** ANTI-GAMING CHECK (mandatory after any segment-narrowing): "Does the narrowed segment that saved G4 still pass G6 (ceiling) ON ITS OWN?" If narrowing to dodge competitors breaks the ceiling, the idea is GAMED — mark it and either widen (and re-face competitors) or accept it as MODE-CASH/niche only. KILL-TYPES: T1 commodity · T2 incumbent-owns-wedge · T3 capital/license/data wall · T4 ceiling-low · - T5 wrong-team/no-distribution · T6 bandwidth-incompatible-now. + T5 wrong-team/no-distribution · T6 bandwidth-incompatible-now · T7 pure-hygiene-PUSH-WTP-no-forcing-function. ═══════════════════════════════════════════════════════════════ ## 4. HARDEN (make survivors stronger; same loop, now adversarial) @@ -110,6 +130,11 @@ STATUSES: 🟢 BUILD · 🔬 VALIDATE · 🟡 PARK(+revive-trigger) · 🔴 KILL LEDGER row per idea: name · one-sentence · mode · axis · gate(+kill-type) · harden-shape · bull/bear · buyer-signals(n) · money/exec · status · validate-or-die · revive-trigger. CROSS-IDEA SYNTHESIS every 3-4 ideas: same-buyer? same-moat-shape? convergence? repeatable-edge? +**CONCENTRATION CHECK (v2.3): if ≥3 surviving ideas share the SAME load-bearing bear, that bear is now a +PORTFOLIO risk, not a per-idea watch-risk. STOP generating same-engine ideas → either (a) VALIDATE the bear +cheaply, or (b) DIVERSIFY into a new hunting ground (rotate axis + deploy a different moat). "Convergence on +one engine" can be a strength signal OR a single-point-of-failure — distinguish them by asking: is the shared +moat strong (converge) or is the shared BEAR fatal (diversify)? (Cycle 17→24 failed this: 5 survivors, 1 bear.)** **META-LOOP (the framework improves itself — run every 8 ideas or monthly):** 1. HIT-RATE AUDIT: which gates did the actual killing? Which NEVER fired (dead weight → cut them)? @@ -121,17 +146,22 @@ CROSS-IDEA SYNTHESIS every 3-4 ideas: same-buyer? same-moat-shape? convergence? The methods are NOT fixed. This loop is how they change. Nothing here is sacred except the Prime Directive. ═══════════════════════════════════════════════════════════════ -## 8. THE 8 LAWS (current best; the META-LOOP may rewrite any of them) +## 8. THE 10 LAWS (current best; the META-LOOP may rewrite any of them) ═══════════════════════════════════════════════════════════════ L1 build the accountability layer, not the commodity primitive · L2 loud pain in a mature market = served · L3 warm channel ≠ winnable without buildable product · L4 durable moat = liability-entity + data-byproduct- of-a-regulated-SERVICE (customer-handed/siloed, NOT scale-aggregated — refined v2.1) · L5 own the unglamorous -workflow regulation creates, not the headline tool · L6 no teeth/grace -period = no WTP yet; need enforcement-sensitive buyer · L7 SEED≠MOAT; write the moat sentence · L8 -hardware/build skill = capability, not moat · L9 (provisional, v2.2; calibrated c23) enforcement-gated market + -unproven software-WTP but proven service-WTP → enter as service-wrapped-software (managed service). Precedent: -KR FIs already buy MSSP/RegTech on subscription. Boundary: managed service triggers 업무위탁 규정 + cannot -replace the internal 준법감시인 (it supports, not replaces); ACV still buyer-unvalidated. +workflow regulation creates, not the headline tool · **L6 (sharpened v2.3) no teeth / grace period / 시범·한시 +운영 = no PULL-WTP yet → revenue would rest on hygiene-PUSH, which is presumed ABSENT (see G9); a why-now built +on a teeth-less deadline is a MIRAGE. Need an enforcement-sensitive buyer with a LIVE forcing function.** · +L7 SEED≠MOAT; write the moat sentence · L8 hardware/build skill = capability, not moat · L9 (provisional, +v2.2; calibrated c23) enforcement-gated market + unproven software-WTP but proven service-WTP → enter as +service-wrapped-software (managed service). Precedent: KR FIs already buy MSSP/RegTech on subscription. +Boundary: managed service triggers 업무위탁 규정 + cannot replace the internal 준법감시인 (it supports, not +replaces); ACV still buyer-unvalidated. · **L10 (new v2.3) a documented TRUE-MOAT asset left UNUSED across +cycles is a generation failure, not neutral: every cycle's GENERATE must run the moat-deployment sweep and +bias an axis toward deploying an idle moat. (Trigger: David's site-selection IP sat unused 24 cycles while +the board mined a moat-less FI hunting ground.)** (Each Law is provisional. If the META-LOOP finds a counter-example, it gets rewritten with a logged reason.) ═══════════════════════════════════════════════════════════════ @@ -141,7 +171,21 @@ MON pick MODE(s) + generate from rotating axis · TUE symmetric gate (kill+win s WED harden + bull/bear adversary · THU score + ledger + (every 8) META-LOOP audit · FRI VALIDATE: send 3-5 buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. -## CURRENT STATE (updated cycle 23 — May 30 2026; full detail in LEDGER.md) +## CURRENT STATE (updated cycle 26 — May 30 2026; full detail in LEDGER.md) +🧭 CYCLE 26 = GENERATE (new hunting ground, anti-tunnel). NEW LEAD: **EXPAND-1** — AI store-revenue forecasting + & next-site selection for KR multi-unit / franchise operators. Axis-B (asset-pair: David's site-selection IP + × Ace's AI/full-stack), MODE-MOAT w/ a MODE-CASH per-site-report leg. WHY IT'S HEALTHIER THAN THE FI ENGINE: + G2 finally has a real moat (David's site-IP); G9 passes as ROI/loss-PULL (a bad 출점 = real $ lost, enforcement- + independent); proven WTP (KR franchises already pay 맥세스/유재은-type consultants; global comps SiteZeus/Buxton/ + Placer.ai = paid category, $200/mo→$30K+/yr). Survives gate (contingent). Validate-or-die: (1) will a KR 본사 pay + for brand-specific AI forecasting vs their consultant? (2) is David's IP productizable or tacit? 0/3 signals. +🔴 CYCLE 25 = META-LOOP (operator-triggered). REJECTED the FI engine's load-bearing hypothesis ("mid-tier FIs + proactively pay recurring SaaS pre-enforcement") on EVIDENCE: 책무구조도 제재 has a 2-yr 한시-운영 grace + sanctions + only post-submission ⇒ no teeth til ~2028; 증권사 call it 공시 부담; compliance SaaS globally is fine-PULL not + hygiene-PUSH. → ENTIRE FI ENGINE (CASH-D1/LEAD-1/MOAT-A1/MOAT-D1/DATA-C1) demoted 🔬→🟡 PARK. Revive-trigger: + a real buyer signal from the (cheap, still-valid) discovery kit, OR ~2028 grace-expiry/exam-wave, OR Q0 proves + software-WTP. Methodology fixes → OS v2.3: G9 WTP-TRIGGER gate, T7 kill-type, refined L6, new L10, CONCENTRATION + CHECK, mandatory COMP-SCAN. The framework's own blind spot (24 cycles ignoring David's site-IP) is now gated. 🔁 CYCLE 23 = META-LOOP. Pressure-tested L9 before it hardened: CONFIRMED via MSSP/RegTech subscription precedent (FIs do buy managed services); REFINED with the 업무위탁-규정 + internal-준법감시인 boundary; ACV (₩100-300M) left buyer-unvalidated. No version bump (calibration, not a new rule). New watch-risk: @@ -171,22 +215,19 @@ buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. governance 소명 is the #1 pain. Watch-risk = vendor-risk paradox. 🔴 CYCLE-18 KILL: 중대재해처벌법 5-49인 SME safety SaaS (K17, T1/T5) — anti-tunnel test confirmed leaving the FI buyer forfeits both moat (G2) and distribution (G5). 3 laws now converge on the same FI shape. -⚠️ STANDING CONSTRAINT: bottleneck is VALIDATION, not idea supply — 3 surviving FI ideas, 0 buyer signals. -🔬 LEAD-1 AI-compliance execution layer → consolidated compliance ops, mid-tier KR FIs. NOW DUAL-CLASSED. - ANTI-GAMING FLAG → RESOLVED (cycle 17): confirmed GAMED. Mid-tier-FI-AI-compliance ALONE ≈ $4-8M ARR - (does NOT clear $30M). Fix = SPLIT LEGS: (a) WEDGE → MODE-CASH (fast, profitable, passes lowered - ceiling); (b) $30M MODE-MOAT thesis now rests only on same-buyer expansion + the cross-client - FSS-exam BENCHMARK data product (MOAT-D1) — both UNVALIDATED. Two validate-or-die questions now. - 0 buyer signals → CANNOT exceed YELLOW until validated. -🔬 CASH-D1 책무구조도 (Responsibilities-Map) maintenance SaaS for mid-tier KR FIs. MODE-CASH, Axis-D. - Survived all gates; YELLOW; FASTEST/HOTTEST path on the board — statutory mid-tier deadline Jul 2 2026 - (33 days out), untooled, CEO personal liability. → LEADS LEAD-1's discovery as the concrete hook. -⚙️ MOAT-D1 FSS-exam-readiness BENCHMARK / consolidated-compliance data layer. MODE-MOAT, Axis-D. - Survived (contingent); folded into LEAD-1 as its resolved MOAT leg. Load-bearing risk = cross-client - data CONSENT. Not a separate 21st idea (respects anti-loop + synthesis). -🟡 PARK: TAAMS-service · David's algorithm · NuvoVet (GTM-blocked). -🔴 KILL: 18 ideas (see ledger). -NEXT ACTION (per Prime Directive — ANTI-LOOP active: do NOT generate new ideas): VALIDATE. Send 3-5 - discovery messages via Ace's dad + Coxwave led by the 책무구조도 Jul-2 deadline. Ask what they DID. - Listen for both validate-or-die signals (subscription WTP; vendor-consolidation + data-consent). - Decision date = 2026-07-02. +🟢 EXPAND-1 (NEW LEAD, cycle 26) AI store-revenue forecasting + next-site selection for KR multi-unit/franchise + operators. Axis-B (David site-IP × Ace AI), MODE-MOAT + MODE-CASH per-site-report leg. Passes G2 (real moat) + + G9 (ROI/loss-PULL, enforcement-free) + proven WTP comps (SiteZeus/Buxton/Placer.ai; KR 맥세스/유재은 consult). + Validate-or-die: (1) KR 본사 pay for brand-specific AI forecast vs consultant? (2) David's IP productizable? 0/3. +🟡 PARK — FI ACCOUNTABILITY ENGINE (demoted 🔬→🟡 cycle 25, hypothesis rejected, G9/L6 fail, no teeth til ~2028): + CASH-D1 (책무구조도 maint) · LEAD-1 (AI-compliance ops) · MOAT-D1 (FSS benchmark) · MOAT-A1 (보안 소명) · DATA-C1 + (민원). Revive-trigger: a real buyer signal from the (cheap, still-valid) DISCOVERY_KIT, OR ~2028 grace-expiry / + exam wave, OR discovery Q0 proves software-WTP. The discovery kit can still be sent opportunistically — but the + board NO LONGER WAITS on it. Only the L9 managed-service framing has a near-term (lumpy, services) cash path. +🟡 PARK (other): TAAMS-service · David's algorithm · NuvoVet (GTM-blocked). +🔴 KILL: 19 ideas (see ledger; +책무구조도-proactive-SaaS-hypothesis T7). +NEXT ACTION (per Prime Directive): (1) HARDEN EXPAND-1 (bull/bear + attack vectors) — partly done cycle 26. + (2) VALIDATE EXPAND-1 via DAVID'S network (his domain) — ask a 본사/multi-unit operator what they DID on their + last 출점 decision, what it cost, how often they regret a site. (3) Keep generating 1 more NON-retail axis to + refill the funnel + de-concentrate. GATING UNKNOWNS for the founders: David's site-IP specifics (domain + + productizable?) + which expansion buyers he can reach. Decision date for EXPAND-1 validation = 2026-07-15. diff --git a/LEDGER.md b/LEDGER.md index b461054..3ba43bf 100644 --- a/LEDGER.md +++ b/LEDGER.md @@ -1,6 +1,6 @@ # VENTURE LEDGER > Living state of the Founder's OS. Updated every cycle. Read alongside `CLAUDE.md`. -> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 24) | **OS version:** 2.2 +> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 26) | **OS version:** 2.3 --- @@ -75,6 +75,118 @@ Ran the G6 anti-gaming math on the narrowed segment *on its own*: --- +## CYCLE 26 — GENERATE: new hunting ground (anti-tunnel) · EXPAND-1 · May 30, 2026 +**Why generation is now sanctioned** (anti-loop was active): Cycle 25 demoted the FI engine 🔬→🟡, so no +🔬 idea has un-sent gating conversations *that the board waits on*. The CONCENTRATION CHECK (new v2.3) + +operator directive both demand diversification into a NEW hunting ground that deploys a different moat. + +**MOAT-DEPLOYMENT SWEEP (new v2.3 / L10):** edge-map TRUE MOATs = (a) KR liability entity [used by the +parked FI engine], (b) data-byproduct-of-a-service [generic], (c) **David's site-selection IP — UNUSED for +24 cycles.** → bias the axis to deploy (c). **AXIS-B asset-pair: David's site-IP × Ace's full-stack+AI.** + +### EXPAND-1 — one-sentence (GENERATE output) +> **KR multi-unit & franchise operators (본사 expansion teams, 10–300 stores)** pay for **"where do we open +> the next store, and how much will it make?"** today via **외주 컨설팅 (맥세스·유재은·한국경영시스템즈) + gut + +> free 소상공인/나이스비즈맵 tools**; we do **brand-specific AI store-revenue forecasting + site scoring trained +> on the chain's OWN store performance + KR commercial-geo data**; defensible because **David's proprietary +> site-selection IP encodes the model + each client's siloed store-performance↔location data is an L4 service- +> byproduct that compounds per brand**; fits **MODE-MOAT (with a MODE-CASH per-site-report leg)**; first revenue +> in **~60–90 days (paid pilot / per-site report)** within **~10–13 hr/wk (Ace builds, David sells the domain)**. + +### COMP-SCAN (new v2.3 — evidence before gating) +- **Global analog exists & is PAID** → category is real, WTP proven: SiteZeus & Buxton (franchise/multi-unit + predictive site selection, custom-quote enterprise), Placer.ai (foot-traffic, ~$30K+/yr), GrowthFactor + (from ~$200/mo). The differentiated value the comps converge on = **brand-specific *revenue* forecasting** + ("how much will THIS site make for MY brand"), not generic foot-traffic. (Sources logged in chat.) +- **Local substitute** → KR 소상공인 상권분석 is commoditized/FREE (소상공인진흥공단 상권정보, 서울 골목상권, + 나이스비즈맵) — so the *solo 소상공인* segment is T1 commodity. The PAID gap is the **multi-unit 본사** that + evaluates many sites/yr and currently buys *consulting* (proven, enforcement-independent WTP). + +### SYMMETRIC GATE +- KILL-SEARCH: incumbents have data scale (Buxton since 1994, NICE owns KR 상권 data); generic 상권분석 is free; + "IP" might be tacit consulting skill, not productizable; forecast errors destroy trust fast. +- WIN-SEARCH: KR mid-size chains too small for Buxton/SiteZeus enterprise pricing + underserved by free DIY + tools + consultants don't leave a compounding model behind → a brand-tuned, consultant-grade-but-software- + priced forecaster, sold via David's domain relationships, can take the mid-market slice. +| Gate | Verdict | Note | +|---|---|---| +| G0 foreigner-proof | ✅ | KR commercial geography + KR franchise relationships + Korean 상권 data | +| G1a budget exists | ✅ | 본사 already pay 출점 consultants (맥세스 등) — proven, recurring spend | +| G1b budget winnable | 🟡 | must beat consultant + free tools on brand-specific forecast value | +| G2 named TRUE-MOAT | ✅✅ | **David's site-selection IP — the first idea in 24 cycles with a real, non-SEED moat** | +| G3 moat type | 🟡 | IP + per-client siloed data (L4-clean) — NOT data-scale; BUT watch incumbents' scale (harden A2/A4) | +| G4 disconfirm survived | ✅ (narrowed) | survives by narrowing to mid-market multi-unit 본사 (not free-served solos, not enterprise Buxton turf) | +| G5 distribution <90d | 🟡 | rides **David's** network (his domain) — load-bearing unknown = which 본사 he can reach | +| G6 ceiling (MODE-MOAT) | 🟡→✅ | SiteZeus/Buxton are $XXM+ co's; KR franchise mkt is huge; clears $30M IF it becomes the brand-forecast standard + data network. Anti-gaming: mid-market slice alone may be MODE-CASH-sized — moat leg needs multi-brand expansion. | +| G7 why-now | 🟡 | **softer than a regreg deadline** — AI makes brand-specific forecasting cheap now (vs Buxton's heavy-data 1994 model); post-COVID 출점 risk-aversion. Honestly logged as the weakest gate. | +| G8 bandwidth-fit | ✅ | MVP = forecasting model + thin UI fed by a chain's store data + public geo; async-buildable | +| **G9 WTP-TRIGGER (new)** | ✅ **PULL** | a bad 출점 = a failed store = quantified $ loss → ROI/loss-PULL, **enforcement-INDEPENDENT.** The exact opposite of the FI engine's T7 problem. This is why EXPAND-1 is structurally healthier. | +**Gate result:** SURVIVES (contingent on G5 distribution + the David-IP unknown). No kill-type. + +### HARDEN — bull/bear adversary (built-in) +- **#1 limitation:** forecast *credibility* — one wrong revenue prediction on a real lease and trust dies; + and David's "IP" may be tacit (a productizability risk). MUTATE: ship as **decision-support + confidence + bands + "compare to your existing stores"**, not a single magic number; price the *pilot* on back-testing + against the chain's OWN historical openings (cheap proof, no lease risk). Regression-check: still bandwidth-fit ✅. +- **BULL:** finally deploys the dormant TRUE MOAT; enforcement-independent ROI-WTP; a proven consulting market + to displace with software margin; data compounds per brand (L4); perfect Ace/David division of labor. +- **BEAR (logged as permanent watch-risk):** incumbents' data-scale + NICE's ownership of KR 상권 data could + commoditize the inputs; mid-size KR chains may not evaluate enough sites/yr to sustain recurring SaaS (→ may + be per-report MODE-CASH, not subscription); **David's IP may not productize** (tacit expertise ≠ model). +- **Resolution:** promote to validate. Watch-risks carried: forecast-accuracy-trust, IP-productizability, recurring-vs-per-report. + +### VALIDATE-OR-DIE (for the discovery David must run) +1. **(WTP / recurring)** Will a KR 본사 pay for brand-specific AI forecasting *instead of / on top of* their + current consultant — recurring, or only per-site? *(ask what they DID on the last 출점 + what it cost.)* +2. **(moat productizability)** Is David's site-selection IP encodable into a model, or is it tacit judgment? + +**Status:** 🟢 **NEW LEAD / 🌱→⚙️ HARDENING→🔬 VALIDATE.** 0/3 buyer signals (cannot exceed YELLOW on desk +research — §5). Validation channel = **David's network** (his domain), NOT Ace's dad/Coxwave (that was the FI +channel). Decision date for EXPAND-1 = **2026-07-15**. +**Seeded for next cycle (funnel refill, de-concentration):** EXPAND-2 (post-opening 부진점 turnaround analytics, +same buyer — expansion SKU, do NOT split focus) + 1 deliberately NON-retail axis TBD (needs David's network map). + +--- + +## CYCLE 25 — META-LOOP (operator-triggered): reject the FI proactive-SaaS hypothesis · May 30, 2026 +**Trigger:** operator (Ace) stress-tested the board's load-bearing assumption — *"mid-tier firms (small/mid +asset managers, regional savings banks, small fintechs) will proactively pay a recurring SaaS subscription +out of good hygiene BEFORE a regulatory forcing function."* Correct instinct → ran the META-LOOP on it. + +### FALSE-POSITIVE HUNT (the assumption the board over-trusted) +Desk EVIDENCE gathered this cycle (sources logged in chat): +1. **책무구조도 제재 운영지침**: a **2-year 한시(限時) 운영** grace period per cohort, and sanctions apply **only to + violations AFTER submission** (지배구조법 부칙 §4). Mid-tier files Jul-2026 → **effectively no teeth until ~2028.** +2. **실증 behavior**: securities firms that already implemented 책무구조도 call it a **공시 부담** (a cost to + minimize), not a capability to invest in. +3. **Category pattern**: compliance software globally is bought on **enforcement-PULL** ("an OFAC solution costs + <20% of the fine"); proactive/hygiene WTP appears only under *continuous* audit pressure (SOC 2 etc.), which + the KR mid-tier FI does NOT face for 2 more years. +**VERDICT: hypothesis REJECTED.** The pure recurring-SaaS cash leg is **T7 (pure hygiene-PUSH, no forcing +function)** for the 2026–2028 window. Symmetric WIN-SEARCH: only the **L9 managed-service** framing survives +(rides proven *consulting*-WTP) — a lumpy services business, NOT the venture the board over-weighted. + +### PORTFOLIO VERDICT (CONCENTRATION CHECK, new v2.3) +Cycles 17→24 produced 5 survivors (CASH-D1, LEAD-1, MOAT-D1, MOAT-A1, DATA-C1) = **same engine, one shared +bear (L6 software-WTP).** That shared bear is now confirmed fatal for the near term → it's a PORTFOLIO risk, +not a per-idea watch-risk. **Action: DEMOTE the entire FI engine 🔬→🟡 PARK** with revive-trigger = (a) a real +buyer signal from the (cheap, still-valid) DISCOVERY_KIT, OR (b) ~2028 grace-expiry / an actual exam wave, OR +(c) discovery Q0 proves software-WTP. The discovery kit stays usable opportunistically; **the board no longer +waits on it.** + +### METHODOLOGY FIXES (the valuable output) → OS v2.2 → v2.3 +1. **G9 WTP-TRIGGER gate** + **kill-type T7**: classify revenue as enforcement/ROI-**PULL** vs hygiene-**PUSH**; + PUSH without a forcing function is presumed ABSENT (max YELLOW, desk research can't grant it). *This gate would + have flagged the FI engine at Cycle 17 instead of Cycle 25.* +2. **L6 sharpened**: grace-period / 시범·한시운영 / no-teeth-yet = PUSH; a why-now built on a teeth-less deadline is a MIRAGE. +3. **L10 (new)**: an unused named TRUE-MOAT is a generation failure → mandatory moat-deployment sweep each GENERATE. + (David's site-IP sat idle 24 cycles → directly produced Cycle 26's EXPAND-1.) +4. **CONCENTRATION CHECK** (§7): ≥3 survivors sharing one bear ⇒ stop same-engine generation, diversify. +5. **COMP-SCAN** (§3, mandatory): evidence (global analog + local substitute) before gating, not armchair. +**No new idea generated this cycle (this is a meta/audit cycle); EXPAND-1 generated next, in Cycle 26.** + +--- + ## CYCLE 24 — VALIDATE: send-ready discovery kit + signal log (no generation) · May 30, 2026 **Why this cycle:** anti-loop is active and the board is blocked on the SAME constraint it has been for 5 cycles — buyer signals 0/3. So this cycle produces the operational artifacts that turn "messages drafted" From 58b23e611bc4e3122d01e91033d9ab706f70d703 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 30 May 2026 07:53:39 +0000 Subject: [PATCH 3/5] Cycles 27-28: OS v3.0 (kill the kill-machine), over-kill re-audit, pivot to CASH-F1 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cycle 27 (META-LOOP, operator-triggered, MAJOR framework rebuild -> v3.0): - CAPTURABLE > BIG: a small profitable slice is a terminal WIN; no idea is a unicorn; competition is acceptable; kill only structural WALLs. - RESCUE-BEFORE-KILL; WALL-vs-SPEED-BUMP (T2 retired -> T2-WALL); DEEP COMP-SCAN with research-depth proportional to irreversibility; WIN-TYPES; standing over-kill re-audit. New laws L11/L12/L13. - Over-kill re-audit (deep KR research, 2 agents): K9 REVIVED (true false-negative), K15 partial-revive, K8 reason-corrected (stays killed), K4/5/6 stay killed & stronger -> EXPAND-1 PARKED (Finda 오픈업 프로 Oct-2025 occupies the slice on a NICE card-data wall; Cycle-26 gen was shallow). Cycle 28 (GENERATE under v3.0): - NEW LEAD CASH-F1: F&B operator margin/cost-leak AI (invoice-OCR price-drift + delivery-commission-aware per-menu net profit). ROI-PULL, enforcement- independent, capturable. Secondary cash portfolio: CASH-F2/F3 + EXPAND-1b. https://claude.ai/code/session_01A1BJYL2PvZh683Gh8KvuG8 --- CLAUDE.md | 175 +++++++++++++++++++++++++++++++++++++++--------------- LEDGER.md | 129 +++++++++++++++++++++++++++++++++------- 2 files changed, 234 insertions(+), 70 deletions(-) diff --git a/CLAUDE.md b/CLAUDE.md index 85bbe95..16cfa4a 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -1,4 +1,4 @@ -# THE FOUNDER'S OPERATING SYSTEM (v2.3) +# THE FOUNDER'S OPERATING SYSTEM (v3.0) ### An agentic protocol that HUNTS, HARDENS, VALIDATES, and COMMITS to an idea — and improves its own methods. ### For: Ace Yeom + David Mun. Self-contained — hand to any AI/instance and it runs. Supersedes v1.0. ### v2.0 changelog: added forcing function, symmetric judgment, buyer-signal-as-input, anti-gaming check, @@ -16,12 +16,34 @@ ### risk ⇒ STOP same-engine generation, diversify); (5) mandatory COMP-SCAN in GENERATE/GATE (find the global ### analog + local substitute as EVIDENCE before gating). Evidence that the FI 책무구조도 cash leg is PUSH/no- ### teeth-til-2028 → entire FI engine demoted 🔬→🟡 PARK; lead is now EXPAND-1 (site-selection, MODE-MOAT). +### v3.0 changelog (cycle 27, OPERATOR-TRIGGERED — MAJOR, philosophy shift): operator found the OS was still a +### KILL-MACHINE that over-researches the death case, under-researches the path THROUGH, and worships unicorns — +### proof: it killed David's site-IP 3× (K4/5/6 "incumbents exist") then re-generated it as the lead (EXPAND-1). +### CORE RESET — "CAPTURABLE > BIG": a small, profitable, capturable slice is a terminal WIN, not a consolation; +### no idea is a flawless unicorn (they don't exist); competition/challenges are EXPECTED and ACCEPTABLE; we do +### NOT need to capture a whole market — just a paying slice that clears our bandwidth. Fixes: (1) RESCUE-BEFORE- +### KILL (a gate fail triggers a mandatory mutation attempt, not a kill); (2) WALL-vs-SPEED-BUMP (kill ONLY behind +### a structural WALL — capital/license/winner-take-all/true-commodity-no-edge; a mere incumbent = SPEED-BUMP → +### needs an overcome-strategy, not a tombstone); T2 retired → T2-WALL (much higher bar); (3) DEEP COMP-SCAN with +### research-depth ∝ irreversibility (kills demand the DEEPEST, KR-specific research — never kill on one shallow +### US-centric search); (4) CAPTURABLE-SLICE reframes G6 (cash: "enough revenue for 10-13 hr/wk?" not "$30M?"); +### narrowing to a winnable niche is FOCUS, not gaming; (5) WIN-TYPES taxonomy (W1-W6) for true symmetry; (6) +### STANDING OVER-KILL RE-AUDIT with teeth. New laws L11 (capturable>big) · L12 (kill walls not speed-bumps) · +### L13 (research depth ∝ irreversibility). The Prime Directive is unchanged; everything else served the kill bias. ═══════════════════════════════════════════════════════════════ ## 0. PRIME DIRECTIVE & STOP/COMMIT RULES (read first — fixes the "kill machine" flaw) ═══════════════════════════════════════════════════════════════ The goal is NOT to avoid bad ideas. It is to COMMIT to the best available idea by a deadline and produce a buyer-validated BUILD PLAN. Killing is a means, not the product. +**★ CAPTURABLE > BIG (v3.0 prime frame — read before every gate):** the win condition is a CAPTURABLE, + PROFITABLE slice that clears our bandwidth — even small revenue (~₩100-500M/yr) is a WIN, not a consolation + prize. We do NOT need to dominate or capture a whole market. NO IDEA IS A FLAWLESS UNICORN — they do not + exist; every real idea has competitors and challenges, and that is ACCEPTABLE. The question is NEVER "is this + uncontested / could this be a $30M venture?" — it is "can THIS 2-person team take a paying, defensible-enough + slice and ship revenue?" Kill an idea only when it cannot make MEANINGFUL revenue for US behind a STRUCTURAL + WALL — never because a competitor exists or because it won't be huge. (This frame overrides any gate that + smuggles unicorn-bias back in. MODE-MOAT still chases ceiling; MODE-CASH does NOT.) - FORCING FUNCTION: set a DECISION DATE at the start. On that date, the highest-scoring surviving idea is PROMOTED to BUILD PLAN even if imperfect. "Best by [date]" > "perfect, never." - BUILD QUOTA: at most 3 generate→gate→harden cycles before one idea MUST advance to buyer-validation. @@ -67,28 +89,54 @@ deploy it (e.g., David's site-IP → AXIS-B asset-pair). Don't keep mining a hun weak while a strong moat sits idle.** ═══════════════════════════════════════════════════════════════ -## 3. GATE — SYMMETRIC (fixes "biased toward killing" + "gameable") -═══════════════════════════════════════════════════════════════ -For EACH idea run BOTH searches, equal effort. **COMP-SCAN FIRST (v2.3, mandatory, EVIDENCE not armchair): -find (i) the GLOBAL analog — does this category exist & get paid for elsewhere? what's the price band & -proven WTP trigger? and (ii) the LOCAL incumbent / free substitute. No comp-scan → gating is just opinion.** - (a) KILL-SEARCH: who owns this / why does it fail? - (b) WIN-SEARCH: under what real, specific conditions could a 2-person team take a profitable slice? -KILL ONLY IF win-conditions are ALSO absent. Then check gates in order; one fail = stop + log kill-type. - G0 hunting-ground (foreigner-proof?) · G1a budget exists · G1b budget winnable · - G2 named TRUE-MOAT asset (not "relationships") · G3 moat type (not data-scale/GPU/capital wall) · - G4 disconfirm survived · G5 distribution <90-day test path · G6 ceiling (per MODE) · - G7 why-now catalyst · G8 BANDWIDTH-FIT (path to revenue fits ~10-13 hr/wk now?) · - **G9 WTP-TRIGGER (v2.3): is revenue enforcement/ROI-PULL (a live fine, exam-with-teeth, deadline-with-teeth, - or a quantified loss-event compels the buy) or hygiene-PUSH (buyer pays proactively for good practice)? - PUSH-WTP without a forcing function is PRESUMED ABSENT → cap at YELLOW + flag high-risk; desk research - CANNOT grant it (only a buyer signal can). Grace-period / 시범운영 / 한시운영 / no-teeth-yet = PUSH, not - PULL — the why-now is a mirage. Kill-type for a pure-PUSH idea with no signal = T7.** -ANTI-GAMING CHECK (mandatory after any segment-narrowing): "Does the narrowed segment that saved G4 -still pass G6 (ceiling) ON ITS OWN?" If narrowing to dodge competitors breaks the ceiling, the idea is -GAMED — mark it and either widen (and re-face competitors) or accept it as MODE-CASH/niche only. -KILL-TYPES: T1 commodity · T2 incumbent-owns-wedge · T3 capital/license/data wall · T4 ceiling-low · - T5 wrong-team/no-distribution · T6 bandwidth-incompatible-now · T7 pure-hygiene-PUSH-WTP-no-forcing-function. +## 3. GATE — SYMMETRIC + RESCUE-FIRST (v3.0 rebuild: kills the kill-machine) +═══════════════════════════════════════════════════════════════ +**STEP 1 — DEEP COMP-SCAN (mandatory, EVIDENCE not armchair; research-depth ∝ irreversibility, L13).** +Because a KILL is near-irreversible (we stop thinking about it), competitor research for a kill must be the +DEEPEST work in the cycle — never kill on one shallow or US-centric search. Produce a COMPETITOR MAP: + • the GLOBAL analog (does this category get paid for elsewhere? price band? proven WTP trigger?) AND + • the LOCAL (KR) incumbents BY NAME — what they ACTUALLY do, who they actually serve, their pricing, and — + most important — their SPECIFIC GAP (segment they ignore, job they don't do, price tier they're too heavy + for). "An incumbent exists in the category" is NOT a finding; "incumbent X serves segment Y at price Z and + is absent from slice W" is. + +**STEP 2 — CLASSIFY EACH OBSTACLE: WALL or SPEED-BUMP (the core anti-overkill rule, L12).** + • WALL (structural, justifies a kill): capital wall · license/regulatory monopoly · genuine winner-take-all + network effect · true commodity where WE have zero edge · mass-market with no founder distribution. + • SPEED-BUMP (does NOT justify a kill): a competitor exists but serves a different/larger segment, is too + expensive/heavy for the slice, has a product gap, or can be undercut / unbundled / partnered / service- + wrapped. A SPEED-BUMP requires an OVERCOME-STRATEGY, not a tombstone. Most "T2 incumbent" kills were + really speed-bumps (proof: K4/5/6 site-selection killed, then revived as EXPAND-1). + +**STEP 3 — SYMMETRIC SEARCH, equal effort:** + (a) KILL-SEARCH: what's the WALL (not just "who exists")? (b) WIN-SEARCH: what's the smallest CAPTURABLE, + PAYING slice a 2-person team could take, and the overcome-strategy + WIN-TYPE for it? + +**STEP 4 — GATES (check in order). A gate FAIL does NOT kill → it triggers STEP 5 (RESCUE).** + G0 hunting-ground (foreigner-proof?) · G1a budget exists (someone pays for this job today) · G1b budget + winnable (is there a slice below/beside the incumbent?) · G2 edge (MODE-MOAT: a named TRUE-MOAT; **MODE-CASH: + a 12-month head-start / founder-domain edge / service-leverage is ENOUGH — a durable moat is NOT required**) · + G3 edge-type (not behind a WALL) · G4 disconfirm survived · G5 distribution <90-day test path · G6 + **CAPTURABLE-SLICE (replaces ceiling-worship): MODE-CASH asks "is the reachable revenue worth ~10-13 hr/wk? + (small is fine)"; MODE-MOAT asks "ceiling $30M+?". Narrowing to a winnable niche is FOCUS, not gaming** · + G7 why-now (or "evergreen pain" — absence of a catalyst is not fatal for MODE-CASH) · G8 BANDWIDTH-FIT · + **G9 WTP-TRIGGER: enforcement/ROI-PULL vs hygiene-PUSH. PUSH-WTP without a forcing function is presumed + ABSENT (max YELLOW); grace-period/시범·한시운영/no-teeth = PUSH, why-now is a mirage. Desk research can't + grant PULL — only a buyer signal can.** + +**STEP 5 — RESCUE-BEFORE-KILL (mandatory; this is the kill-machine fix).** Before logging ANY kill, attempt +to mutate past the failing gate with the operators: narrow · change-buyer · change-wedge · re-price · +service-wrap (L9) · bundle · partner-with-the-incumbent · pick-a-smaller-but-paying-slice. KILL ONLY IF: + (i) no mutation clears the gate, AND (ii) WIN-SEARCH finds NO capturable paying slice, AND (iii) the failure + is a true WALL (Step 2). Otherwise the idea SURVIVES (possibly re-scoped/MODE-switched) → HARDEN. + +WIN-TYPES (name one on every survival — true symmetry vs kill-types): W1 underserved-slice-below-incumbent- + pricing · W2 unbundle-one-painful-job · W3 service-wrap-with-software-leverage · W4 partner/channel-with- + incumbent · W5 founder-domain-edge-in-a-niche-incumbents-ignore · W6 AI/speed cost-advantage-over-legacy. +KILL-TYPES (all now require a WALL, not mere competition): T1 true-commodity-no-edge · T3 capital/license/ + data-SCALE wall · T4 revenue-too-small-EVEN-as-a-niche (not "ceiling < $30M" — that's fine for cash) · T5 + mass-market-no-founder-distribution · T6 bandwidth-incompatible-now · T7 pure-hygiene-PUSH-no-forcing- + function · **T2-WALL incumbent owns it AND no reachable slice survives rescue (replaces old T2; HIGH bar).** ═══════════════════════════════════════════════════════════════ ## 4. HARDEN (make survivors stronger; same loop, now adversarial) @@ -121,7 +169,10 @@ MONEY/100: ACVx3 · underserved-painx3 · WTP-EVIDENCEx3 (0 if no buyer signal) EXECUTION/70: Ace-demo-in-bandwidthx3 · data-cheap-day1x3 · named-warm-buyersx3 · no-license-gapx2 · defensible-12mox2 · low-burnx1. GRID: MoneyHi+ExecHi=GREEN(only with buyer signal) · mixed=YELLOW · LowLow=RED. -MODE re-weights: MODE-CASH triples time-to-rev + bandwidth; MODE-MOAT triples ceiling + defensibility. +MODE re-weights: MODE-CASH triples time-to-rev + bandwidth **and ZEROES the ceiling + market-growth weights +(v3.0: a cash idea is NOT penalized for being small — capturable revenue that clears bandwidth scores full);** +MODE-MOAT triples ceiling + defensibility. **Never let a low ceiling pull a MODE-CASH idea to RED — that's the +unicorn-bias the Prime Directive forbids. RED is reserved for "no capturable revenue / behind a WALL," not "small."** ═══════════════════════════════════════════════════════════════ ## 7. DECIDE + LEDGER + the SELF-IMPROVEMENT META-LOOP (fixes "static methods") @@ -138,15 +189,21 @@ moat strong (converge) or is the shared BEAR fatal (diversify)? (Cycle 17→24 f **META-LOOP (the framework improves itself — run every 8 ideas or monthly):** 1. HIT-RATE AUDIT: which gates did the actual killing? Which NEVER fired (dead weight → cut them)? - 2. FALSE-NEGATIVE HUNT (most valuable): did any idea we KILLED later get funded / succeed for someone - else? If yes, that gate/judgment was WRONG → find the flawed assumption, rewrite the rule. + 2. **STANDING OVER-KILL RE-AUDIT (v3.0 — now has teeth, most valuable):** walk the ENTIRE kill list. For + every kill, ask: was this a true WALL (Step 2) or a category-dismissal SPEED-BUMP? Any kill whose reason + is "an incumbent/competitor exists" (old T2), or that cited only US incumbents / a single shallow search, + or that pre-dates a relevant change (new tech, new reg, cheaper AI) → RE-OPEN with a fresh DEEP COMP-SCAN. + Track a `revived?` column. Also the classic FALSE-NEGATIVE hunt: did any killed idea get funded/succeed + for someone else? If yes the gate was wrong → rewrite it. (Trigger: K4/5/6 over-killed → became EXPAND-1.) 3. CALIBRATION: were HARDEN's "validate-or-die" predictions borne out by real buyer conversations? Where desk-research disagreed with buyers, TRUST BUYERS and down-weight that desk signal. - 4. VERSION IT: rewrite the offending gate/Law/weight, bump version (v2.1, v2.2…), LOG what changed & why. + 4. **RESEARCH-DEPTH AUDIT (v3.0, L13): did any decision (esp. a kill) rest on shallow / US-only / single- + search evidence? Flag it as under-researched and re-do the comp-scan before trusting the verdict.** + 5. VERSION IT: rewrite the offending gate/Law/weight, bump version, LOG what changed & why. The methods are NOT fixed. This loop is how they change. Nothing here is sacred except the Prime Directive. ═══════════════════════════════════════════════════════════════ -## 8. THE 10 LAWS (current best; the META-LOOP may rewrite any of them) +## 8. THE 13 LAWS (current best; the META-LOOP may rewrite any of them) ═══════════════════════════════════════════════════════════════ L1 build the accountability layer, not the commodity primitive · L2 loud pain in a mature market = served · L3 warm channel ≠ winnable without buildable product · L4 durable moat = liability-entity + data-byproduct- @@ -161,7 +218,15 @@ Boundary: managed service triggers 업무위탁 규정 + cannot replace the inte replaces); ACV still buyer-unvalidated. · **L10 (new v2.3) a documented TRUE-MOAT asset left UNUSED across cycles is a generation failure, not neutral: every cycle's GENERATE must run the moat-deployment sweep and bias an axis toward deploying an idle moat. (Trigger: David's site-selection IP sat unused 24 cycles while -the board mined a moat-less FI hunting ground.)** +the board mined a moat-less FI hunting ground.)** · **L11 (new v3.0, CAPTURABLE>BIG) a small, profitable, +capturable slice is a terminal WIN, not a consolation; no idea is a flawless unicorn; the test is "can we +take a paying slice?" not "is this uncontested / huge?" — small revenue that clears bandwidth counts.** · +**L12 (new v3.0) kill WALLS, not SPEED-BUMPS: a competitor existing is a speed-bump needing an overcome- +strategy, not a tombstone. Kill only behind a structural wall (capital/license/winner-take-all/true-commodity- +no-edge/mass-market-no-distribution). "An incumbent exists" is the single most over-used false kill (K2→T2).** · +**L13 (new v3.0) research depth ∝ irreversibility: a KILL is near-irreversible, so it demands the DEEPEST, +KR-specific, multi-source research — never kill on one shallow or US-centric search. Cheap/reversible moves +get light research; expensive/sticky ones get deep research.** (Each Law is provisional. If the META-LOOP finds a counter-example, it gets rewritten with a logged reason.) ═══════════════════════════════════════════════════════════════ @@ -171,14 +236,22 @@ MON pick MODE(s) + generate from rotating axis · TUE symmetric gate (kill+win s WED harden + bull/bear adversary · THU score + ledger + (every 8) META-LOOP audit · FRI VALIDATE: send 3-5 buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. -## CURRENT STATE (updated cycle 26 — May 30 2026; full detail in LEDGER.md) -🧭 CYCLE 26 = GENERATE (new hunting ground, anti-tunnel). NEW LEAD: **EXPAND-1** — AI store-revenue forecasting - & next-site selection for KR multi-unit / franchise operators. Axis-B (asset-pair: David's site-selection IP - × Ace's AI/full-stack), MODE-MOAT w/ a MODE-CASH per-site-report leg. WHY IT'S HEALTHIER THAN THE FI ENGINE: - G2 finally has a real moat (David's site-IP); G9 passes as ROI/loss-PULL (a bad 출점 = real $ lost, enforcement- - independent); proven WTP (KR franchises already pay 맥세스/유재은-type consultants; global comps SiteZeus/Buxton/ - Placer.ai = paid category, $200/mo→$30K+/yr). Survives gate (contingent). Validate-or-die: (1) will a KR 본사 pay - for brand-specific AI forecasting vs their consultant? (2) is David's IP productizable or tacit? 0/3 signals. +## CURRENT STATE (updated cycle 28 — May 30 2026; full detail in LEDGER.md) +🧭 CYCLE 28 = GENERATE under v3.0. NEW LEAD: **CASH-F1** — F&B operator margin / cost-leak AI (invoice-OCR + supplier-price-drift + delivery-commission-aware per-menu net profit + repricer) for KR independent/small-chain + F&B ops, via David's network. MODE-CASH, Axis-D. Two deep-research passes CONVERGED here (revived K9 + fresh + scan). SPEED-BUMP (도도카트 free=cost-tracking only; 푸드로직 costing-only & young; nobody does price-drift+plate- + cost). Passes G9 (ROI-PULL, enforcement-INDEPENDENT: delivery fees 16.9–29.3% of sales) + G6 capturable + (~₩100–300M/yr). 🔬 VALIDATE, 0/3 signals, decision 2026-07-15. Secondary cash portfolio (same buyer): CASH-F2 + (franchise-HQ QSC for sub-50-store brands), CASH-F3 (managed review service), EXPAND-1b (interior rollout-PM). +🔧 CYCLE 27 = META-LOOP (operator-triggered, MAJOR). Operator critique: the OS is a kill-machine that worships + unicorns + misses KR competitors. REBUILT → **OS v3.0**: CAPTURABLE>BIG (small profitable slice = a WIN; L11), + RESCUE-BEFORE-KILL, WALL-vs-SPEED-BUMP (kill only structural walls; T2 retired→T2-WALL; L12), DEEP COMP-SCAN + w/ research-depth ∝ irreversibility (L13), WIN-TYPES, standing over-kill re-audit. OVER-KILL RE-AUDIT (deep KR + research): **K9 REVIVED → CASH-F1** (true false-negative); K15 partial-revive (EXPAND-1b); K8 stays killed + (reason corrected: KR 시프티/알밤/토스, not US players); **K4/5/6 stay killed, now STRONGER → EXPAND-1 PARKED** + (오픈업 프로/Finda launched Oct-2025 = exact product on a NICE card-data WALL; my Cycle-26 gen was shallow — + v3.0 caught its author's own error). 🔴 CYCLE 25 = META-LOOP (operator-triggered). REJECTED the FI engine's load-bearing hypothesis ("mid-tier FIs proactively pay recurring SaaS pre-enforcement") on EVIDENCE: 책무구조도 제재 has a 2-yr 한시-운영 grace + sanctions only post-submission ⇒ no teeth til ~2028; 증권사 call it 공시 부담; compliance SaaS globally is fine-PULL not @@ -215,19 +288,25 @@ buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. governance 소명 is the #1 pain. Watch-risk = vendor-risk paradox. 🔴 CYCLE-18 KILL: 중대재해처벌법 5-49인 SME safety SaaS (K17, T1/T5) — anti-tunnel test confirmed leaving the FI buyer forfeits both moat (G2) and distribution (G5). 3 laws now converge on the same FI shape. -🟢 EXPAND-1 (NEW LEAD, cycle 26) AI store-revenue forecasting + next-site selection for KR multi-unit/franchise - operators. Axis-B (David site-IP × Ace AI), MODE-MOAT + MODE-CASH per-site-report leg. Passes G2 (real moat) - + G9 (ROI/loss-PULL, enforcement-free) + proven WTP comps (SiteZeus/Buxton/Placer.ai; KR 맥세스/유재은 consult). - Validate-or-die: (1) KR 본사 pay for brand-specific AI forecast vs consultant? (2) David's IP productizable? 0/3. +🟢 CASH-F1 (NEW LEAD, cycle 28) F&B operator margin/cost-leak AI — invoice-OCR price-drift + delivery-commission- + aware per-menu net profit + repricer. MODE-CASH, Axis-D, via David's F&B network. SPEED-BUMP; W1/W2/W3. G9 + ROI-PULL (enforcement-free). Capturable ~₩100–300M/yr. Validate-or-die (1 q): "last time 식자재 prices jumped, + how did you decide whether/how much to reprice, and what did you use?" 🔬 VALIDATE, 0/3, decision 2026-07-15. +🌱 SECONDARY CASH PORTFOLIO (same buyer = David's F&B/retail ops; one discovery tests all; build ≤1 pre-signal): + CASH-F2 franchise-HQ QSC/SV audit for sub-50-store brands · CASH-F3 managed review/reputation service · + EXPAND-1b HQ multi-store interior rollout-PM (narrow K15 revive, NOT a marketplace=captive-channel wall). +🟡 PARK — EXPAND-1 (site-selection/forecasting, parked cycle 27): T3 data WALL — 오픈업 프로/Finda (Oct-2025, ~₩825k + /mo) on NICE card-data; software play walled, only a David-consulting residue survives. Revive only if 오픈업 + fails on accuracy → a service-led "interpret the data for your 출점 committee" play. 🟡 PARK — FI ACCOUNTABILITY ENGINE (demoted 🔬→🟡 cycle 25, hypothesis rejected, G9/L6 fail, no teeth til ~2028): CASH-D1 (책무구조도 maint) · LEAD-1 (AI-compliance ops) · MOAT-D1 (FSS benchmark) · MOAT-A1 (보안 소명) · DATA-C1 (민원). Revive-trigger: a real buyer signal from the (cheap, still-valid) DISCOVERY_KIT, OR ~2028 grace-expiry / exam wave, OR discovery Q0 proves software-WTP. The discovery kit can still be sent opportunistically — but the board NO LONGER WAITS on it. Only the L9 managed-service framing has a near-term (lumpy, services) cash path. -🟡 PARK (other): TAAMS-service · David's algorithm · NuvoVet (GTM-blocked). -🔴 KILL: 19 ideas (see ledger; +책무구조도-proactive-SaaS-hypothesis T7). -NEXT ACTION (per Prime Directive): (1) HARDEN EXPAND-1 (bull/bear + attack vectors) — partly done cycle 26. - (2) VALIDATE EXPAND-1 via DAVID'S network (his domain) — ask a 본사/multi-unit operator what they DID on their - last 출점 decision, what it cost, how often they regret a site. (3) Keep generating 1 more NON-retail axis to - refill the funnel + de-concentrate. GATING UNKNOWNS for the founders: David's site-IP specifics (domain + - productizable?) + which expansion buyers he can reach. Decision date for EXPAND-1 validation = 2026-07-15. +🟡 PARK (other): TAAMS-service · NuvoVet (GTM-blocked). +🔴 KILL: 19 ideas (see ledger; K9 REVIVED→CASH-F1; K15 partial-revive; re-audit column added). +NEXT ACTION (per Prime Directive): VALIDATE CASH-F1 via DAVID'S F&B network. ONE discovery question tests the + whole cash portfolio: *"Last time 식자재 prices jumped, how did you decide whether/how much to reprice a menu, + and what did you use?"* — listen for a painful manual process described BEFORE naming any product (a polite + "sounds useful" = ZERO). Same conversation can probe CASH-F3 (review retainer) + CASH-F2 (QSC). Build ≤1 thing + pre-signal. Decision date = 2026-07-15. (FI engine + EXPAND-1 parked; do not build either.) diff --git a/LEDGER.md b/LEDGER.md index 3ba43bf..b7511f5 100644 --- a/LEDGER.md +++ b/LEDGER.md @@ -1,6 +1,6 @@ # VENTURE LEDGER > Living state of the Founder's OS. Updated every cycle. Read alongside `CLAUDE.md`. -> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 26) | **OS version:** 2.3 +> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 27) | **OS version:** 3.0 --- @@ -75,7 +75,90 @@ Ran the G6 anti-gaming math on the narrowed segment *on its own*: --- +## CYCLE 28 — GENERATE under v3.0 (new lead) + secondary cash portfolio · May 30, 2026 +**Hunting ground:** David's F&B/retail operator network (AXIS-D buyer-back). Two independent deep-research +passes (the over-kill re-audit's K9 revival + a fresh 2nd-vertical scan) CONVERGED on the same unowned, +ROI-driven, AI-shaped job → strong signal. Generated under the v3.0 rules (deep comp-scan, WALL-vs-speed-bump, +capturable>big, rescue-first). + +### CASH-F1 (NEW LEAD) — F&B operator margin / cost-leak AI +> **KR independent + small-chain F&B operators** pay today for margin/costing via **Excel templates (크몽/오투잡 +> ₩5–30k), 도도카트 (FREE invoice-OCR cost-tracking), gut**; we do **invoice-OCR supplier-price-drift detection + +> delivery-commission-aware per-menu NET-profit + a "raise this dish ₩X" repricer**; edge = **a painful job no KR +> player does (도도카트 stops at cost-tracking; 푸드로직 is costing-only & launched Sept-2025; 그랜터 is store-level +> P&L)** + **David's operator network as channel + service-wrapped onboarding**; fits **MODE-CASH**; first revenue +> **60–90 days**, **~10–13 hr/wk** (Ace builds the OCR/anomaly pipeline; David sells; onboarding is the paid wedge). +- **DEEP COMP-SCAN:** 도도카트(스포카) FREE, ~100k users, cost-tracking/benchmark only (monetizes data, not analytics); + 푸드로직 recipe→cost + 메뉴엔지니어링 but NOT delivery-commission-aware net margin, very early; 그랜터 store-level + P&L not per-dish; 마켓봄/미리 = ordering/inventory, no leak-detection (distributor sets price → no incentive). + US analog MarginEdge proves the job is real & paid; **no KR equivalent.** +- **WALL-vs-SPEED-BUMP:** SPEED-BUMP (fragmented incumbents, the per-dish + delivery-aware + price-drift slice is + genuinely open). **WIN-TYPES:** W2 (unbundle the painful job) · W1 (below 푸드로직 / beside free 도도카트) · W3 + (service-wrapped recipe/cost setup = the #1 adoption barrier, sold as paid onboarding). +- **GATES (v3.0):** G1a ✅ (people pay for Excel templates today) · G1b ✅ (paid analytics gap above free tools) · + G2 ✅ as cash (12-mo head-start + David channel — no durable moat REQUIRED) · G6 ✅ CAPTURABLE (~₩30–50k/mo × + few-hundred reachable = **₩100–300M/yr**, clears bandwidth — small is FINE) · G8 ✅ (solo-buildable OCR+rules) · + **G9 ✅ ROI-PULL, ENFORCEMENT-INDEPENDENT** (margin pain: delivery fees now 16.9–29.3% of sales; "sell a lot, + still lose money"). · G7 🟡 evergreen pain (no catalyst — acceptable for cash per v3.0). +- **BULL/BEAR:** BULL — acute felt pain, weak substitute (static Excel ignores delivery economics), pure-software + Ace ships solo, David's network is the exact buyer. BEAR (watch-risk) — **SMB recurring software-WTP** (the + board's perennial bear; but here cheaply testable & the substitute is weak) + 푸드로직 could extend. +- **VALIDATE-OR-DIE (1 question, David's network):** *"Last time 식자재 prices jumped, how did you decide whether/ + how much to raise a menu price — and what did you use?"* Listen for a painful manual process, not "sounds useful." +- **Status:** 🟢 **NEW LEAD → 🔬 VALIDATE.** 0/3 signals. Decision date = **2026-07-15.** + +### Secondary cash candidates (same buyer = David's F&B/retail ops — a deliberate CASH PORTFOLIO, one discovery tests all) +- **CASH-F2** — franchise-HQ QSC/SV store-audit app for **sub-50-store emerging brands** (below FC다움+Kakao / + 맥스오피스 ERP). SPEED-BUMP (most contested; fight on focus+distribution, not whitespace). W5 (niche incumbents + ignore) + David relationship sale. ACV higher (~₩100–300k/mo), fewer logos. 🌱 GENERATED, secondary. +- **CASH-F3** — done-for-you review/reputation **managed SERVICE** (service-wrap commodity AI-reply tools; sell + David's venues a ₩100–200k/mo retainer). Weakest: software layer ≈ commodity (T1) so ONLY the service-wrap + saves it; 네이버 absorbing the job = decay risk. 🌱 GENERATED — cash-flow warm-up / wedge into David's venues. +- **EXPAND-1b (interior, narrow revive of K15)** — HQ-side multi-store **rollout project-management** tool/service + (NOT a marketplace — that's a captive-channel WALL). Thin MODE-CASH niche. 🌱 noted, low priority. +**CONCENTRATION CHECK (v3.0):** CASH-F1/F2/F3 share ONE buyer + ONE bear (SMB software-WTP). Acceptable now +(a single discovery conversation tests all three; lead with F1's margin pain), but do NOT build >1 until a signal. + +--- + +## CYCLE 27 — META-LOOP (operator-triggered): rebuild the framework to v3.0 + over-kill re-audit · May 30, 2026 +**Trigger:** operator's brutal critique — the OS is a KILL-MACHINE that over-researches the death case, +under-researches the path THROUGH, misses key (esp. KR) competitors, and worships unicorns when **capturable +revenue (even small) is the actual goal.** Proof of the bias: the OS killed David's site-IP 3× (K4/5/6, "T2 +incumbents exist") then re-generated it as the Cycle-26 lead. + +### A) FRAMEWORK REBUILT → OS v2.3 → v3.0 (see CLAUDE.md changelog; major, philosophy shift) +1. **CAPTURABLE > BIG** (Prime-Directive frame + L11): a small profitable slice is a terminal WIN; no idea is a + unicorn; competition is acceptable; kill only "no capturable revenue behind a WALL," never "not huge." +2. **RESCUE-BEFORE-KILL** (§3 Step 5): a gate fail triggers a mandatory mutation attempt, not a kill. +3. **WALL vs SPEED-BUMP** (L12 / Step 2): kill only behind a structural wall; a mere incumbent = speed-bump needing + an overcome-strategy. Old **T2 retired → T2-WALL** (high bar). +4. **DEEP COMP-SCAN, research-depth ∝ irreversibility** (L13 / Step 1): kills demand the deepest KR-specific + multi-source research — never a single shallow US-centric search. SCORE: a low ceiling can NO LONGER pull a + MODE-CASH idea to RED. WIN-TYPES W1–W6 added for true symmetry. STANDING OVER-KILL RE-AUDIT given teeth. + +### B) OVER-KILL RE-AUDIT (deep KR-specific research, 2 agents) — verdicts +| Kill | Original reason | v3.0 deep-research verdict | Action | +|---|---|---|---| +| **K9** SME/F&B supplier price-drift | "US MarginEdge does it" | **SPEED-BUMP — genuine false-negative.** No KR player does invoice-OCR→price-drift→plate-cost. 마켓봄=ordering (no incentive), 미리=inventory-first. | **REVIVE → became CASH-F1 (new lead).** | +| **K15** retail interior/fit-out | "숨고/916디자인 own it" | **Mixed:** marketplace = WALL (captive franchise channel + 공정위 litigation); HQ-rollout-PM niche = thin SPEED-BUMP. | **Partial revive → EXPAND-1b (low priority).** | +| **K8** F&B labor scheduling | "US Restaurant365 etc." | **WALL — stays killed, REASON CORRECTED:** real wall is KR incumbents 시프티/알밤/토스플레이스 = commoditized, near-free, payments-subsidized. Not the US reason. | Stay killed; reason → T1/T5. | +| **K4/5/6** site-selection / forecasting | "gov-free + Placer/Buxton" | **WALL — stays killed, NOW STRONGER:** 오픈업 프로 (Finda) launched **Oct-2025** = the EXACT product (~₩825k/mo) on a NICE-class card-data wall (T3). Reason was incomplete but conclusion correct & now stronger. | Stay killed; **downgrade EXPAND-1 (see below).** | +**Meta-finding:** in all 4, the real threat was a DOMESTIC incumbent or structural channel, never the cited US +software. 2 kills landed correctly for the wrong reason; 1 was a true false-negative (revived); 1 (K9-adjacent +EXPAND-1) was generated on MY OWN shallow research and is now correctly walled. **v3.0's deep-comp-scan rule +caught the framework author's own Cycle-26 error on its first run — exactly its purpose.** + +--- + ## CYCLE 26 — GENERATE: new hunting ground (anti-tunnel) · EXPAND-1 · May 30, 2026 +> **⚠️ SUPERSEDED by Cycle 27/28:** EXPAND-1 was generated on shallow research. The v3.0 deep comp-scan found +> **오픈업 프로 (Finda, launched Oct-2025)** occupies the exact paid-HQ slice (~₩825k/mo) on a NICE card-data WALL +> (T3). The *software* play is walled; only a pure-CONSULTING residue survives (David's tacit IP as a service — +> doesn't scale in 10–13 hr/wk, competes with 맥세스, no software moat). **EXPAND-1 → 🟡 PARK** (revive only if +> 오픈업 프로 fails on accuracy → a service-led "we interpret the data for your 출점 committee" play, David's lane). +> Honest re-grade: my Cycle-26 G2 "real moat" claim was wrong (David's IP is tacit/SEED per the operator + L7), +> AND I missed the funded incumbent. New lead = CASH-F1. The original Cycle-26 gate table below is left for record. **Why generation is now sanctioned** (anti-loop was active): Cycle 25 demoted the FI engine 🔬→🟡, so no 🔬 idea has un-sent gating conversations *that the board waits on*. The CONCENTRATION CHECK (new v2.3) + operator directive both demand diversification into a NEW hunting ground that deploys a different moat. @@ -480,27 +563,29 @@ Tiny KR F&B importers source over KakaoTalk and spreadsheets; a human-in-the-loo --- ## KILLED — do not re-chase (kill-type logged to prevent loops) - -| # | Idea | T | Reason | -|---|------|---|--------| -| K1 | AI HR training videos | T1 | GPU/model-quality war — Synthesia (90% Fortune 100), HeyGen, Runway ($3B) | -| K2 | AI lease abstraction | T1 | Claude/GPT do extraction natively; 10+ funded players at 90-97% accuracy | -| K3 | "Beat Leasecake" lease ops | T2 | Feature war vs 15+ incumbents; ASC842 anchor needs a CPA | -| K4 | Generic site selection US | T2 | Placer (unicorn $100M+ ARR), Buxton, SiteZeus, GrowthFactor | -| K5 | David algo → US site selection | T2 | Same as K4 | -| K6 | David algo → KR site selection | T2/T1 | Gov gives algo free (상권정보시스템) + 오픈업/Finda funded | -| K7 | Trade finance / LC automation | T2/T5 | Microsoft+HSBC, Cleareye, NeuroGaint; bank sales motion | -| K8 | Restaurant labor/scheduling | T2 | Restaurant365, Crunchtime, SynergySuite entrenched | -| K9 | Supplier price-drift/invoice-leak | T2 | MarginEdge does exact feature (product-level price alerts) | -| K10 | KR crypto VASP AML | T2/T3 | CODE consortium (Bithumb/Coinone/Korbit) + Elliptic/Notabene | -| K11 | KR institutional crypto custody | T3 | KODA/KDAC/BDACS — license + capital + bank wall | -| K12 | AI content watermarking | T1 | Commoditized May 2026 — OpenAI+SynthID, Google Chrome/Cloud, C2PA 6K+ members | -| K13 | KR-Israel connector/fund | T2 | Sovereign funds own it — OurCrowd/NH $80M, KORIL-RDF since 2001 | -| K14 | KR FDI compliance | T4 | Small ceiling; embassy = seed not moat. Possible LEAD-1 bundle SKU only | -| K15 | KR retail interior/fit-out | T2 | 숨고/916디자인/더공 (escrow+dispute referee) own it | -| K16 | TAAMS as platform | T2/T1 | Tridge+Tradlinx+free gov data own all layers end-to-end | -| K17 | 중대재해처벌법 (5-49인) safety evidence SaaS | T1/T5 | Free KOSHA tooling + licensed 안전관리대행 agencies own it; no founder distribution to ~800K SMEs (cycle 18 anti-tunnel test) | -| K18 | SME bookkeeping → lending-signal data | T2/T3 | 캐시노트/한국신용데이터 owns it — 1.7M sites (88%), ₩1.1조 unicorn, already runs CB on the data (cycle 20 anti-tunnel test) | +> **🔁 v3.0 over-kill re-audit (Cycle 27):** all "T2 incumbent-exists" kills re-examined with deep KR research. +> Verdicts in the `re-audit` column. Old T2 retired → kills now require a structural WALL (see CLAUDE.md §3). + +| # | Idea | T | Reason | 🔁 re-audit (c27) | +|---|------|---|--------|--------------------| +| K1 | AI HR training videos | T1 | GPU/model-quality war — Synthesia (90% Fortune 100), HeyGen, Runway ($3B) | WALL ✓ (true commodity, no edge) | +| K2 | AI lease abstraction | T1 | Claude/GPT do extraction natively; 10+ funded players at 90-97% accuracy | WALL ✓ (commodity) | +| K3 | "Beat Leasecake" lease ops | T2 | Feature war vs 15+ incumbents; ASC842 anchor needs a CPA | not re-audited yet (US, low founder fit) | +| K4 | Generic site selection US | T2 | Placer (unicorn $100M+ ARR), Buxton, SiteZeus, GrowthFactor | WALL ✓ (no KR edge / no distribution) | +| K5 | David algo → US site selection | T2 | Same as K4 | WALL ✓ | +| K6 | David algo → KR site selection | T2/T1→**T3** | Gov gives algo free (상권정보시스템) + 오픈업/Finda funded | **WALL, now stronger:** 오픈업 프로 (Finda, Oct-2025) = exact product ~₩825k/mo on NICE card-data wall. (=EXPAND-1, parked) | +| K7 | Trade finance / LC automation | T2/T5 | Microsoft+HSBC, Cleareye, NeuroGaint; bank sales motion | not re-audited (bank sales = T5 wall likely) | +| K8 | Restaurant labor/scheduling | T2→**T1/T5** | Restaurant365, Crunchtime, SynergySuite entrenched | **WALL, reason CORRECTED:** real wall = KR 시프티/알밤/토스플레이스 commoditized + payments-subsidized free; NOT the US players | +| K9 | Supplier price-drift/invoice-leak | ~~T2~~ | MarginEdge does exact feature (product-level price alerts) | **🟢 REVIVED → CASH-F1.** No KR player does invoice-OCR→price-drift→plate-cost. False-negative. | +| K10 | KR crypto VASP AML | T2/T3 | CODE consortium (Bithumb/Coinone/Korbit) + Elliptic/Notabene | WALL ✓ (license/consortium) | +| K11 | KR institutional crypto custody | T3 | KODA/KDAC/BDACS — license + capital + bank wall | WALL ✓ (license+capital) | +| K12 | AI content watermarking | T1 | Commoditized May 2026 — OpenAI+SynthID, Google Chrome/Cloud, C2PA 6K+ members | WALL ✓ (commodity) | +| K13 | KR-Israel connector/fund | T2 | Sovereign funds own it — OurCrowd/NH $80M, KORIL-RDF since 2001 | not re-audited (capital/relationship wall likely) | +| K14 | KR FDI compliance | T4 | Small ceiling; embassy = seed not moat. Possible LEAD-1 bundle SKU only | re-frame: small ceiling is OK per v3.0 — but PUSH-WTP (T7) likely; low priority | +| K15 | KR retail interior/fit-out | T2→**mixed** | 숨고/916디자인/더공 (escrow+dispute referee) own it | **Partial revive → EXPAND-1b:** marketplace=WALL (captive channel); HQ-rollout-PM niche=thin SPEED-BUMP | +| K16 | TAAMS as platform | T2/T1 | Tridge+Tradlinx+free gov data own all layers end-to-end | WALL ✓ (end-to-end owned) | +| K17 | 중대재해처벌법 (5-49인) safety evidence SaaS | T1/T5 | Free KOSHA tooling + licensed 안전관리대행 agencies own it; no founder distribution to ~800K SMEs (cycle 18 anti-tunnel test) | WALL ✓ (free gov + T5 no distribution) | +| K18 | SME bookkeeping → lending-signal data | T2/T3 | 캐시노트/한국신용데이터 owns it — 1.7M sites (88%), ₩1.1조 unicorn, already runs CB on the data (cycle 20 anti-tunnel test) | WALL ✓ (T3 data-scale wall) | --- From 60c66c7ff879ccd5cb10e6dd6fcc42b8c71729f9 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 30 May 2026 09:42:39 +0000 Subject: [PATCH 4/5] Cycles 29-30: recurring-revenue floor ($10k/mo = win) + v3.0 stress-test on 3 new verticals MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Cycle 29 (operator calibration): RECURRING > SIZE — a ~$10k/mo recurring business is a full win; one-time revenue is only a wedge. Folded into the Prime frame, L11, and SCORE. Cycle 30 (stress-test v3.0 on 3 new verticals, anti-tunnel): - NEW-1 KR e-commerce seller tools: SURVIVES-thin via rescue, low priority (T1 commodity + no founder distribution channel). - NEW-2 KR local-service booking/CRM: KILL (K19) on a REAL wall (priced-to- free vertical CRMs + 네이버 absorbs booking), after a failed rescue. - NEW-3 PIPA 처리방침 평가제 (SMB): REJECT on PUSH/no-teeth (G9/T7). - v3.0 behaved correctly in all 3 (no over-kill, killed on a real wall, rejected a teeth-less reg). CASH-F1 remains the lead. https://claude.ai/code/session_01A1BJYL2PvZh683Gh8KvuG8 --- CLAUDE.md | 42 ++++++++++++++++++++++++------------- LEDGER.md | 63 ++++++++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 89 insertions(+), 16 deletions(-) diff --git a/CLAUDE.md b/CLAUDE.md index 16cfa4a..ddd1271 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -36,14 +36,16 @@ ═══════════════════════════════════════════════════════════════ The goal is NOT to avoid bad ideas. It is to COMMIT to the best available idea by a deadline and produce a buyer-validated BUILD PLAN. Killing is a means, not the product. -**★ CAPTURABLE > BIG (v3.0 prime frame — read before every gate):** the win condition is a CAPTURABLE, - PROFITABLE slice that clears our bandwidth — even small revenue (~₩100-500M/yr) is a WIN, not a consolation - prize. We do NOT need to dominate or capture a whole market. NO IDEA IS A FLAWLESS UNICORN — they do not - exist; every real idea has competitors and challenges, and that is ACCEPTABLE. The question is NEVER "is this - uncontested / could this be a $30M venture?" — it is "can THIS 2-person team take a paying, defensible-enough - slice and ship revenue?" Kill an idea only when it cannot make MEANINGFUL revenue for US behind a STRUCTURAL - WALL — never because a competitor exists or because it won't be huge. (This frame overrides any gate that - smuggles unicorn-bias back in. MODE-MOAT still chases ceiling; MODE-CASH does NOT.) +**★ CAPTURABLE > BIG, and RECURRING > SIZE (v3.0 prime frame — read before every gate):** the win condition is + a CAPTURABLE, PROFITABLE slice that clears our bandwidth. **The PRIORITY is RECURRING revenue, not size — a + small RECURRING business is explicitly a FULL WIN. ~$10k/month (~₩13M/mo ≈ ₩150M/yr) of recurring revenue is a + target worth pursuing, not a consolation prize.** Lower floor confirmed by operator (cycle 29). We do NOT need + to dominate a market or hit $30M. NO IDEA IS A FLAWLESS UNICORN — competitors and challenges are EXPECTED and + ACCEPTABLE. The question is NEVER "is this uncontested / could this be huge?" — it is "can THIS 2-person team + take a paying, RECURRING slice?" Kill an idea only when it cannot make MEANINGFUL RECURRING revenue for US + behind a STRUCTURAL WALL — never because a competitor exists or because it won't be big. One-time/project + revenue is fine as a wedge, but a path to RECURRING is what promotes an idea. (Overrides any gate that smuggles + unicorn-bias back in. MODE-MOAT still chases ceiling; MODE-CASH optimizes recurring + speed + bandwidth-fit.) - FORCING FUNCTION: set a DECISION DATE at the start. On that date, the highest-scoring surviving idea is PROMOTED to BUILD PLAN even if imperfect. "Best by [date]" > "perfect, never." - BUILD QUOTA: at most 3 generate→gate→harden cycles before one idea MUST advance to buyer-validation. @@ -169,8 +171,9 @@ MONEY/100: ACVx3 · underserved-painx3 · WTP-EVIDENCEx3 (0 if no buyer signal) EXECUTION/70: Ace-demo-in-bandwidthx3 · data-cheap-day1x3 · named-warm-buyersx3 · no-license-gapx2 · defensible-12mox2 · low-burnx1. GRID: MoneyHi+ExecHi=GREEN(only with buyer signal) · mixed=YELLOW · LowLow=RED. -MODE re-weights: MODE-CASH triples time-to-rev + bandwidth **and ZEROES the ceiling + market-growth weights -(v3.0: a cash idea is NOT penalized for being small — capturable revenue that clears bandwidth scores full);** +MODE re-weights: MODE-CASH triples time-to-rev + bandwidth **+ RECURRING (v3.0: recurring is the priority metric) +and ZEROES the ceiling + market-growth weights — a cash idea is NOT penalized for being small; even ~$10k/mo +recurring scores full. A purely one-time-revenue idea with no recurring path is capped (it's a wedge, not a win);** MODE-MOAT triples ceiling + defensibility. **Never let a low ceiling pull a MODE-CASH idea to RED — that's the unicorn-bias the Prime Directive forbids. RED is reserved for "no capturable revenue / behind a WALL," not "small."** @@ -218,9 +221,11 @@ Boundary: managed service triggers 업무위탁 규정 + cannot replace the inte replaces); ACV still buyer-unvalidated. · **L10 (new v2.3) a documented TRUE-MOAT asset left UNUSED across cycles is a generation failure, not neutral: every cycle's GENERATE must run the moat-deployment sweep and bias an axis toward deploying an idle moat. (Trigger: David's site-selection IP sat unused 24 cycles while -the board mined a moat-less FI hunting ground.)** · **L11 (new v3.0, CAPTURABLE>BIG) a small, profitable, -capturable slice is a terminal WIN, not a consolation; no idea is a flawless unicorn; the test is "can we -take a paying slice?" not "is this uncontested / huge?" — small revenue that clears bandwidth counts.** · +the board mined a moat-less FI hunting ground.)** · **L11 (new v3.0, CAPTURABLE>BIG, RECURRING>SIZE) a small, profitable, +RECURRING slice is a terminal WIN, not a consolation; no idea is a flawless unicorn; the test is "can we take a +paying RECURRING slice?" not "is this uncontested / huge?" — even ~$10k/mo recurring counts (operator floor, +c29). Priority order: recurring > one-time; capturable-now > big-someday. A one-time job is only a WEDGE to a +recurring relationship.** · **L12 (new v3.0) kill WALLS, not SPEED-BUMPS: a competitor existing is a speed-bump needing an overcome- strategy, not a tombstone. Kill only behind a structural wall (capital/license/winner-take-all/true-commodity- no-edge/mass-market-no-distribution). "An incumbent exists" is the single most over-used false kill (K2→T2).** · @@ -236,7 +241,14 @@ MON pick MODE(s) + generate from rotating axis · TUE symmetric gate (kill+win s WED harden + bull/bear adversary · THU score + ledger + (every 8) META-LOOP audit · FRI VALIDATE: send 3-5 buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. -## CURRENT STATE (updated cycle 28 — May 30 2026; full detail in LEDGER.md) +## CURRENT STATE (updated cycle 30 — May 30 2026; full detail in LEDGER.md) +🧪 CYCLE 30 = STRESS-TEST v3.0 on 3 NEW verticals. v3.0 passed its own test: NEW-1 (e-comm seller tools) 🟡 + SURVIVES-thin via rescue (low priority — T1 commodity + no David channel, G5 weak); NEW-2 (local-service + booking/CRM 미용실·필라테스) 🔴 KILL K19 on a REAL wall (priced-to-free vertical CRMs + 네이버 absorbs booking — + not "incumbent exists"); NEW-3 (PIPA 처리방침 평가제 SMB) 🔴 REJECT on PUSH/no-teeth (G9/T7). None beat CASH-F1. + Repeatable winning shape confirmed = recurring + ROI-PULL + unowned AI-shaped job + a buyer David can reach. +🎯 CYCLE 29 = CALIBRATION (operator): win floor lowered — **RECURRING > SIZE; ~$10k/mo recurring is a FULL WIN.** + One-time revenue = a wedge only; a path to recurring promotes an idea. Folded into Prime frame + L11 + SCORE. 🧭 CYCLE 28 = GENERATE under v3.0. NEW LEAD: **CASH-F1** — F&B operator margin / cost-leak AI (invoice-OCR supplier-price-drift + delivery-commission-aware per-menu net profit + repricer) for KR independent/small-chain F&B ops, via David's network. MODE-CASH, Axis-D. Two deep-research passes CONVERGED here (revived K9 + fresh @@ -304,7 +316,7 @@ buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. exam wave, OR discovery Q0 proves software-WTP. The discovery kit can still be sent opportunistically — but the board NO LONGER WAITS on it. Only the L9 managed-service framing has a near-term (lumpy, services) cash path. 🟡 PARK (other): TAAMS-service · NuvoVet (GTM-blocked). -🔴 KILL: 19 ideas (see ledger; K9 REVIVED→CASH-F1; K15 partial-revive; re-audit column added). +🔴 KILL: 20 ideas (see ledger; K19 local-service CRM new c30; K9 REVIVED→CASH-F1; K15 partial-revive). NEXT ACTION (per Prime Directive): VALIDATE CASH-F1 via DAVID'S F&B network. ONE discovery question tests the whole cash portfolio: *"Last time 식자재 prices jumped, how did you decide whether/how much to reprice a menu, and what did you use?"* — listen for a painful manual process described BEFORE naming any product (a polite diff --git a/LEDGER.md b/LEDGER.md index b7511f5..e1ceada 100644 --- a/LEDGER.md +++ b/LEDGER.md @@ -1,6 +1,6 @@ # VENTURE LEDGER > Living state of the Founder's OS. Updated every cycle. Read alongside `CLAUDE.md`. -> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 27) | **OS version:** 3.0 +> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 30) | **OS version:** 3.0 --- @@ -75,6 +75,66 @@ Ran the G6 anti-gaming math on the narrowed segment *on its own*: --- +## CYCLE 30 — STRESS-TEST v3.0 across 3 NEW verticals (anti-tunnel) · May 30, 2026 +**Purpose:** operator asked to "go through new verticals and test this new framework." Ran v3.0 (deep comp-scan, +WALL-vs-SPEED-BUMP, rescue-before-kill, G9 PULL/PUSH, recurring-floor) on 3 deliberately DIFFERENT hunting +grounds. The point was as much to TEST the framework's calibration as to find a winner. Result: the framework +behaved correctly in all 3 (rescued one, killed one on a real WALL, rejected one on PUSH) — and **none beat the +current lead CASH-F1**, which is itself a useful comparative result. (Note: deep-research agents were rate-limited +this cycle; comp-scans done via direct search — flagged as MEDIUM depth, re-audit before any kill hardens, L13.) + +### NEW-1 · KR e-commerce seller tools (스마트스토어/쿠팡) — 🟡 SURVIVES THIN, low priority +- One-sentence: KR online sellers pay for keyword/listing optimization via 아이템스카우트/판다랭크 (subscription); + we'd do AI 상세페이지 + review-response generation under the 2025 algo shift (whole-store quality > keyword stuffing). +- COMP-SCAN: 아이템스카우트 OWNS keyword/product analytics (dominant, data-driven); 사방넷/플레이오토/이지어드민 own + order/inventory 통합관리. The keyword slice = speed-bump-toward-WALL. AI content-generation slice is more open BUT + generic LLMs do it natively → **T1 commodity risk**, and a 2-person team has **no distribution edge** (David's + network ≠ e-comm sellers → G5 weak). +- RESCUE: narrow to an AI 상세페이지+리뷰대응 bundle niche → SURVIVES, but thin moat + product-led distribution. +- VERDICT: 🟡 YELLOW, **low priority.** Not killed (rescue works), but ranks below CASH-F1 on G5 (no warm channel) + + moat. WIN-TYPE would be W6 (AI cost-advantage) only. Park unless a seller-channel appears. + +### NEW-2 · KR local-service booking/CRM (미용실·필라테스·PT) — 🔴 KILL on a real WALL (K19) +- COMP-SCAN: SATURATED with entrenched, cheap/FREE vertical CRMs — 바디코디 (fitness, 4,000+ centers, 1.7M users), + 공비서 (nail/hair, **free 알림톡, ₩0 가입비**), 핸드SOS, 비타민CRM, 어시스트핏, 이야기CRM, 뷰티인프라; **네이버 예약 + absorbs basic booking at 3.19%/no signup fee**; 카카오/마메드네 for hair. Pricing floor ≈ FREE. +- WALL test: commoditized + entrenched vertical incumbents + **platform (네이버) absorbs the core job** + near-zero + pricing + **no founder edge** = a genuine structural WALL (T1-commodity + platform-absorption), NOT a mere + "incumbent exists." RESCUE (AI no-show prediction / AI retention messaging) fails — incumbents already do 알림톡 + reminders + retention, and a me-too vertical CRM has no wedge or distribution. +- VERDICT: 🔴 **KILL → K19, type T1 (commodity + platform-absorbed + priced-to-free, no edge).** ✅ This is v3.0 + killing CORRECTLY — on a real wall after a rescue attempt, not on category-dismissal. (Contrast K9, revived.) + +### NEW-3 · KR PIPA 개인정보 처리방침 평가제 (SMB) — 🔴 REJECT on PUSH/no-teeth (G9/T7) +- COMP-SCAN: 처리방침 평가제 (since Sep-2023) lets PIPC **RECOMMEND** improvements; the carrot is 과태료 *감경* for + good 처리방침, not a direct SMB fine. **Deep evaluation targets ~50 LARGE companies** (AI/IoT/edutech), not SMBs. + And the 처리방침-generation job is already being templated/commoditized (free tools, even a Claude-Code skill). +- G9 verdict: for the SMB segment this is **PUSH / no-teeth (T7)** — no live forcing function compels an SMB to pay + for ongoing 처리방침 tooling; the big-company deep-eval segment has teeth-ish but is **Big-law/enterprise turf (T5)**. +- VERDICT: 🔴 **REJECT (don't even generate), type T7 + T5.** ✅ This is v3.0's G9 correctly flagging a regulation + with no teeth on FRESH non-FI ground — exactly the rule added after the 책무구조도 lesson. (Note: actual PIPA + *breach* enforcement IS real, but that's incident-response/security tooling — a different, walled market.) + +**CYCLE SYNTHESIS:** v3.0 passed its own stress-test — it did NOT over-kill (NEW-1 rescued), it killed on a REAL +wall (NEW-2), and it rejected a teeth-less reg (NEW-3). **CASH-F1 remains the lead** (only board idea with +ROI-PULL + an unowned slice + a real distribution channel via David). Funnel note: the repeatable winning shape +is "recurring, ROI-PULL, unowned AI-shaped job, buyer David can actually reach" — keep hunting that shape, not +generic SMB-SaaS where we have no channel. + +--- + +## CYCLE 29 — CALIBRATION (operator): recurring-revenue floor lowered · May 30, 2026 +**Operator input:** "I don't mind a ~$10k/month business — I just want anything RECURRING. Update it." +**Change (calibration within v3.0, no version bump):** the win floor is explicitly lowered and the PRIORITY is +reframed: **RECURRING > SIZE.** A ~$10k/mo (~₩13M/mo ≈ ₩150M/yr) recurring business is a FULL WIN, not a +consolation. One-time/project revenue is only a WEDGE; a path to RECURRING is what promotes an idea. Folded into: +Prime-Directive frame (CAPTURABLE>BIG + RECURRING>SIZE), L11, and SCORE (MODE-CASH now triples RECURRING and +caps purely one-time ideas). This sharpens — does not replace — the v3.0 anti-overkill rebuild. +**Implication for the board:** CASH-F1 (₩30–50k/mo × few-hundred = recurring ✅) fits cleanly; CASH-F3 (managed +service) must show a RETAINER not a one-time setup to count; EXPAND-1's parked consulting residue is one-time → stays parked. + +--- + ## CYCLE 28 — GENERATE under v3.0 (new lead) + secondary cash portfolio · May 30, 2026 **Hunting ground:** David's F&B/retail operator network (AXIS-D buyer-back). Two independent deep-research passes (the over-kill re-audit's K9 revival + a fresh 2nd-vertical scan) CONVERGED on the same unowned, @@ -586,6 +646,7 @@ Tiny KR F&B importers source over KakaoTalk and spreadsheets; a human-in-the-loo | K16 | TAAMS as platform | T2/T1 | Tridge+Tradlinx+free gov data own all layers end-to-end | WALL ✓ (end-to-end owned) | | K17 | 중대재해처벌법 (5-49인) safety evidence SaaS | T1/T5 | Free KOSHA tooling + licensed 안전관리대행 agencies own it; no founder distribution to ~800K SMEs (cycle 18 anti-tunnel test) | WALL ✓ (free gov + T5 no distribution) | | K18 | SME bookkeeping → lending-signal data | T2/T3 | 캐시노트/한국신용데이터 owns it — 1.7M sites (88%), ₩1.1조 unicorn, already runs CB on the data (cycle 20 anti-tunnel test) | WALL ✓ (T3 data-scale wall) | +| K19 | KR local-service booking/CRM (미용실·필라테스·PT) | T1 | Entrenched cheap/FREE vertical CRMs (바디코디·공비서·핸드SOS) + 네이버 예약 absorbs booking + priced-to-free + no founder edge (cycle 30 v3.0 test) | WALL ✓ (commodity + platform-absorbed; rescue failed) | --- From 379c5bf1e1104040cab024c51131c7524ba1ba61 Mon Sep 17 00:00:00 2001 From: Claude Date: Sat, 30 May 2026 11:34:30 +0000 Subject: [PATCH 5/5] Cycle 31: deep-research round 2 (connections + greenfield) -> new co-leads MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Ran 4 parallel deep-research agents. 3 new contenders beat the re-scoped CASH-F1; two are compliance plays that pass G9 because enforcement is LIVE (the gate the FI engine failed): - CASH-R1 'RAG-care' (NEW CO-LEAD): managed RAG-accuracy retainer via Ace's Coxwave channel. No wall, commoditization is a tailwind, complements Coxwave. ~₩13-25M/mo recurring. - CASH-FR1 필수품목 compliance trail (NEW CO-LEAD): franchise-HQ 사전협의·통지 via David's network. Documented fines (₩2.1B/₩94M) = real teeth. - CASH-G1 의료광고 AI 검수: strong teeth, but cold channel -> validate distribution first. - CASH-F1 re-scoped to 'ingredient-drift->repricing engine' (cost-tracking half is priced-to-free via 도도카트); ranked below the co-leads. - KILLS: K20 generic CRM (Naver Place absorbed it 2026-05-14), K21 학원 (priced-to-free). Synthesis: compliance works WHEN G9 teeth are live. STOP generating -> VALIDATE via our connections. Decision 2026-07-15. https://claude.ai/code/session_01A1BJYL2PvZh683Gh8KvuG8 --- CLAUDE.md | 42 ++++++++++++++++-------- LEDGER.md | 97 ++++++++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 124 insertions(+), 15 deletions(-) diff --git a/CLAUDE.md b/CLAUDE.md index ddd1271..7f8b3a9 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -241,7 +241,15 @@ MON pick MODE(s) + generate from rotating axis · TUE symmetric gate (kill+win s WED harden + bull/bear adversary · THU score + ledger + (every 8) META-LOOP audit · FRI VALIDATE: send 3-5 buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. -## CURRENT STATE (updated cycle 30 — May 30 2026; full detail in LEDGER.md) +## CURRENT STATE (updated cycle 31 — May 30 2026; full detail in LEDGER.md) +🧭 CYCLE 31 = GENERATE round 2 (4 deep-research agents: connection-anchored + greenfield). 3 NEW contenders beat + the re-scoped CASH-F1; 2 are compliance plays with LIVE teeth (the gate the FI engine failed). NEW CO-LEADS: + **CASH-R1 "RAG-care"** (managed RAG-accuracy retainer; Ace↔Coxwave channel; no wall, commoditization is a + TAILWIND, complements Coxwave; ₩13–25M/mo) + **CASH-FR1 필수품목 compliance trail** (franchise-HQ 사전협의·통지; + David's buyer; DOCUMENTED fines ₩2.1B/₩94M = real G9 PULL; ₩4.5–18M/mo). Strong-but-cold-channel: **CASH-G1 + 의료광고 AI 검수** (live 행정처분; no AI tool exists; but clinics ≠ David's F&B channel → validate distribution). + Re-scoped CASH-F1 → "ingredient-drift→repricing engine" (cost-tracking half is priced-to-free via 도도카트). + KILLS: K20 generic CRM (Naver Place absorbed it 2026-05-14), K21 학원 (priced-to-free). STOP generating → VALIDATE. 🧪 CYCLE 30 = STRESS-TEST v3.0 on 3 NEW verticals. v3.0 passed its own test: NEW-1 (e-comm seller tools) 🟡 SURVIVES-thin via rescue (low priority — T1 commodity + no David channel, G5 weak); NEW-2 (local-service booking/CRM 미용실·필라테스) 🔴 KILL K19 on a REAL wall (priced-to-free vertical CRMs + 네이버 absorbs booking — @@ -300,13 +308,19 @@ buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. governance 소명 is the #1 pain. Watch-risk = vendor-risk paradox. 🔴 CYCLE-18 KILL: 중대재해처벌법 5-49인 SME safety SaaS (K17, T1/T5) — anti-tunnel test confirmed leaving the FI buyer forfeits both moat (G2) and distribution (G5). 3 laws now converge on the same FI shape. -🟢 CASH-F1 (NEW LEAD, cycle 28) F&B operator margin/cost-leak AI — invoice-OCR price-drift + delivery-commission- - aware per-menu net profit + repricer. MODE-CASH, Axis-D, via David's F&B network. SPEED-BUMP; W1/W2/W3. G9 - ROI-PULL (enforcement-free). Capturable ~₩100–300M/yr. Validate-or-die (1 q): "last time 식자재 prices jumped, - how did you decide whether/how much to reprice, and what did you use?" 🔬 VALIDATE, 0/3, decision 2026-07-15. -🌱 SECONDARY CASH PORTFOLIO (same buyer = David's F&B/retail ops; one discovery tests all; build ≤1 pre-signal): - CASH-F2 franchise-HQ QSC/SV audit for sub-50-store brands · CASH-F3 managed review/reputation service · - EXPAND-1b HQ multi-store interior rollout-PM (narrow K15 revive, NOT a marketplace=captive-channel wall). +🟢 CASH-R1 (NEW CO-LEAD, cycle 31) "RAG-care" managed RAG-accuracy/maintenance retainer. Axis-B (Ace's AI × + Coxwave channel). No wall (commoditization = tailwind); COMPLEMENTS Coxwave (they detect, we fix). ~₩13–25M/mo. + Validate-or-die: "since your chatbot went live, last time it answered wrong — what did you DO to fix it?" 🔬 0/3. + OPEN Q for operator: will Coxwave refer customers down to us, or only lend credibility? (changes G5 strength.) +🟢 CASH-FR1 (NEW CO-LEAD, cycle 31) 필수품목 사전협의·통지 compliance trail for franchise HQs. Axis-D (David's buyer). + REAL G9 teeth — documented fines ₩2.1B/₩94M; recurs on every 식자재 price change. ~₩4.5–18M/mo. W2/W3/W5. + Validate-or-die: "지난번 필수품목 공급가 올렸을 때 가맹점 사전협의·통지 어떻게 처리하셨어요?" Bear = teeth latent vs felt. 🔬 0/3. +🔬 CASH-G1 (cycle 31, channel-gated) 의료광고 AI 검수 for 비급여 clinics. LIVE 행정처분; no AI tool exists; ~₩7.5–25M/mo. + STRONG teeth but clinics ≠ David's F&B channel → validate DISTRIBUTION first (friendly 원장 / association / agency W4). +🔬 CASH-F1 (re-scoped cycle 31, ranked below co-leads) → "ingredient-drift→repricing engine" (drop cost-leak framing; + 도도카트 gives cost-tracking free + 0.61% feel delivery-commission pain). MODE-CASH, David's F&B network. 🔬 0/3. +🌱 SECONDARY (build ≤1 pre-signal): CASH-G2 정부지원 R&D 정산 (lumpy recurrence, dev's startup network) · Coxwave-#2 + AI build-&-run retainer (voucher-funded; watch T5 enterprise-sales) · CASH-F2 QSC · CASH-F3 review service · EXPAND-1b. 🟡 PARK — EXPAND-1 (site-selection/forecasting, parked cycle 27): T3 data WALL — 오픈업 프로/Finda (Oct-2025, ~₩825k /mo) on NICE card-data; software play walled, only a David-consulting residue survives. Revive only if 오픈업 fails on accuracy → a service-led "interpret the data for your 출점 committee" play. @@ -316,9 +330,9 @@ buyer messages for any 🔬 idea. Decision date enforced. Build quota enforced. exam wave, OR discovery Q0 proves software-WTP. The discovery kit can still be sent opportunistically — but the board NO LONGER WAITS on it. Only the L9 managed-service framing has a near-term (lumpy, services) cash path. 🟡 PARK (other): TAAMS-service · NuvoVet (GTM-blocked). -🔴 KILL: 20 ideas (see ledger; K19 local-service CRM new c30; K9 REVIVED→CASH-F1; K15 partial-revive). -NEXT ACTION (per Prime Directive): VALIDATE CASH-F1 via DAVID'S F&B network. ONE discovery question tests the - whole cash portfolio: *"Last time 식자재 prices jumped, how did you decide whether/how much to reprice a menu, - and what did you use?"* — listen for a painful manual process described BEFORE naming any product (a polite - "sounds useful" = ZERO). Same conversation can probe CASH-F3 (review retainer) + CASH-F2 (QSC). Build ≤1 thing - pre-signal. Decision date = 2026-07-15. (FI engine + EXPAND-1 parked; do not build either.) +🔴 KILL: 22 ideas (see ledger; K20 generic CRM platform-absorbed + K21 학원 priced-to-free, new c31). +NEXT ACTION (per Prime Directive — STOP generating, idea supply now ample, 0/3 signals): VALIDATE in 2 parallel + tracks via OUR CONNECTIONS. (1) Ace probes CASH-R1 via Coxwave-network / AI-agency teams: "since your chatbot + went live, last time it answered wrong — what did you DO to fix it, who, how long?" (2) David probes CASH-FR1 + via franchise-HQ owners: "지난번 필수품목 공급가 올렸을 때 가맹점 사전협의·통지 어떻게 처리하셨어요?" CASH-G1 needs a + distribution test first (can David/Ace reach 3-5 clinic 원장?). Build ≤1 Phase-0 demo pre-signal. Decision 2026-07-15. diff --git a/LEDGER.md b/LEDGER.md index e1ceada..2e0491c 100644 --- a/LEDGER.md +++ b/LEDGER.md @@ -1,6 +1,6 @@ # VENTURE LEDGER > Living state of the Founder's OS. Updated every cycle. Read alongside `CLAUDE.md`. -> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 30) | **OS version:** 3.0 +> **Team:** Ace Yeom · David Mun | **Last updated:** May 30, 2026 (Cycle 31) | **OS version:** 3.0 --- @@ -75,6 +75,99 @@ Ran the G6 anti-gaming math on the narrowed segment *on its own*: --- +## CYCLE 31 — GENERATE round 2: connection-anchored + greenfield (deep research, v3.0) · May 30, 2026 +**Purpose:** operator asked to re-run deep-research agents on the strongest verticals + explore NEW fields, +"make sure some use our connections." Ran 4 parallel deep-research agents (David's F&B/franchise network · +Ace's Coxwave/AI relationship · greenfield · deepen CASH-F1). Result: **3 new contenders that BEAT the +re-scoped CASH-F1**, two of them compliance plays with LIVE teeth (the gate the FI engine failed). All gated +under v3.0. Still 0/3 buyer signals — so this is the LAST generate round before validation (Prime Directive). + +### A) CONNECTION-ANCHORED WINNERS + +#### CASH-R1 · "RAG-care" — managed RAG-accuracy/maintenance retainer (Ace ↔ Coxwave channel) — 🟢 NEW CO-LEAD +> KR companies that bought/built a RAG chatbot pay to KEEP IT ACCURATE today via ad-hoc dev firms or nobody; +> we do a monthly retainer (re-index changed docs, regression-test answers, hallucination triage, accuracy +> report); recurring **~₩1.5–4M/mo × 5–8 clients = ₩13–25M/mo**; MODE-CASH, Axis-B (Ace's AI skill × Coxwave channel). +- COMP-SCAN: build agencies (메이크봇/올거나이즈) chase new projects + neglect post-launch; eval tools (Coxwave + Align/Langfuse/Galileo) DETECT drift but hand you a dashboard, not the fix. The recurring *fix-work* is unowned. +- WALL-vs-SPEED-BUMP: SPEED-BUMP. **Foundation-model commoditization is a TAILWIND** (more cheap deployments → more + to maintain). **Complements Coxwave (they detect, we fix) — does NOT compete with their roadmap.** WIN-TYPE W3+W2. +- G9: ROI-PULL (a wrong-answer bot = real business cost / abandoned project) + evergreen. G5: Coxwave referral + + agencies + AI-Voucher pool. **OPEN QUESTION for operator: will Coxwave refer customers DOWN to us, or only lend + credibility?** (materially changes G5 strength.) G8: productizable as fixed-scope SLA → fits 10–13 hr/wk. +- Validate-or-die: *"Since your chatbot went live, last time it gave a wrong/outdated answer — what did you DO to + fix it, who did it, how long?"* (manual pain / "we just live with it" = strong). **Status: 🔬 VALIDATE, 0/3.** + +#### CASH-FR1 · 필수품목 사전협의·통지 compliance trail (David ↔ franchise-HQ network) — 🟢 NEW CO-LEAD +> KR franchise HQs (sub-50-store) pay to avoid 가맹사업법 필수품목 violations today via 가맹거래사/law firms + +> 모두싸인 + Excel; we do a change-triggered "사전협의 record + notify all 가맹점 + audit trail" workflow; recurring +> **~₩150–300k/mo × 30–60 brands = ₩4.5–18M/mo**; MODE-CASH, Axis-D (David's exact buyer). +- COMP-SCAN: 외식UP/Maxvisor ERPs ignore this legal workflow; 모두싸인 = signing only (unbundle target); substitute + = expensive per-change 가맹거래사 review. No cheap, recurring, change-triggered tool for SMALL brands. +- **G9: ROI/enforcement-PULL with DOCUMENTED TEETH** — 2024 가맹사업법 amendment makes 사전협의+통지 mandatory on every + 필수품목 change; regulator already fined a donut HQ **₩2.136B** + a 족발 HQ **₩94M**. Recurs on every 식자재 price move. +- WALL-vs-SPEED-BUMP: SPEED-BUMP. WIN-TYPE W2+W3+W5. Service-wrap entry (David sells "we keep you audit-ready"). +- BEAR (validate first): teeth may be LATENT (small HQs ignore until a 공정위 letter) vs actively felt. +- Validate-or-die: *"지난번 필수품목 공급가 올렸을 때 가맹점 사전협의·통지 어떻게 처리하셨어요? 누가, 뭘로?"* **🔬 VALIDATE, 0/3.** + +### B) GREENFIELD WINNERS (strong on teeth, WEAK on warm channel — distribution is their bear, G5) + +#### CASH-G1 · 의료광고 컴플라이언스 AI 검수 (clinic medical-ad pre-screen) — 🔬 VALIDATE (channel-gated) +> KR 비급여 clinics pay for 의료법 ad-safety today via ₩800k–1M/mo 마케팅 대행사 + manual 자율심의; we do an AI checker +> that flags 위반 문구/이미지 BEFORE posting + tracks 심의 obligations; recurring **~₩100–300k/clinic/mo (50–100 = ₩7.5–25M/mo)**. +- COMP-SCAN: incumbents = marketing agencies (라온하제 580 clinics, 애드리절트 1272) bundling "review" into ₩800k+ + retainers; 자율심의기구 do manual per-submission review. **No dedicated AI self-check tool exists** (searched). +- G9: STRONG ROI-PULL — LIVE enforcement (366 illegal ads caught Dec-2023–Feb-2024 → 행정처분 + 형사고발). W2+W6. +- **G5 WEAK: clinics are NOT David's F&B channel** → needs a new channel (friendly 원장 / 비급여 association / agency + reseller W4). This is the #1 watch-risk. Validate DISTRIBUTION before building. +- Validate-or-die: *"마지막으로 비급여 후기/광고 올릴 때 누가 의료법 검수했고, flag 났을 때 어떻게 됐어요?"* **🔬 VALIDATE (channel-gated), 0/3.** + +#### CASH-G2 · 정부지원사업/R&D 정산·사후관리 (RCMS 증빙 automation) — 🌱 GENERATED (secondary) +> SMBs running multiple gov grants pay 회계법인/컨설턴트 for RCMS 정산·증빙 or eat 환수 risk; we do AI evidence-to-세목 +> organization + deadline tracking; ~₩100–300k/co/mo WHILE projects active. Teeth real (환수). +- LIMITATION: recurrence is LUMPY (project-duration-bounded, not evergreen) → ranks below evergreen plays. Reachable + via the dev's own startup/TIPS network (warm-ish). Confidence MEDIUM. Keep as secondary. + +### C) EXISTING LEAD RE-SCOPED (from the CASH-F1 deepening agent) +**CASH-F1 → re-scoped + downgraded.** Deep scan found the cost-tracking half is **priced-to-free** (도도카트 free, +~100k users, marketplace-subsidized) and "savings-finder" analytics WTP is weak (F&B ops pay for things that MOVE +money — POS/settlement/캐시노트 ₩16.5–55k tiers). Worse, the delivery-commission angle (my "novel" piece) is the +LEAST-felt pain (survey: 90.3% reprice on ingredient cost vs **0.61%** on delivery commission). **Re-scope: from +"cost-leak finder" → "ingredient-drift→repricing DECISION engine"** (the action incumbents won't ship); demote +delivery-commission to a feature. Portfolio watch-risk = same priced-to-free bear that killed K19. **🔬 VALIDATE, +now ranked BELOW CASH-R1/CASH-FR1.** + +### D) ELIMINATED THIS ROUND (evidence-based — logged to prevent re-chase) +- **Generic 단골/CRM/멤버십 messaging → WALL (platform-absorbs-job):** Naver Place absorbed CRM/membership natively + **2026-05-14** (plugs in 도도포인트/페이히어). → K20. +- **식자재 group-buying/공동구매 → WALL (capital/logistics):** 마켓보로/푸드팡/오더플러스 run cold-chain + working capital. +- **학원 admin/parent-comms → WALL (priced-to-free):** 학원조아/클래스업/랠리즈 free, monetize 알림톡 markup. → K21. +- **물류 WMS · 제조 MES · 보험 GA tooling · 건물 관리비 · generic 미수금(청구스) · 세무사 payroll core →** walls/commodity + (SI-heavy, license-adjacent, or priced-to-free). 세무사 *intake-only* survives only as a thin 12-mo-head-start play + (더존 위하고-T is absorbing it) — not pursued. +- **AI 기본법 compliance ops + 식당 HACCP →** PUSH/no-teeth (G9/T7): AI법 has a ≥1-yr 계도기간 (no fines til ~2027); + restaurant 위생등급제 is voluntary. Both PARKED as ~2027 revive / bundle add-ons, not leads. ✅ G9 working again. + +### E) SYNTHESIS + DECISION +- **REPEATABLE WINNING SHAPE (now strongly confirmed across 3 winners):** recurring + **ROI/enforcement-PULL with + LIVE teeth** + an **unowned AI-shaped job** + **a buyer we can actually reach**. +- **REFINEMENT (compliance is fine — NO-TEETH was the FI killer, not "compliance"):** two new winners (CASH-FR1, + CASH-G1) are compliance plays that PASS G9 because enforcement is LIVE + fines are documented. The FI engine + failed only because its teeth were 2 yrs away. → AXIS-A is productive *when* G9 PULL is real; verify enforcement + is live + fined, not pending. +- **CONCENTRATION/CHANNEL CHECK:** the two co-leads use DIFFERENT moats + DIFFERENT channels (CASH-R1 = Ace/Coxwave/AI; + CASH-FR1 = David/franchise/teeth) → healthy diversification, not single-point-of-failure. CASH-G1's shared bear = + no warm channel (G5). +- **LEAD RANKING (by gate-fit × channel-fit, MODE-CASH, recurring floor):** 1) **CASH-R1 (RAG-care)** — no wall, + tailwind, Ace's core skill, Coxwave channel; 2) **CASH-FR1 (필수품목)** — live teeth + David's exact buyer; + 3) **CASH-G1 (의료광고)** — best teeth but cold channel (validate distribution first); then CASH-F1 (re-scoped), + CASH-G2, CASH-F2/F3 secondary. +- **NEXT ACTION (Prime Directive — STOP generating, VALIDATE):** two cheap parallel discovery tracks — Ace probes + CASH-R1 via Coxwave-network/agency AI teams; David probes CASH-FR1 via franchise-HQ owners. One question each + (above). Decision date = **2026-07-15.** Build ≤1 Phase-0 demo pre-signal. + +--- + ## CYCLE 30 — STRESS-TEST v3.0 across 3 NEW verticals (anti-tunnel) · May 30, 2026 **Purpose:** operator asked to "go through new verticals and test this new framework." Ran v3.0 (deep comp-scan, WALL-vs-SPEED-BUMP, rescue-before-kill, G9 PULL/PUSH, recurring-floor) on 3 deliberately DIFFERENT hunting @@ -647,6 +740,8 @@ Tiny KR F&B importers source over KakaoTalk and spreadsheets; a human-in-the-loo | K17 | 중대재해처벌법 (5-49인) safety evidence SaaS | T1/T5 | Free KOSHA tooling + licensed 안전관리대행 agencies own it; no founder distribution to ~800K SMEs (cycle 18 anti-tunnel test) | WALL ✓ (free gov + T5 no distribution) | | K18 | SME bookkeeping → lending-signal data | T2/T3 | 캐시노트/한국신용데이터 owns it — 1.7M sites (88%), ₩1.1조 unicorn, already runs CB on the data (cycle 20 anti-tunnel test) | WALL ✓ (T3 data-scale wall) | | K19 | KR local-service booking/CRM (미용실·필라테스·PT) | T1 | Entrenched cheap/FREE vertical CRMs (바디코디·공비서·핸드SOS) + 네이버 예약 absorbs booking + priced-to-free + no founder edge (cycle 30 v3.0 test) | WALL ✓ (commodity + platform-absorbed; rescue failed) | +| K20 | Generic 단골/CRM/멤버십 messaging | T1 | **Naver Place absorbed CRM/membership natively 2026-05-14** (도도포인트/페이히어 plug-ins) → platform-absorbs-job (cycle 31) | WALL ✓ (platform-absorbed) | +| K21 | 학원 admin / parent-comms SaaS | T1 | Priced-to-FREE — 학원조아·클래스업·랠리즈 free, monetize 알림톡 markup (cycle 31 greenfield) | WALL ✓ (priced-to-free) | ---