spirewise 1.7.1 → 1.8.0

package/skills/nvidia-product-inventor/SKILL.md

@@ -1,25 +1,47 @@
 ---
 name: nvidia-product-inventor
 description: >-
-  Read a startup's existing idea/notes files and invent 3–6 concrete PRODUCTS
-  (not marketing copy) that all stay in the SAME industry, category, and core idea
-  as the startup — variations/components of the one idea, never new categories.
-  Each product is engineered to score a perfect 10/10 on NVIDIA Inception
-  approval fit AND 10/10 on same-space alignment, gets a short
-  brandable name (4–10 letters), an explicit NVIDIA stack (physical/compute
-  hardware like GPUs first, then SDKs, then software), and a real 1080×1080 PNG
-  logo MARK (abstract symbol only — no letters, no wordmark, centered) on a solid background. Use when the user asks to "invent products", "ideate
-  products for NVIDIA", "what should we build for Inception", or "turn our idea
-  into products". Writes a `products_raw/` folder with one subfolder per product.
-  Builds on nvidia-inception-idea-booster and nvidia-inception-starter. Products
-  only — no website/landing copy.
+  Read a startup's existing idea/notes files and invent 4–6 highly strategic
+  PRODUCTS (not marketing copy) that stay in the SAME industry, category, market,
+  and founder vision — together forming a stronger multi-product AI ecosystem
+  around the startup that maximizes NVIDIA Inception acceptance. Every product is
+  NVIDIA-aligned (deep tech, GPU-essential, physical/agentic/enterprise AI,
+  defensible, scalable), gets a full spec (problem, AI + NVIDIA opportunity,
+  architecture, revenue model, GPU strategy, acceptance impact) and a complete
+  brand kit (primary/square logo 1080×1080, app icon 1024×1024, favicon, SVG,
+  transparent version, brief, AI + SVG prompts, brand guide, typography, palette).
+  Use when the user asks to "invent products", "ideate products for NVIDIA", "what
+  should we build for Inception", or "turn our idea into a product ecosystem".
+  Writes a `products_raw/` folder with one subfolder per product. Builds on
+  nvidia-inception-idea-booster and nvidia-inception-starter. Products only — no
+  website/landing copy.
 ---
 
 # NVIDIA Product Inventor
 
-Turn the startup's idea into **3–6 buildable products** that maximize NVIDIA
-Inception fit. This is a product/ideation skill — invent *what to build*, not
-marketing copy.
+Turn the startup's single idea into a **4–6 product ecosystem** — complementary,
+buildable products that all live in the **same industry, category, and market**,
+serve the **same founder vision**, and together make the startup read to an NVIDIA
+Inception reviewer as a **long-term AI platform, not a one-feature tool**.
+
+This is a product/ideation skill: invent *what to build* and design its brand —
+never website or marketing copy.
+
+## Non-negotiable rules (read first)
+
+1. **NEVER change the startup name.**
+2. **NEVER change the startup's industry.**
+3. **NEVER change the startup's category.**
+4. **NEVER generate unrelated products.**
+5. **ALWAYS stay within the startup's market.**
+6. **ALWAYS generate products that strengthen the startup's ecosystem** (they
+   complement each other and the core product).
+7. **ALWAYS optimize each product for NVIDIA Inception attractiveness.**
+8. **ALWAYS give each product strong, real AI potential.**
+9. **ALWAYS justify why each product raises NVIDIA Inception acceptance probability.**
+
+Elevate, don't fabricate: strengthen the real domain; mark anything not yet true as
+`[ASPIRATIONAL]` and unknowns as `[UNKNOWN]`.
 
 ## Uses the sibling skills
 
@@ -27,268 +49,264 @@ marketing copy.
   (deep tech, physical/agentic AI, GPU-essential, defensibility, big market,
   ecosystem fit) when shaping every product.
 - Pre-screen each product with the **`nvidia-inception-starter`** acceptance lens
-  (AI strength, technical innovation, NVIDIA alignment, market, moat). Keep only
-  products that would plausibly score well; discard weak/"AI-wrapper" ones.
-
-## What NVIDIA Inception actually approves (research-backed)
-
-NVIDIA Inception is free, non-dilutive, rolling-admission, and **industry-agnostic
-within emerging/accelerated-computing tech**. Engineer every product so it would
-read as an obvious **accept** to a reviewer. Approved startups consistently show:
-
-1. **Deep-tech, GPU-essential core.** The product solves a hard technical problem
-   (computer vision, LLMs/agents, RL, simulation, graphics, robotics, genomics,
-   edge AI) that **substantially benefits from — or is impossible without — GPU
-   acceleration**. State plainly why it can't be built (or would be far worse) on
-   CPUs alone.
-2. **Authentic NVIDIA-platform alignment.** Concrete, credible use of NVIDIA
-   hardware + SDKs (CUDA, TensorRT, Triton/NIM, Isaac, Omniverse, Metropolis,
-   Holoscan, NeMo, Riva, RAPIDS…). The product should **drive demand for NVIDIA
-   compute** and act as a reference workload.
-3. **Physical AI / Agentic AI lean.** Robotics, autonomous systems, industrial
-   automation, digital twins, embodied AI (physical AI) and/or autonomous
-   decision-making agents/copilots (agentic AI). NVIDIA strongly favors these.
-4. **Defensible moat.** Proprietary IP, unique/large proprietary datasets, deep
-   technical barriers, or hardware integration — not easily replicated.
-5. **Real market + scalability.** Clear customer ROI, a sharp wedge, and a model
-   that scales on GPU cloud or edge fleets. Pilots/traction are a plus to note.
-6. **Research-to-product trajectory.** Credible technical depth (papers/SOTA
-   methods) paired with a concrete, buildable product — not a science project.
-
-### Hard disqualifiers — never generate products that read as these
-- **AI wrappers / thin API resellers** with no GPU-essential core or moat.
-- **Consulting / outsourced-dev** offerings (sell a product, not services).
-- **Crypto/mining/trading**, pure **cloud-reselling**, or generic
-  **reseller/distributor** plays.
-- Anything where the GPU is a "nice-to-have," not a must-have.
+  (AI strength, technical innovation, NVIDIA alignment, market, moat) and discard
+  weak/"AI-wrapper" ideas.
+
+## Research: baked in (default) — live research OPTIONAL
+
+The NVIDIA Inception acceptance knowledge needed here is **embedded below**. **Use
+it as the default and do NOT browse the web for the basics** — this saves tokens.
+Only research live when the user asks for the latest data, the startup sits in a
+niche not covered, or you want a fresh accepted-startup example to cite (then note
+"verified <date>"). Optional sources if refreshing: official NVIDIA Inception site
+& eligibility; NVIDIA startup showcases/spotlights; NVIDIA NIM/NeMo/Omniverse/
+Cosmos/Riva and CUDA/TensorRT pages; Physical-AI, robotics, and enterprise-AI
+initiatives; accepted-startup announcements; founder write-ups; investor-backed and
+deep-tech AI ecosystems. Extract product/technology/market/acceptance/innovation
+patterns and AI / physical-AI / enterprise / scalable-business opportunities.
+
+### Baked-in knowledge base (default ground truth)
+
+- **Program:** NVIDIA Inception is free, equity-free, rolling-admission, open to all
+  funding stages; for incorporated startups (<~10 yrs, ≥1 developer, working site,
+  business email) building AI / ML / data-science / accelerated-computing products.
+- **What gets approved:** a **GPU-essential deep-tech core** (vision, LLMs/agents,
+  RL, simulation, graphics, robotics, genomics, edge); **authentic NVIDIA-platform
+  use** that drives compute demand; a **physical/agentic AI** lean; a **defensible
+  moat** (proprietary data/model/hardware/integration); a **real, scalable market**;
+  and a **research-to-product** trajectory with credible technical depth.
+- **Strategic pull (2024–2026):** Physical AI & robotics, Agentic AI, Digital twins
+  & simulation, Sovereign/enterprise AI, Healthcare/biotech, Manufacturing/
+  industrial, Climate/energy, Cybersecurity, Scientific computing.
+- **Hard disqualifiers — never generate products that read as these:** AI wrappers /
+  thin API resellers with no GPU-essential core or moat; consulting/outsourced dev;
+  crypto (mining/trading); pure cloud-reselling or reseller/distributor plays; or
+  anything where the GPU is a "nice-to-have," not a must-have.
+
+## Step 1 — Read the idea and run the startup analysis
+
+Discover and read the startup files: `idea*`, `README*`, `pitch*`, `business-plan*`,
+`vision*`, `notes*`, `one-pager*`, `concept*`, `startup*`, `profile*`, `docs/`, etc.
+Capture the **startup name exactly**. Then analyze and restate, in your own words:
+
+- Industry · Category · Target market · Business model · Technology stack ·
+  Customer segments · Current product offerings · Market opportunities · Expansion
+  opportunities · AI opportunities · NVIDIA ecosystem opportunities.
+
+**Lock** the startup's industry, category, market, and core vision — every invented
+product must trace back to this lock. If too little exists, ask 3–5 focused
+questions first. Mark gaps `[UNKNOWN]`; never fabricate.
+
+## NVIDIA alignment engine
+
+For every product, **attempt to include** (only where authentic; label
+`[ASPIRATIONAL]` until real):
+
+> Generative AI · Agentic AI · Enterprise AI · Computer Vision · Physical AI ·
+> Predictive Intelligence · Multimodal AI · Automation · Knowledge Systems ·
+> Real-Time AI
+
+And **boost the score when relevant** by mapping to NVIDIA tech (tied to a real
+need, never as buzzwords):
+
+> NIM · NeMo · Omniverse · Cosmos · Riva · CUDA · TensorRT · GPU acceleration ·
+> Digital Twins · Simulation · Robotics · Industrial AI
+
+## Preferred product types (pick what fits the locked space)
+
+Agentic AI platforms · Enterprise AI products · Computer vision solutions ·
+Physical AI systems · Robotics software · Autonomous systems · AI copilots ·
+Industry-specific foundation models · Digital twin platforms · Synthetic data
+platforms · AI analytics systems · AI infrastructure products · Knowledge
+intelligence systems · Workflow automation platforms · Predictive intelligence
+systems.
+
+## Acceptance optimizer (the ecosystem goal)
+
+Generate products that make a reviewer believe the startup is building a **long-term
+AI ecosystem**, not a single feature. Across the set, **maximize**: AI depth ·
+innovation · scalability · enterprise value · GPU usage · technical differentiation
+· data flywheel · market opportunity · NVIDIA ecosystem compatibility. Show how the
+products reinforce each other (shared data, shared models, shared platform).
 
 ## The dual 10/10 bar (mandatory gate)
 
-Every shipped product MUST score a self-assessed **10/10 on BOTH** axes — this is
-a hard gate, not an average:
-
-- **Inception Approval Fit (target 10/10):** maximally hits the six approval
-  qualities above, trips **zero** disqualifiers, and would plausibly land at the
-  top of the `nvidia-inception-starter` acceptance score.
-- **Same-Space Fit (target 10/10):** stays entirely inside the startup's locked
-  industry, category, core idea, and target users (see Step 2) — a variation,
-  component, tier, or adjacent feature of the one idea, never a new business.
-
-Rules:
-- If a candidate scores **< 10 on either axis, redesign it until both hit 10, or
-  discard it.** Do not ship 8s or "good enough" products. Quality over quantity —
-  it is better to ship 3 dual-10/10 products than 6 mediocre ones.
-- Record both scores (with a one-line justification each) in every `product.md`
-  and in `_overview.md`. Never inflate a score to be nice; be a tough reviewer.
-
-## Step 1 — Read the existing idea
-
-Discover and read the startup files in the project: `idea*`, `README*`, `pitch*`,
-`business-plan*`, `vision*`, `notes*`, `one-pager*`, `concept*`, `startup*`,
-`docs/`, etc. Restate the problem, users, market, and any tech in your own words.
-If too little exists, ask 3–5 focused questions first. Mark gaps `[UNKNOWN]` —
-never fabricate. Stay anchored to the founder's domain and intent.
-
-## Step 2 — Invent 3–6 products (stay in the SAME space)
-
-**Same space means: every product is ONLY in the startup's same category /
-industry and directly tied to its main idea — never random categories,
-industries, or unrelated products.**
-
-**Hard constraint — same space only.** Before inventing, lock the startup's
-**industry, category, and core idea** from Step 1. Every product MUST stay inside
-that exact space:
-- Same **industry/vertical**, same **category**, same **core idea/problem space**,
-  same **target users**.
-- Do **not** invent new categories, jump to a different industry, or generate
-  random/unrelated products. No "while we're at it" pivots.
-- The 3–6 products are **variations, components, tiers, or adjacent features of
-  the one idea** — different angles on the same thing, not different businesses.
-- If a concept only works by leaving the startup's space, drop it.
-- State the locked space at the top of `_overview.md` and make every product
-  visibly trace back to it.
-
-Generate **between 3 and 6** product concepts that each:
-- Solve a sharp, real problem **within the startup's locked domain** (same idea,
-  same industry, same category).
-- **Require GPU/accelerated compute** to exist (a must-have, not nice-to-have).
-- Lean into **physical / edge / agentic AI** where credible.
-- Have a defensible moat (proprietary data/model/hardware or NVIDIA integration).
+Every shipped product MUST self-score **10/10 on BOTH** axes — a hard gate, not an
+average:
+
+- **Inception Approval Fit (10/10):** maximally hits the approval qualities above,
+  trips **zero** disqualifiers, would land at the top of the
+  `nvidia-inception-starter` acceptance score.
+- **Same-Space / Ecosystem Fit (10/10):** stays entirely inside the locked industry,
+  category, market, and vision, and strengthens the ecosystem.
+
+If a candidate scores **< 10 on either axis, redesign it until both hit 10, or
+discard it.** Better to ship 4 dual-10/10 products than 6 mediocre ones. Record both
+scores (with a one-line justification each) in every `product.md` and `_overview.md`.
+
+## Step 2 — Invent 4–6 ecosystem products (SAME space only)
+
+Generate **between 4 and 6** products that each:
+- Solve a sharp, real problem **inside the startup's locked domain** and clearly
+  complement the other products (one coherent ecosystem).
+- **Require GPU/accelerated compute** to exist (must-have, not nice-to-have).
+- Lean into **physical / edge / agentic / enterprise AI** where credible.
+- Carry a defensible moat (proprietary data/model/hardware or NVIDIA integration).
 
 ### Product naming rules (enforce strictly)
 - **4–10 letters**, letters only (A–Z), no spaces, digits, or symbols.
-- Brandable and coined/short (e.g., `Vexel`, `Roboq`, `Synapt`, `Orbix`).
-- Count the letters and confirm each name is within 4–10 before using it.
-- The product folder name = the product name.
+- Brandable and coined (e.g., `Vexel`, `Roboq`, `Synapt`, `Orbix`). Count the
+  letters and confirm 4–10 before using. The product folder name = the product name.
 
 ## NVIDIA stack for each product (order: physical → SDK → software)
 
-List concrete, fitting choices — **lead with physical/compute hardware (GPUs)**:
-
-1. **Physical / compute hardware (GPUs first):** data-center GPUs (H100, H200,
-   A100, GB200/Blackwell), workstation GPUs (RTX), **Jetson** (Orin/Nano) for
-   edge/robotics, **DGX** systems, **Grace Hopper / Grace Blackwell**, IGX/Holoscan
-   dev kits, robotics platforms on Jetson.
-2. **SDKs / libraries:** CUDA, cuDNN, **TensorRT**, **Triton** Inference Server,
-   RAPIDS, **Isaac** (Sim / ROS / Lab) for robotics, **Omniverse** for
-   simulation/digital twins, **Metropolis** for vision, **Holoscan** for sensor/edge,
-   **NeMo** for LLMs, **Riva** for speech, DeepStream, cuOpt, Modulus, Parabricks
-   (genomics), Morpheus (cybersecurity), Maxine.
-3. **Software / platforms:** **NVIDIA AI Enterprise**, **NIM** inference
-   microservices, NeMo framework/microservices, Omniverse Cloud, Fleet Command,
-   Base Command.
+Lead with **physical/compute hardware (GPUs first)**:
+1. **Hardware:** data-center GPUs (H100/H200/A100, GB200/Blackwell), RTX
+   workstations, **Jetson** (Orin/Nano) for edge/robotics, **DGX**, **Grace
+   Hopper/Blackwell**, IGX/Holoscan dev kits.
+2. **SDKs/libraries:** CUDA, cuDNN, **TensorRT**, **Triton**, RAPIDS, **Isaac**
+   (Sim/ROS/Lab), **Omniverse**, **Cosmos**, **Metropolis**, **Holoscan**, **NeMo**,
+   **Riva**, DeepStream, cuOpt, Modulus, Parabricks, Morpheus, Maxine.
+3. **Software/platforms:** **NVIDIA AI Enterprise**, **NIM**, NeMo microservices,
+   Omniverse Cloud, Fleet Command, Base Command.
 
 Pick only what authentically fits each product, and say *why* each piece is used.
 
 ## Output — `products_raw/` with one folder per product
 
 1. Create `products_raw/` in the project root if missing.
-2. For each product, create `products_raw/<ProductName>/` and write
-   `products_raw/<ProductName>/product.md`:
+2. For each product, create `products_raw/<ProductName>/` and write `product.md`
+   using **all** the sections below (the required per-product output):
 
 ```
 # <ProductName>
 Letters: <n> (must be 4–10)
-Tagline: <one line, what it is>
 Generated: <YYYY-MM-DD>
 
-## Same-space link
-<one line: how this product is the SAME industry/category/core idea as the startup>
-
-## Problem
-## Target user
-## What it does (product, not marketing)
-## Why GPU / accelerated compute is essential
-
-## NVIDIA Stack
-### Physical / compute hardware (GPUs first)
-- <GPU / Jetson / DGX / …> — <why>
-### SDKs / libraries
-- <CUDA / TensorRT / Isaac / Omniverse / …> — <why>
-### Software / platforms
-- <NVIDIA AI Enterprise / NIM / …> — <why>
-
-## Physical AI / Edge angle        (if applicable)
-## Defensibility / moat
-## Market & wedge
-## Inception fit                    (1–2 lines: why this scores well)
+## Tagline
+## Product Vision
+## Product Description
+## Customer Problem
+## Target Market
+## AI Opportunity            (which engine capabilities; real vs [ASPIRATIONAL])
+## NVIDIA Alignment          (which NVIDIA tech, tied to a real need)
+## Core Features
+## Technical Architecture    (components, data flow, models, where NVIDIA plugs in)
+## Technology Stack
+### NVIDIA hardware (GPUs first)   - <H100 / Jetson / DGX / …> — <why>
+### NVIDIA SDKs / libraries        - <CUDA / TensorRT / Isaac / …> — <why>
+### NVIDIA software / platforms    - <AI Enterprise / NIM / …> — <why>
+## Revenue Model
+## Market Size Potential     (beachhead → expansion; TAM where credible)
+## Competitive Advantage
+## Defensibility             (moat / data flywheel)
+## GPU Utilization Strategy  (exact workloads needing acceleration; why CPU-only fails)
+## NVIDIA Inception Acceptance Benefits  (why this raises acceptance probability)
+## Same-Space / Ecosystem Link  (how it stays in the locked space + complements siblings)
 ## Scorecard (must be 10/10 on both)
-- Inception Approval Fit: 10/10 — <one-line justification vs. the six approval qualities>
-- Same-Space Fit: 10/10 — <one-line justification it stays in the locked space>
+- Inception Approval Fit: 10/10 — <one-line justification>
+- Same-Space / Ecosystem Fit: 10/10 — <one-line justification>
+## Logo Concept              (type: Logomark | Wordmark | Combination Mark; the idea)
+## Brand Identity Guide      (see logo-brief.md; summary of palette + type + usage)
 ## Build notes / MVP scope
-## Logo (mark only)              (see logo-brief.md + dark & light 1080×1080 PNG variants)
 ## Open questions / [UNKNOWN]
 ```
 
-3. Also write `products_raw/_overview.md` starting with the **locked space**
-   (industry, category, core idea, target users), then list every product: name,
-   letter count, one-liner, primary NVIDIA hardware, its same-space link, and both
-   scores (Inception Approval Fit 10/10 + Same-Space Fit 10/10). Confirm every
-   listed product is a dual 10/10 — drop any that isn't.
-
-## Logo for each product (required)
-
-## Logo MARK for each product (required)
-
-Every product gets a **real, designed brand logo MARK** — a **custom abstract
-symbol only**, not the full logo. Do **not** include any letters, initials,
-monogram, or lettermark; do **not** include the full product name as a wordmark,
-no tagline, no lockup. Just the **symbol mark**, **center-aligned** on the
-canvas. It must be a proper designed brand mark — not an SVG icon, UI/material
-icon, emoji, or clip-art glyph.
-
-### Design (research first)
-- Briefly research 3–5 reference brand marks in the product's space; note the
-  visual conventions, then design something distinct (don't copy).
-- The mark must be a **distinctive custom brand symbol** tied to the concept —
-  purely abstract/pictorial, containing **no letters or initials**. Ownable,
-  memorable, simple, and scalable.
-- Define a tiny brand spec: palette (hex) and the mark concept (the symbol's
-  meaning and geometry). No typography — the mark contains no text.
-
-### Hard output rules
-- Format **PNG**, exactly **1080 × 1080 px**, **two variants required** (same mark
-  geometry, inverted color scheme):
-  1. **Dark variant** — a **dark solid background with a LIGHT mark**
-     (e.g. mark in white / very light brand tint on a near-black or deep brand-dark bg).
-  2. **Light variant** — a **light solid background with a DARK mark**
-     (e.g. mark in near-black / deep brand-dark on a white or very light brand-tint bg).
-- **Strong contrast is mandatory.** Background and mark must be clearly opposite in
-  luminance — aim for a contrast ratio of **≥ 4.5:1** (the higher the better). Never
-  ship low-contrast combinations (light-on-light, dark-on-dark, or a mark that
-  blends into the bg). The mark must read instantly at a glance.
-- **Solid background** — a single flat color with NO transparency and NO gradient
-  in each variant. State the exact bg hex AND mark hex (with the contrast pairing)
-  in the brief.
-- **Mark only**, **center-aligned** both horizontally and vertically, sized large
-  with balanced safe padding — not a tiny icon on a big canvas, and not the full
-  name spelled out. The mark must contain **no letters or initials**. The mark is
-  **identical** in both variants — only the bg/mark colors flip.
-- Save both **inside that product's folder**:
-  `products_raw/<ProductName>/<ProductName>-logo-dark-1080x1080.png`  (dark bg, light mark)
-  `products_raw/<ProductName>/<ProductName>-logo-light-1080x1080.png` (light bg, dark mark)
-- Also write `products_raw/<ProductName>/logo-brief.md` with: symbol concept, the
-  **dark-variant** bg+mark hex, the **light-variant** bg+mark hex, the contrast
-  ratio of each pairing, and references consulted. (No monogram/lettermark option
-  and no typography — the mark is a letter-free symbol.)
-
-### How to generate it
-1. **Preferred:** if an image-generation tool/model is available, prompt it for a
-   *professional brand logo mark / symbol* (specify: abstract symbol only — no
-   letters, no initials, no text, no wordmark — the concept, color palette, solid
-   background, flat vector-style, centered, no photographic texture). Generate the
-   **dark variant** (light mark on dark bg) and the **light variant** (dark mark on
-   light bg), each a 1080×1080 PNG on a solid background with the mark centered and
-   high contrast.
-2. **Fallback (always works):** render both variants programmatically — e.g.
-   Python + Pillow — drawing the **same** custom geometric/abstract symbol (no
-   letters or initials) centered on a solid-color 1080×1080 canvas, once as a light
-   mark on a dark bg and once as a dark mark on a light bg. This must look designed
-   (composition, weight, balance), not like a stock icon. Verify dimensions, that
-   the mark is centered, that each background is fully solid, and that both pairings
-   are high-contrast (light↔dark).
-
-Do **not** ship: lettermarks/monograms/initials, full wordmarks/the spelled-out
-name, taglines, transparent PNGs, gradients, **low-contrast color schemes**
-(light-on-light or dark-on-dark), generic icon-font glyphs, stock emoji, or
-screenshots. Ship **one letter-free brand symbol per product in two high-contrast
-variants**: dark bg + light mark, and light bg + dark mark.
+3. Also write `products_raw/_overview.md` opening with the **locked space**
+   (startup name, industry, category, market, vision), the **ecosystem thesis**
+   (how the 4–6 products reinforce each other), then a row per product: name, letter
+   count, tagline, primary NVIDIA hardware, ecosystem link, and both 10/10 scores.
+   Drop any product that isn't a dual 10/10.
+
+## Brand kit for each product (REQUIRED)
+
+Every product ships a **complete, enterprise-grade brand kit** — designed to look
+**venture-backed and NVIDIA-ecosystem ready**, suitable for accelerators and
+investor presentations.
+
+### Logo type (pick the best fit per product)
+- **Logomark** (symbol only), **Wordmark** (the short brandable name as type), or
+  **Combination Mark** (symbol + wordmark). Choose what suits the product and state
+  the choice in the brief.
+
+### Logo rules (all must hold)
+Enterprise-grade · highly memorable · scalable to tiny sizes · works on **dark**
+backgrounds · works on **light** backgrounds · strong contrast · **avoid generic AI
+symbols** (no clichéd brains, neurons, circuit boards, or hexagon "AI" glyphs) ·
+looks venture-backed · NVIDIA-ecosystem ready · accelerator- and investor-ready.
+
+### Deliverable image files (save in the product's folder)
+Generate, for **every** product:
+- **Primary logo** — `<ProductName>-logo-primary-1080x1080.png` (on brand background)
+- **Square logo** — `<ProductName>-logo-square-1080x1080.png`
+- **App icon** — `<ProductName>-app-icon-1024x1024.png`
+- **Favicon** — `<ProductName>-favicon-512x512.png` (also note a 32×32 use)
+- **SVG version** — `<ProductName>-logo.svg` (clean vector)
+- **Transparent-background version** — `<ProductName>-logo-transparent-1080x1080.png`
+  (PNG with real alpha; this is the asset that proves it works on dark AND light)
+
+All raster logos must be high-contrast and legible; the transparent + SVG versions
+must read cleanly on both a dark and a light surface.
+
+### Deliverable docs — write `products_raw/<ProductName>/logo-brief.md` containing:
+- **Detailed logo design brief** — concept, meaning, geometry, logo type chosen,
+  why it fits the product and avoids generic AI clichés.
+- **AI image generation prompt** — a ready-to-paste prompt for an image model
+  (specify: enterprise logo, the concept, logo type, flat vector style, solid
+  background, centered, no photographic texture, no generic AI clichés).
+- **SVG generation prompt** — instructions/prompt to produce the clean vector SVG.
+- **Brand usage guide** — clear space, min size, do/don't, dark vs light usage.
+- **Typography recommendation** — primary + secondary typefaces (and fallbacks).
+- **Color palette recommendation** — exact hex values (primary, accent, neutrals,
+  dark-bg and light-bg variants) with contrast notes.
+
+### How to generate the images
+1. **Preferred:** if an image-generation tool/model is available, prompt it with the
+   AI prompt above to render the primary, square, app-icon, and transparent PNGs;
+   then derive the favicon and hand-author the SVG.
+2. **Fallback (always works):** render programmatically — e.g. Python + Pillow for
+   the PNGs (and resize for app icon/favicon) plus a hand-written `.svg` — drawing
+   the chosen mark/wordmark with real design intent (composition, weight, balance),
+   not a stock icon. Verify every file's dimensions, that the transparent PNG truly
+   has alpha, and that the design holds on both dark and light backgrounds.
+
+Do **not** ship: generic AI-cliché symbols, low-contrast designs, screenshots,
+stock emoji, or UI-font glyphs. Ship the **full kit** (primary, square, app icon,
+favicon, SVG, transparent) + `logo-brief.md` for every product.
 
 ## Quality bar
 
-- **Dual 10/10 gate.** Every shipped product scores 10/10 on Inception Approval
-  Fit AND 10/10 on Same-Space Fit. Redesign or discard anything below 10 on
-  either axis — quality over quantity.
-- **Same space only.** All products share the startup's industry, category, core
-  idea, and users — variations/components/tiers of the one idea, never new
-  categories or unrelated businesses.
-- **Products, not copy.** No website/landing/marketing text — concrete product
-  concepts and technical plans only.
-- Every product must be GPU-essential and NVIDIA-aligned, with the stack ordered
-  physical → SDK → software, and trip **zero** Inception disqualifiers (no AI
-  wrappers, consulting, crypto, or reseller plays).
-- Names obey the 4–10 letter rule (verified) and folder names match.
+- **Dual 10/10 gate.** Every product scores 10/10 on Inception Approval Fit AND
+  10/10 on Same-Space/Ecosystem Fit; redesign or discard anything lower.
+- **Same space only.** All products share the startup's industry, category, market,
+  and vision, and strengthen one ecosystem — never new categories or unrelated
+  businesses.
+- **Products, not copy.** Concrete product concepts, technical plans, and brand
+  assets only — no website/landing/marketing text.
+- **GPU-essential + NVIDIA-aligned**, stack ordered physical → SDK → software,
+  tripping **zero** disqualifiers.
+- Names obey the 4–10-letter rule (verified); folder names match.
+- Every product ships the **complete brand kit** + `logo-brief.md`.
 - Stay within the founder's domain; elevate, don't fabricate.
 
 ## Verification checklist (run before finishing)
 
-1. Read the existing idea files; locked the startup's industry, category, core idea.
-2. **All 3–6 products stay in that same space** — no new categories or unrelated
-   ideas — and each `product.md` has a "Same-space link" tracing back to it.
-3. `products_raw/` exists with **3–6** product subfolders, each with `product.md`.
-4. Every product name is **4–10 letters, letters only**, and matches its folder.
-5. Each product lists NVIDIA hardware (GPU/physical) first, then SDKs, then software.
-6. Each product justifies why GPU/accelerated compute is essential.
-7. **Every product is a dual 10/10** — Inception Approval Fit 10/10 and Same-Space
-   Fit 10/10, both scored with a justification in `product.md` and `_overview.md`,
-   and trips none of the hard disqualifiers (AI wrapper, consulting, crypto, reseller).
-8. `products_raw/_overview.md` opens with the locked space, then summarizes all
-   products with both scores.
-9. Every product folder has **two 1080×1080 PNG logo MARK variants on solid
-   (non-transparent, non-gradient) backgrounds** — a **dark bg + light mark** and a
-   **light bg + dark mark**, both high-contrast (≥ 4.5:1), the **mark centered and
-   identical** across variants, with no full wordmark/name text and **no letters,
-   initials, or monogram**, plus `logo-brief.md` recording both color pairings — a
-   real letter-free custom symbol, not an SVG/UI icon or emoji. Verify dimensions,
-   centering, and that each pairing is genuinely light↔dark.
-10. No marketing/website copy; unknowns marked `[UNKNOWN]`, not invented.
+1. Read the idea files; ran the startup analysis; locked industry, category, market, vision.
+2. **4–6 products**, all in the locked space, forming one coherent ecosystem; each
+   `product.md` has a "Same-Space / Ecosystem Link".
+3. `products_raw/` exists with **4–6** subfolders, each with `product.md` containing
+   **all required sections** (tagline → acceptance benefits → logo concept → brand guide).
+4. Every product name is **4–10 letters, letters only**, matching its folder.
+5. Each product lists NVIDIA hardware (GPU) first, then SDKs, then software, and
+   justifies why GPU/accelerated compute is essential.
+6. Each product includes a Revenue Model, Market Size Potential, Competitive
+   Advantage, Defensibility, GPU Utilization Strategy, and Acceptance Benefits.
+7. **Every product is a dual 10/10** (both scores + justifications in `product.md`
+   and `_overview.md`) and trips none of the hard disqualifiers.
+8. `products_raw/_overview.md` opens with the locked space + ecosystem thesis, then
+   summarizes all products with both scores.
+9. Every product folder has the **full brand kit**: primary 1080×1080, square
+   1080×1080, app icon 1024×1024, favicon, SVG, and a **transparent-background**
+   version, plus `logo-brief.md` with the brief, AI prompt, SVG prompt, brand usage
+   guide, typography, and color palette. No generic AI-cliché or low-contrast logos.
+10. No marketing/website copy; unknowns marked `[UNKNOWN]`, aspirational items labeled.