# 00 — Script Canon Audit · v3 script + v4 consolidation (Option B video, 10-page deck)

> Companion to [BRIEFING.html](./BRIEFING.html). Proves every spoken line traces to canon and the script carries zero forbidden phrases.
> Date: 2026-05-30 · **Script v3: 357 words, ~2:44 at 130 wpm**. **Deck v4: 10-page consolidated**, single source for both J-StarX SV Extended (deadline 6/22, deck + 3-min video) and Alchemist Enterprise B2B Course (deadline 5/31, deck only). **Video format: Option B** — talking head + lower-third text cards (no slide cutaways), based on research into Alchemist application-video conventions.

---

## Target program

**J-StarX Silicon Valley Extended Program (B214)** — JETRO. ~10 weeks: Tokyo bootcamp (Aug 17–28), pre-departure mentoring (Sep 7–25), **Silicon Valley on-site (Oct 5–30)**, follow-up from November. Pre-seed to Series A. Required deliverables: English entry form + English pitch deck + 3-min English pitch video (per the form). **Application deadline: 2026-06-22 17:00 JST** (secondary round window 6/23–7/13). The program's stated value (SV market validation + VC meetings + customer acquisition + mentoring) maps cleanly onto the script's three asks, so no rewording is needed on S15 — the program-neutral phrasing also covers any other program you apply to.

---

## Forbidden-claims grep result (v3 spoken text)

Ran a case-insensitive grep of the full spoken script against every EN phrase in
[`kashi/docs/canon/FORBIDDEN_CLAIMS_REGISTRY.md`](../../../kashi/docs/canon/FORBIDDEN_CLAIMS_REGISTRY.md)
(§1 HARM_VERDICT, §2 MARKETING_OVERCLAIM, §3 PREDICTION_OVERCLAIM, §4 NO_SIGNAL_FALSE_REASSURANCE, §9 LEGAL_OVERCLAIM):

```
RESULT: 0 forbidden-phrase matches — CLEAN
```

Also re-grepped against the jargon-removal list (`database layer | row-level security | RLS | longitudinal`) — **0 hits**. The spoken script and the deck share the same plain-language register.

### Phrases deliberately handled (not violations, but worth knowing)

| Spoken phrase | Why it is safe |
|---|---|
| "preventing workplace harassment" | Legal-concept reference tied to MHLW law (the §6.1 legal-citation shape). NOT the banned "harassment detected" / "detects harassment". |
| "restricts workplace emotion-recognition" | Describes the EU AI Act, not a Kashi capability. NOT the banned "emotion detected". |
| "isn't surveillance software" | Surveillance used in NEGATION — the canonical §3.3 move, not surveillance positioning. |
| "designed to align with Article 5" | Uses the canon-approved "designed to align with"; avoids banned "fully compliant" / "legally safe". |
| "Built into the data itself, not into the screen" | Plain-language equivalent of v1's "architecture not policy at the database layer". Same meaning, no jargon — matches deck S7. |
| "private mirror" | Canon-approved phrasing; avoids the banned "manager self-observation report". |
| "We never judge what your teams say" | Accurate (Kashi never judges on content); avoids the banned "we do not transcribe" sleight-of-hand. |
| "From an acceleration partner" | Program-neutral phrasing — script is reusable for any program, including J-StarX SV Extended. Matches deck S15 header. |

---

## Section → canon source map (cues updated for v3 deck order)

| § | Spoken claim | Canon source | Deck slide cue |
|---|---|---|---|
| 1 Hook + Cover + Problem | "Before anyone says a word, trouble in a team shows up." Then 6-yr SMB board, surveys came back without a flag, breakdown visible in meeting structure. | Cover tagline (deck S1) + Founder bio [10-application-narratives §8B](../ACCELERATOR_RESEARCH/10-kashi-application-narratives.md) | **SLIDE 1 → SLIDE 2** |
| 2 Existing tools | MHLW law mandates harassment prevention, expanded to SMB; surveys/hotlines/content classifiers/meeting AI depend on someone speaking up | Elevator pitch — [INVESTOR_NARRATIVE_CANON §1](../../../kashi/docs/canon/INVESTOR_NARRATIVE_CANON.md) | **SLIDE 3** |
| 3 Insight + solution | Kashi reads structure not content; turn timing, interruption, no-response; one meeting noise / many meetings pattern; floor-share drop, proposals repeatedly deferred; 30/90/180-day windows | §1 + deck S4/S5 verbatim | **SLIDE 4 → SLIDE 5** |
| 4 Three views + boundaries | Output bounded by role: executives cross-team, managers private mirror, members own pattern; content/scores/sentiment not shown; built into the data itself, not the screen | §1 (private weekly mirror, aggregate 5+) + §7.5 pillar 3 (architecture not policy, expressed in plain language) | **SLIDE 6 → SLIDE 7** |
| 5 Why now + Moat | Japan mandate expanded; EU AI Act Art.5; content-reading tools face exposure; deterministic, rule-based, no LLM in judgment path; no emotion or sentiment; content not stored for eval; align with Art.5 + APPI | §1 (regulatory posture) + §7.5 (pillars 1, 4) | **SLIDE 8 → SLIDE 9** *(was 14→13 in v2 deck)* |
| 6 Traction + Team | Pre-pilot, no paying logos; 908 tests + determinism-lock; simulated-meeting eval (patterns surface cleanly, zero false positives no-signal); demo tenant live; 50-meeting / 5000-turn stress test; Justine Acaylar, Keio, solo, Filipino-Japanese | [§2.3 traction caveats](../../../kashi/docs/canon/INVESTOR_NARRATIVE_CANON.md) + §8B bio | **SLIDE 10** *(was 8 in v2 deck; founder-intro stays on talking head)* |
| 7 Ask + Closer | Trust infrastructure, privacy-protective + regulation-aligned, travels to US/EU; three asks: NA market-fit validation, enterprise PoC introductions, AI-governance mentoring; closer tagline | [10-narratives §A program ask](../ACCELERATOR_RESEARCH/10-kashi-application-narratives.md) (program-neutralized) | **SLIDE 15 → SLIDE 16** |

---

## Slide cue map (v3 deck — 16 slides, 12 referenced in cutaways)

| Slide | Title (new position) | Was at | Used in cutaway? | Spoken section |
|---|---|---|---|---|
| 1 | Cover ("Before anyone says a word.") | 1 | ✓ open | §1 |
| 2 | Problem | 2 | ✓ | §1 |
| 3 | Existing tools | 3 | ✓ | §2 |
| 4 | Insight | 4 | ✓ | §3 |
| 5 | Solution | 5 | ✓ | §3 |
| 6 | Three views | 6 | ✓ | §4 |
| 7 | Boundaries | 7 | ✓ | §4 |
| **8** | **Why now** | was 14 | ✓ | §5 |
| **9** | **Moat / Differentiation** | was 13 | ✓ | §5 |
| **10** | **N=1 Traction** | was 8 | ✓ | §6 |
| **11** | **6-month plan** | was 9 | — deck only | — |
| **12** | **Initial customers** | was 10 | — deck only | — |
| **13** | **Market (TAM/SAM/SOM)** | was 11 | — deck only | — |
| **14** | **Business model** | was 12 | — deck only | — |
| 15 | Three asks | 15 | ✓ | §7 |
| 16 | Team + closer | 16 | ✓ | §7 |

**Cue sequence end-to-end: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 15, 16** — strictly ascending. Slides 11–14 are review-only content (the application committee reads the full PDF; the video skips them to keep narrative flow under 3:00).

---

## Numbers used (all canon, none fabricated)

| Number | Source | Notes |
|---|---|---|
| 908 tests + determinism-lock | INVESTOR_NARRATIVE_CANON §2.2 | shipped |
| zero false positives in no-signal scenarios | §2.3 | shipped product-page copy |
| 50 meetings / 5,000 turns stress test | §2.3 / founder positioning | approved |
| 5-person minimum (k-anonymity ≥5) — implicit in "aggregate patterns" | §7.5 pillar 3 | architecture |
| Demo tenant live in production | §2.3 / ops memory | true |
| 30 / 90 / 180-day windows | product canon | true |

**Deliberately NOT spoken in the video** (conservative, per investor-canon hedge): hard TAM/SAM/SOM figures (deck S13 shows ¥164.2B / ¥57.5B / ¥720M labelled "model — sizing under NDA"). Hard pricing (¥100k/¥400k/¥800k+) shown on deck S14, not spoken. The video stays defensible.

---

## Delta log

### v4 consolidation (2026-05-30 · late) — Option B video + 10-page deck
| Change | Why |
|---|---|
| Deck consolidated from 16 pages → **10 pages** | (a) Alchemist Enterprise B2B Course asks for "approximately 10 pages". (b) Single deck for both programs avoids version drift. Adjacent slides merged: Problem + Existing tools → S2; Insight + Solution → S3; Three views + Boundaries → S4; Traction + 6-month plan → S7; Market + Business model → S8; Three asks + Closer → S10. Canon unchanged. |
| 16-page deck (`deck-en-full-16slide.html`) marked **superseded** | Legacy reference only; 10-page deck is now the source for submissions. |
| 9-slide cutaway deck (`deck-en-9slide.html`) marked **superseded** | Not needed under Option B (talking head + text cards, no slide cutaways). |
| Video format: **Option B** (talking head + lower-third text cards, no slide cutaways) | Per Alchemist insider-guide research: application-video screen is about *the founder* (#1 selection criterion is team composition). Pure delivery + occasional text accents demonstrates the sales ability they're looking for in B2B founders. Slide-heavy cutaways feel like a sales/product video — weaker signal for an application. (Full-slide style is Demo Day convention, not application.) |
| `[SLIDE n]` cues in BRIEFING + teleprompter swapped for **`[CARD: …]`** markers | 6 text cards total over 3 minutes: opening lower-third name card · "Kashi" wordmark · "Reading the SHAPE, not the content" · "Limits built into the data — not the screen" · "EU AI Act · Article 5" · "908 tests · determinism lock" · final full-bleed tagline. |
| Spoken script byte-for-byte unchanged | The text card choice is editorial / production. The 357-word script content is identical to v3 trim. |
| S9 of the 10-page deck has a **LinkedIn URL placeholder** | Enterprise B2B Course application checklist explicitly requires "LinkedIn information of founders/managers". Awaiting Justine's URL to swap in before PDF export. |

### v3 script trim (2026-05-30 · earlier same day)
| Change | Why |
|---|---|
| Word count 431 → **357** (~17% trim) | Justine requested slower delivery for a non-technical audience. At 130 wpm the script now lands at ~2:44 with ~15s of pause headroom; at 140 wpm it lands at ~2:33 with 27s of headroom. |
| Dropped "Designed to align with Article 5, APPI-aware on retention." | Avoided making a compliance claim. The "Article 5" mention earlier in §5 is enough; not claiming alignment is safer than claiming it. Canon-clean (the v2 "designed to align with" was the safe alternative to "fully compliant" — removing the claim entirely is also safe). |
| Dropped "Japan HQ, the Philippines as a second base, Asia to the US." from §7 | Already trimmed in v2; kept removed in v3. Geography intent lives on deck S12. |
| Simplified §4: "These boundaries are built into the data itself — not into the screen, so they can't be lifted by changing a setting." → "The limits are built into the data — they can't be turned off." | Same meaning, shorter, more colloquial. |
| Simplified §5: "deterministic, rule-based detectors, no language model in the judgment path. No emotion or sentiment inference." → "rule-based, deterministic, no emotion or sentiment." | "No language model in the judgment path" was technically precise but jargonier. The shorter form preserves the moat without the engineering term. |
| Simplified §6: "bit-for-bit deterministic on transcript input, 908 tests, including a determinism-lock. On simulated meetings, it surfaced repeated interruption patterns…" → "908 tests, including a determinism lock. On simulated meetings, the patterns came through cleanly…" | "Bit-for-bit deterministic" was redundant with "determinism lock". Tightened. |
| Removed "MHLW" name from §2 (kept the legal concept) | "MHLW law" is opaque to a non-Japanese audience. The sentence "In Japan, preventing workplace harassment is now a legal obligation — even for small firms" carries the same regulatory weight without the acronym. |
| Removed "across many meetings" phrase from §3 product description | Still says "Across many meetings, it's a pattern" — that's enough; the redundant earlier "across many meetings" was cut. |
| Spoken script otherwise byte-equivalent to v2 narrative arc | No new claims, no fabricated numbers, no forbidden phrases. |

### v3 deck reorder (2026-05-30 · earlier same day)
| Change | Why |
|---|---|
| Slides 8–14 reordered (cues now ascend 1→16) | The script's narrative arc (problem → tools-fail → insight → solution → product → boundaries → why-now → moat → traction → ask → closer) is the correct pitch flow. The deck now follows it; cues no longer jump backward. |
| Section 5 cue: SLIDE 14 → 13 became SLIDE 8 → 9 | Why-now is now S8, Moat is now S9 (after the swap + 13→9 shift). |
| Section 6 cue: SLIDE 8 → 16 (face) became SLIDE 10 only | N=1 traction is now S10; the founder-intro line ("I'm Justine Acaylar…") stays on a talking-head shot — no cutaway to S16 mid-section. |
| Slides 11–14 (Plan / Customers / Market / Biz Model) became deck-only | They live in the static PDF for the committee's review but don't appear as cutaways in the 3-min video. |
| Spoken script byte-for-byte unchanged | No content rewrites. Only cue labels updated. |

### v2 script (2026-05-30) — for reference
| Change | Why |
|---|---|
| Opens with "Before anyone says a word, trouble in a team shows up." (was: "For six years…") | Echoes the deck cover slide; burns the message in immediately. Founder story still anchors the rest of §1. |
| "enforced at the database layer" → "built into the data itself, not into the screen" | Matches v2 deck S7's jargon-free phrasing. Same meaning. |
| "From JETRO and Alchemist" → "From an acceleration partner" | Program-neutral — script is reusable for any program. |
| Removed "Japan HQ, the Philippines as a second base, Asia to the US." | Tightened section 7 to gain pause headroom; the geography intent lives on deck S12 ("Japan first, then US/EU"). |
| 447 words / 2:53 → 431 words / 2:47 | ~13s of pause headroom under 3:00 (was ~7s). |

---

## Open item flagged for Justine

- The video script and the deck are now cue-aligned. The **9-slide cutaway deck** ([deck-en-9slide.html](./deck-en-9slide.html)) from v1 still exists in the folder but is **not used by the v2 script / v3 deck**. Leave it as a legacy backup or delete; it does not affect anything else.

---

## Production-process provenance

- v1 script generated 2026-05-29 via Workflow (3 hook drafts → judges → synthesis → adversarial audit → clean).
- v2 derived 2026-05-30 by main session, restructuring v1 to align with the v2 deck. Same canon sources; new cover-tagline opener; jargon stripped; program-neutralized.
- v3 deck reorder 2026-05-30 by main session, repositioning slides 8–14 so cues ascend monotonically.
- **v3 script trim 2026-05-30 (later same day), 431 → 357 words (~17% cut)** so Justine can deliver more slowly and clearly for a non-technical audience. Target pace 130 wpm = 2:44 with ~15s pause headroom. Same canon sources; no new claims.
- **v4 consolidation 2026-05-30 (late)**: deck collapsed 16 → 10 pages as the single source for both J-StarX SV Extended + Alchemist Enterprise B2B Course. Video format changed to Option B (talking head + 6 text cards, no slide cutaways), based on Alchemist application-video convention research. Script content unchanged.
- Final self-grep on the v3 script + v4 deck: **0 forbidden hits, 0 jargon hits, 0 program-name hits, 357 words, ~2:44 at 130 wpm**.

*End of audit · v3 deck.*