spirewise 1.7.1 → 1.8.0

package/skills/nvidia-inception-idea-booster/README.md

@@ -1,13 +1,13 @@
 # nvidia-inception-idea-booster
 
-Reads your **existing startup-idea files** in the project and elevates the idea
-into a stronger, defensible, **NVIDIA-Inception-ready** version — using exactly
-the traits NVIDIA prefers (deep tech, physical/agentic AI, essential GPU
-acceleration, defensibility, large markets, NVIDIA ecosystem fit). It elevates
-your real idea; it does not invent a different company.
+Reads your **existing startup-idea, pitch, PRD, business-plan, and profile files**,
+researches what NVIDIA Inception actually rewards, **scores the idea 0–100**, then
+**rewrites the idea and its documents** into a stronger, **NVIDIA-Inception-ready**
+version with a higher chance of acceptance — **without changing your startup name
+or core vision**. It elevates your real idea; it never invents a different company.
 
-> Pairs with **`nvidia-inception-starter`**: boost the idea here, then score and
-> audit it there.
+> Pairs with **`nvidia-inception-starter`**: boost + rewrite the idea here, then
+> score and audit it there.
 
 ## Install
 
@@ -28,28 +28,47 @@ npx spirewise remove -s nvidia-inception-idea-booster -a claude,cursor -sc both
 ## Use
 
 Put your idea notes anywhere in the project (`idea.md`, `README`, `pitch`,
-`vision`, `notes`, a one-pager, etc.), then ask your agent:
+`prd`, `business-plan`, `vision`, `profile`, a one-pager, etc.), then ask your agent:
 
-> "Make my startup idea stronger for NVIDIA Inception"
+> "Boost my startup idea for NVIDIA Inception"
 
-It reads those files and writes **`nvidia-inception/elevated-idea.md`**.
+It reads those files, researches NVIDIA acceptance patterns, scores the idea, and
+writes a master report plus rewritten documents under `nvidia-inception/`.
+
+## What it guarantees
+
+- **Never changes** your startup name.
+- **Always preserves** your core founder vision.
+- **Additive only** — keeps every existing strength.
+- **Justifies every change** with a "why this raises acceptance odds."
+- Only adds improvements that increase NVIDIA Inception attractiveness; labels
+  anything not yet true as `[ASPIRATIONAL]`.
 
 ## Output
 
-`nvidia-inception/elevated-idea.md` — original idea (restated), an elevated
-one-liner and pitch, a **before → after** upgrade table across every
-NVIDIA-preference lever, a recommended NVIDIA stack (tech → use case → reason),
-a moat plan, a beachhead→expansion market path, a proof plan, and honest gaps.
+- `nvidia-inception/boosted-idea.md` — an 18-section master report: original
+  summary, **before→after 0–100 score**, acceptance-probability analysis, missing
+  requirements, NVIDIA alignment, upgraded concept, AI / Physical-AI / Agentic-AI /
+  enterprise expansion, GPU plan, technical architecture, data strategy, moat,
+  market plan, investor narrative, optimized pitch, and final score with a
+  per-change justification table.
+- `nvidia-inception/rewrites/` — rewritten **pitch, startup profile, PRD, business
+  plan, and summaries**, each preserving your name + vision (originals untouched).
 
-## NVIDIA-preference levers it optimizes
+## Scoring engine (0–100, weighted)
 
-| Lever | What it pushes for |
+| Category | Weight |
 |---|---|
-| Deep tech / IP | Real model/system/hardware differentiation, not an AI wrapper |
-| Physical / Agentic AI | Robotics, autonomy, embodied/edge, action-taking agents |
-| GPU essential | Workloads that genuinely require accelerated compute |
-| Defensibility | Proprietary data/model/hardware or NVIDIA-stack integration |
-| Market & why-now | Large TAM, clear beachhead, commercialization path |
-| Ecosystem fit | Authentic use of CUDA, Isaac, Omniverse, Jetson, NIM, etc. |
-
-See `SKILL.md` for the full method and output template.
+| AI Depth | 20% |
+| GPU Utilization Potential | 15% |
+| Technical Innovation | 15% |
+| Scalability | 10% |
+| Enterprise Potential | 10% |
+| Physical AI Potential | 10% |
+| Data Flywheel | 5% |
+| Market Size | 5% |
+| Defensibility | 5% |
+| Alignment w/ NVIDIA Strategic Areas | 5% |
+
+See `SKILL.md` for the research phase, transformation framework, 14-step rewrite
+process, and the full output template.

package/skills/nvidia-inception-idea-booster/SKILL.md

@@ -1,151 +1,295 @@
 ---
 name: nvidia-inception-idea-booster
 description: >-
-  Read a startup's existing idea/notes files in the project and elevate the idea
-  into a far stronger, NVIDIA-Inception-ready version — leaning on exactly the
-  traits NVIDIA prefers (deep tech, physical/agentic AI, essential GPU
+  Read a startup's existing idea/pitch/PRD/business-plan/profile files and
+  transform the idea into a far stronger version with a significantly higher
+  probability of acceptance into the NVIDIA Inception Program — WITHOUT changing
+  the startup name or abandoning the founder's core vision. Runs a NVIDIA-acceptance
+  scoring engine (0–100) using a baked-in NVIDIA knowledge base (so live research is
+  optional and token-cheap), identifies gaps, then rewrites the
+  idea docs, pitch, profile, PRD, business plan, and summary toward the traits
+  NVIDIA rewards (deep tech, Physical/Agentic/Generative AI, essential GPU
   acceleration, defensibility, large markets, NVIDIA ecosystem fit). Use when the
-  user asks to "make my startup idea stronger for NVIDIA", "elevate my idea",
-  "improve my startup for Inception", or "rewrite my idea to fit NVIDIA". Writes
-  an elevated idea document to an `nvidia-inception/` folder in the project root.
-  For scoring/auditing an existing company, use nvidia-inception-starter instead.
+  user asks to "boost my idea for NVIDIA Inception", "make my startup
+  Inception-ready", "improve my startup for NVIDIA", or "rewrite my pitch for
+  NVIDIA". Writes outputs to an `nvidia-inception/` folder in the project root.
+  For pure scoring/auditing of an existing company, use nvidia-inception-starter.
 ---
 
 # NVIDIA Inception Idea Booster
 
-Take whatever startup idea already exists in the project, understand it deeply,
-then **transform it into a stronger, defensible, NVIDIA-aligned idea** that maps
-to what NVIDIA Inception actually rewards — without inventing a different company.
+Take whatever startup idea already exists in the project, apply the baked-in
+knowledge of what NVIDIA Inception rewards (live research optional), **score it
+0–100**, then **rewrite the idea and every related document into a stronger,
+NVIDIA-aligned version** with a materially higher
+chance of acceptance — while keeping the **same startup name** and the **same core
+founder vision**.
+
+## Non-negotiable rules (read first)
+
+1. **NEVER change the startup name.** It carries through every rewrite verbatim.
+2. **ALWAYS preserve the core founder vision** — the problem they care about, who
+   they serve, and why they started. Elevate it; never swap it for a trendier idea.
+3. **ALWAYS improve the idea** using real NVIDIA Inception acceptance patterns.
+4. **ALWAYS rewrite the related documents** — idea docs, pitch, startup profile,
+   PRD, business plan, and one-line/short summaries — not just a separate report.
+5. **ALWAYS justify every modification** with a one-line "why this raises
+   acceptance odds."
+6. **NEVER remove an existing strength.** Additive only — keep what already works.
+7. **ONLY add improvements that increase NVIDIA Inception attractiveness.**
+8. **Elevate, don't fabricate.** Strengthen the real idea. Mark anything not yet
+   true as `[ASPIRATIONAL]` and anything unknown as `[UNKNOWN]` — never present a
+   wish as a fact.
+
+Rules 3 and 8 work together: be aggressive about adding AI depth and NVIDIA fit,
+but label what is real vs. a planned/aspirational upgrade so the founder can decide.
 
 ## When to use vs. the sibling skill
 
-- **This skill** reshapes and strengthens the *idea* itself (positioning, wedge,
-  tech depth, NVIDIA fit), starting from the user's existing files.
-- **`nvidia-inception-starter`** scores and audits an existing company against
+- **This skill** strengthens and *rewrites* the idea + its documents and reports a
+  before/after acceptance score, starting from the user's existing files.
+- **`nvidia-inception-starter`** scores/audits an existing company against
   acceptance criteria and writes a readiness report.
 
-Use both together: boost the idea here, then score it there.
-
-## Step 1 — Read the existing idea
-
-Discover and read the startup-related files already in the project. Look for:
-- `idea*.md/.txt`, `README*`, `pitch*`, `business-plan*`, `vision*`, `notes*`,
-  `one-pager*`, `deck*`, `prd*`, `concept*`, `startup*`, and any `docs/` content.
-- Prefer Markdown/text/PDF-exported notes. Scan the repo for whatever describes
-  the problem, product, market, and tech.
-
-Extract and restate, in your own words:
-- Problem, who has it, and current alternatives
-- Proposed product/solution and how it works
-- Target market and business model
-- Any AI/ML, data, or compute involved
-- Team, stage, traction, and named tech stack
-
-If little exists, ask 3–5 focused questions before boosting. **Never discard the
-founder's core intent** — elevate it, don't replace it. Flag unknowns as
-`[UNKNOWN]` rather than inventing facts.
-
-## What NVIDIA prefers (the levers you optimize for)
-
-Pull the idea toward these — but only where it's authentic to the product:
-
-1. **Deep tech / novel IP** — real algorithmic, model, system, or hardware
-   differentiation; not an "AI wrapper" over someone else's API.
-2. **Physical & agentic AI** — robotics, autonomy, embodied/edge AI, perception
-   + navigation + actuation, or agents that take real actions. NVIDIA is leaning
-   hard into Physical AI.
-3. **GPU acceleration is essential** — the workload genuinely needs accelerated
-   compute (training/fine-tuning, large-scale inference, simulation, 3D, RL,
-   vision, LLMs) — GPUs are a *must-have*, not a nice-to-have.
-4. **Defensibility / moat** — proprietary models, a unique/proprietary dataset,
-   custom hardware, or deep integration with the NVIDIA stack for stickiness.
-5. **Large, impactful market** — autonomy, robotics, healthcare AI, advanced
-   manufacturing, smart cities, enterprise AI — clear, massive TAM and a path to
-   commercialize and scale globally.
-6. **Technical rigor & momentum** — prototypes, benchmarks, pilots, publications;
-   a credible, technical founding team.
-7. **NVIDIA ecosystem synergy** — would clearly use and benefit from NVIDIA
-   platforms (CUDA, TensorRT, Triton, NIM, NeMo, Riva, Metropolis, Holoscan,
-   Omniverse, Isaac, Jetson, DGX / NVIDIA AI Enterprise), and could collaborate
-   on reference designs / whitepapers / joint GTM.
-
-Excluded/weak profiles to steer away from: consulting/outsourcing, crypto
-mining/trading, pure reselling, thin AI wrappers with no real compute need.
-
-## Step 2 — Elevate the idea
-
-For each lever, produce a concrete upgrade to the existing idea:
-- **Sharpen the wedge:** a specific, painful, narrow beachhead use case where
-  the product wins decisively before expanding.
-- **Deepen the tech:** identify where to add genuine deep-tech depth (a
-  proprietary model, a data flywheel, a simulation/edge component) so GPUs become
-  essential. Be honest about what is real vs. aspirational.
-- **Add a Physical/Agentic angle where credible:** reframe toward embodied/edge
-  or agentic execution if it fits the problem — don't force it.
-- **Build the moat:** name the defensible asset (data, model, hardware,
-  integration) and how it compounds over time.
-- **Right-size the market:** beachhead → adjacent expansion → full TAM, with the
-  "why now."
-- **Map NVIDIA tech authentically:** for each technical need, the specific NVIDIA
-  technology that serves it and why — tied to a real use case, never as buzzwords.
-
-Always keep two views: the **original idea** and the **elevated idea**, so the
-founder sees exactly what changed and why it raises acceptance odds.
-
-## Output — write the elevated idea
+Use both: boost the idea here, then score it there for a second opinion.
+
+## Phase 1 — Read the existing startup
+
+Discover and read the startup-related files already in the project:
+- `idea*`, `README*`, `pitch*`, `deck*`, `business-plan*`, `prd*`, `vision*`,
+  `notes*`, `one-pager*`, `concept*`, `startup*`, `profile*`, and `docs/` content.
+- Prefer Markdown/text/exported notes. Capture the **startup name exactly**.
+
+Extract and restate in your own words: the problem and who has it; current
+alternatives; the product/solution and how it works; target market and business
+model; any AI/ML, data, or compute; team, stage, traction, and named tech stack.
+List every existing **strength** explicitly so Rule 6 can protect them.
+
+If little exists, ask 3–5 focused questions before boosting. Flag gaps as
+`[UNKNOWN]`, never invent facts.
+
+## Phase 2 — Acceptance knowledge base (baked in; research OPTIONAL)
+
+A distilled NVIDIA Inception knowledge base is **embedded in this skill** below.
+**Use it as the default ground truth and do NOT browse the web for the basics** —
+this saves tokens and time. Only do live research when the user explicitly asks for
+the latest/verified data, the idea sits in a niche the baked-in patterns don't
+cover, or you want a fresh accepted-startup example to cite. When you do, fold in
+what's new and note "verified <date>"; otherwise proceed straight to scoring.
+
+**Optional sources (only if refreshing):** the official NVIDIA Inception site and
+eligibility page; NVIDIA startup **showcase**/**spotlight** companies; NVIDIA
+**Physical AI / Robotics / Agentic AI / Omniverse / Cosmos / NIM / NeMo / Riva /
+AI Enterprise / CUDA** pages; accepted-startup announcements; founder write-ups;
+and VC-backed Physical/Enterprise/Deep-Tech AI startups. Extract: acceptance
+patterns, common categories, preferred tech/workloads, technical-depth indicators,
+and acceptance vs. rejection reasons.
+
+### Baked-in knowledge base (default ground truth — no research needed)
+
+**Program facts & eligibility.** Free, equity-free, no fees; no deadline, rolling
+review (~few days to ~2 weeks). Must be incorporated, have a working website, use a
+business email, and employ ≥1 developer. Generally <10 years old, building AI / ML /
+data-science / accelerated-computing products (not consulting/outsourcing). Open to
+all funding stages. **Not eligible:** consulting/outsourcing shops, crypto
+(mining/trading), pure cloud resellers/distributors, public companies. **Tiers:**
+Emerging → Scale → Premier. **Benefits:** SDKs + dev tools, DLI training, preferred
+hardware pricing, partner cloud credits, Inception Capital Connect / VC Alliance
+investor access, co-marketing, badges, events.
+
+**What gets noticed.** Workloads where **GPUs are essential** (training/fine-tuning,
+large-scale inference, simulation, 3D/graphics, RL, vision, LLMs); **deep tech /
+novel IP**, not thin API wrappers; **Physical AI** (robotics, autonomy, embodied/
+edge, digital twins, simulation); **Agentic & Generative AI** that takes real
+actions; **Enterprise AI** with a clear buyer; a **data flywheel / moat**; a large
+market with a credible technical team.
+
+**Accepted-startup categories.** Healthcare/biotech & drug discovery · Robotics &
+autonomy · Manufacturing/industrial & digital twins · Generative AI / LLM apps &
+infra · Computer vision & video analytics · Agentic AI & copilots · Climate/energy ·
+Cybersecurity · Fintech/fraud · Retail · Agritech · Autonomous vehicles · Edge AI ·
+Data/RAG/vector platforms · Scientific computing/HPC.
+
+**NVIDIA ecosystem reference (tech → use → recommend when).** CUDA/cuDNN (GPU
+compute → custom kernels/training) · TensorRT (fast inference → low-latency prod) ·
+Triton (multi-model serving → high throughput) · RAPIDS (GPU data science → heavy
+ETL/analytics) · NIM (inference microservices → fast enterprise deploy) · NeMo
+(+Retriever/Guardrails) (build/customize/RAG LLMs → custom models) · Riva (speech →
+voice/transcription) · Metropolis (vision AI → video analytics) · Holoscan (sensor/
+edge → medical/industrial streams) · Omniverse (3D sim/digital twins) · Cosmos
+(world foundation models → sim data for robots/AV) · Isaac (Sim/Lab/ROS, GR00T →
+robotics/embodied) · Jetson (edge compute → on-device AI) · DGX/HGX (large-scale
+training) · NVIDIA AI Enterprise (supported platform → regulated enterprises).
+Domain stacks: BioNeMo (drug discovery), Clara (healthcare), DRIVE (AV), Morpheus
+(cybersecurity), Earth-2 (climate), Aerial (telecom), Maxine (media).
+
+**Strategic pull areas (2024–2026).** Physical AI & robotics · Agentic AI · Digital
+twins & simulation · Sovereign/enterprise AI · Healthcare/biotech · Manufacturing/
+industrial · Climate/energy · Cybersecurity · Scientific computing.
+
+**Common rejection signals.** GPUs only nice-to-have; AI-wrapper with no real
+compute; consulting/outsourcing; crypto; pure reselling; vague market; no technical
+depth or differentiation; ineligible category.
+
+## NVIDIA Inception Scoring Engine (0–100)
+
+Score the **original** idea, then the **boosted** idea, with the same rubric.
+Each category is scored 0–100, multiplied by its weight, and summed.
+
+| # | Category | Weight | Scores high when… |
+|---|---|---|---|
+| 1 | AI Depth | 20% | Real models/algorithms/training, not a thin wrapper |
+| 2 | GPU Utilization Potential | 15% | Workload genuinely requires accelerated compute |
+| 3 | Technical Innovation | 15% | Novel IP, architecture, or hard technical edge |
+| 4 | Scalability | 10% | Scales across customers/markets without linear cost |
+| 5 | Enterprise Potential | 10% | Clear enterprise buyer, budget, and deployment path |
+| 6 | Physical AI Potential | 10% | Robotics/autonomy/edge/digital-twin/simulation angle |
+| 7 | Data Flywheel | 5% | Usage compounds proprietary data that improves the product |
+| 8 | Market Size | 5% | Large, growing TAM with a real beachhead |
+| 9 | Defensibility | 5% | Moat: data, model, hardware, integration, switching cost |
+| 10 | Alignment w/ NVIDIA Strategic Areas | 5% | Maps to NVIDIA's current focus areas |
+
+**Acceptance probability bands (boosted score):** 0–39 Low · 40–59 Borderline ·
+60–74 Competitive · 75–89 Strong · 90–100 Exceptional. Always show the **per-
+category table for before and after**, the weighted totals, and the point delta.
+
+## Transformation framework
+
+For every idea, **attempt to add** these AI capabilities where authentic to the
+product (label additions `[ASPIRATIONAL]` until real):
+
+> Agentic AI · Generative AI · Computer Vision · Predictive AI · Enterprise AI ·
+> Multimodal AI · Autonomous Decision Systems · AI Copilots · Digital Twins ·
+> Synthetic Data · Simulation Layer · Real-time AI · Physical AI · AI
+> Infrastructure · AI Workflow Automation · Knowledge Graphs · AI Analytics
+
+And **boost the score when relevant** by mapping to NVIDIA tech and verticals
+(only where it serves a stated need, never as buzzwords):
+
+> NeMo · NIM · Omniverse · Cosmos · Riva · CUDA · TensorRT · Triton · Isaac ·
+> Jetson · Metropolis · Holoscan · GPU acceleration · Edge AI · Robotics ·
+> Industrial AI · Healthcare AI · Climate AI · Manufacturing AI · Supply Chain AI
+> · Cybersecurity AI
+
+## Idea rewrite process (run in order)
+
+1. Analyze the original startup (Phase 1).
+2. Calculate the original NVIDIA score (rubric above).
+3. Identify weaknesses (lowest-weighted-gap categories first).
+4. Identify missing NVIDIA-aligned opportunities.
+5. Generate the upgraded startup concept (same name, same vision).
+6. Generate an AI-enhanced product roadmap.
+7. Generate NVIDIA ecosystem integration opportunities.
+8. Generate the technical architecture.
+9. Generate the GPU utilization strategy.
+10. Generate the enterprise expansion strategy.
+11. Generate the moat/defensibility strategy.
+12. Generate the data strategy (flywheel).
+13. Generate the investor narrative.
+14. Generate the NVIDIA-Inception-optimized pitch, then recompute the score.
+
+## Outputs
 
 1. Create an `nvidia-inception/` folder in the project root if missing.
-2. Write `nvidia-inception/elevated-idea.md` (UTF-8 Markdown):
+2. Write the master report: **`nvidia-inception/boosted-idea.md`** with all 18
+   required sections below.
+3. Write **rewritten documents** into `nvidia-inception/rewrites/` — one file per
+   source artifact that exists (or that the founder needs), each preserving the
+   **startup name** and **core vision**:
+   - `pitch.md`, `startup-profile.md`, `prd.md`, `business-plan.md`,
+     `one-line-summary.txt`, `short-summary.txt`
+   - If a source doc existed, base its rewrite on that file; otherwise generate a
+     fresh NVIDIA-aligned version. Never overwrite the user's originals in place.
+
+### Master report template (`boosted-idea.md`)
+
+Use exactly these 18 section headings, in order:
 
 ```
-# Elevated Startup Idea — <Company / Working Name>
+# NVIDIA Inception Boost — <Startup Name (unchanged)>
 Generated: <YYYY-MM-DD>
-Sources read: <list the files you ingested>
+Sources read: <files ingested>
+Research: <"baked-in KB" by default, or live sources + "verified <date>" if refreshed>
 
-## Original Idea (as understood)
-<faithful restatement of the founder's current idea>
+## 1. Original Startup Summary
+<faithful restatement + explicit list of existing strengths (protected)>
 
-## Elevated One-Liner
-<one sharp sentence, NVIDIA-aligned, no hype>
+## 2. NVIDIA Inception Compatibility Score
+<before/after per-category table (0–100 each), weighted totals, point delta>
 
-## Elevated Pitch (problem → wedge → solution → why now)
+## 3. Acceptance Probability Analysis
+<before vs after band + the few changes that move the needle most>
 
-## NVIDIA-Preference Upgrades
-| Lever | Before | After (elevated) | Why it helps acceptance |
-|---|---|---|---|
-| Deep tech / IP | … | … | … |
-| Physical / Agentic AI | … | … | … |
-| GPU essential | … | … | … |
-| Defensibility / moat | … | … | … |
-| Market & why-now | … | … | … |
-| NVIDIA ecosystem fit | … | … | … |
+## 4. Missing Requirements
+<what's absent vs. what NVIDIA rewards, ranked by score impact>
+
+## 5. NVIDIA Alignment Opportunities
+<specific strategic-area and ecosystem fits, each tied to a real need>
+
+## 6. Upgraded Startup Concept
+<same name, same vision, stronger framing — what changed and why>
+
+## 7. AI Expansion Opportunities
+<which capabilities from the framework to add; real vs [ASPIRATIONAL]>
 
-## Recommended NVIDIA Stack (tech → use case → reason)
+## 8. Physical AI Expansion Opportunities
+<robotics / autonomy / edge / digital-twin / simulation angle, if credible>
 
-## Defensibility / Moat Plan
+## 9. Agentic AI Expansion Opportunities
+<agents that take real actions; copilots; autonomous decision systems>
 
-## Beachhead → Expansion Market Path
+## 10. Enterprise Expansion Opportunities
+<enterprise buyer, deployment, compliance, land-and-expand>
 
-## Proof Plan (prototype, benchmark, pilot, publication)
+## 11. GPU Utilization Plan
+<exact workloads needing acceleration; training/inference/sim; scale>
 
-## Risks & Honest Gaps        (incl. anything still [UNKNOWN])
+## 12. Technical Architecture
+<components, data flow, models, where NVIDIA stack plugs in>
 
-## Next Steps                  (so it's ready for nvidia-inception-starter scoring)
+## 13. Data Strategy
+<data sources, the flywheel, synthetic data, labeling, governance>
+
+## 14. Competitive Moat
+<defensible asset(s) and how they compound; switching costs>
+
+## 15. Market Expansion Plan
+<beachhead → adjacent → full TAM, with "why now">
+
+## 16. Investor Narrative
+<the fundable story: problem, wedge, momentum, vision, ask>
+
+## 17. NVIDIA Inception Optimized Pitch
+<tight pitch written for an NVIDIA reviewer; specific, no hype>
+
+## 18. Final NVIDIA Acceptance Score
+<final weighted score, band, and a justification table:
+ | Change | Section | Why it raises acceptance odds | Real / [ASPIRATIONAL] |>
 ```
 
+Every modification must appear in a justification (Rule 5), and the section-18
+table is where each change is defended.
+
 ## Quality bar
 
-- **Elevate, don't fabricate.** Strengthen the real idea; never swap it for a
-  trendier one or claim capabilities that don't exist.
-- Make GPU dependence and the moat *believable*, with specifics.
-- Recommend NVIDIA tech only where it authentically fits a stated need.
-- Keep the before/after explicit so the founder can choose what to adopt.
+- **Same name, same vision.** Check the name is byte-for-byte unchanged everywhere.
+- **Additive only.** Every original strength still appears in the boosted version.
+- **Believable depth.** Make GPU dependence, the moat, and the data flywheel
+  concrete; label aspirational pieces honestly.
+- **Authentic NVIDIA mapping.** Recommend a specific technology only where it
+  serves a stated need — tech → use case → reason, never buzzword lists.
+- **No banned hype** ("revolutionary", "world-class", "#1", "game-changer").
+  Specific numbers and named use cases beat adjectives.
 
 ## Verification checklist (run before finishing)
 
-1. You actually read the existing idea files and listed them under "Sources read".
-2. `nvidia-inception/elevated-idea.md` exists with every section above.
-3. The elevated idea preserves the founder's core intent.
-4. Every NVIDIA-preference lever has a concrete before → after upgrade.
-5. NVIDIA tech recommendations are tied to real use cases, not buzzwords.
-6. Honest gaps and `[UNKNOWN]`s are stated, not hidden.
+1. The startup **name is unchanged** in the report and in every rewritten document.
+2. The **core founder vision** is intact and visibly preserved.
+3. Phase-1 files were actually read and listed; the baked-in KB (or refreshed research) is reflected.
+4. `nvidia-inception/boosted-idea.md` exists with **all 18 sections**.
+5. Rewritten documents exist under `nvidia-inception/rewrites/` and preserve name + vision.
+6. The scoring engine shows **before and after** per-category scores, weighted
+   totals, and a delta, using the 10 weighted categories.
+7. Every original strength is retained (nothing removed).
+8. Every modification has a justification; aspirational items are labeled, unknowns marked `[UNKNOWN]`.
+9. NVIDIA tech is tied to real use cases; no buzzword stuffing; no banned hype.

package/skills/nvidia-inception-starter/README.md

@@ -1,9 +1,10 @@
 # nvidia-inception-starter
 
 An expert advisor that prepares, evaluates, and optimizes your startup for
-**NVIDIA Inception** acceptance — it researches the program and accepted
-companies, scores you **0–100**, audits website/product/AI/architecture/NVIDIA
-alignment, and writes a full readiness report with a 90-day plan.
+**NVIDIA Inception** acceptance — it ships with a **baked-in knowledge base** of the
+program and accepted companies (so research is optional and token-cheap), scores you
+**0–100**, audits website/product/AI/architecture/NVIDIA alignment, and writes a
+full readiness report with a 90-day plan.
 
 ## Install
 
@@ -27,8 +28,8 @@ After installing, ask your agent:
 
 > "Assess my startup for NVIDIA Inception"
 
-It collects your company details (and researches the program + accepted
-startups), then writes
+It collects your company details (using a **baked-in NVIDIA Inception knowledge
+base** — research is optional), then writes
 **`nvidia-inception/inception-readiness-report.md`** in your project root.
 
 ## Output