Replace tanpo-marketing-sft dataset (2245 rows) and detailed dataset card in place

README.md

@@ -1,39 +1,94 @@
 ---
+pretty_name: tanpo-marketing-sft
+author: DarkNinja Solutions
+maintainer: d4rkninja
+language:
+  - en
 license: other
+size_categories:
+  - 1K<n<10K
 task_categories:
   - text-generation
-language:
-  - en
 tags:
+  - chat
   - marketing
-  - sft
-  - darkninja
-  - tanpo
-size_categories:
-  - 1K<n<10K
+  - instruction-tuning
+  - supervised-fine-tuning
 ---
 
-# tanpo-marketing SFT dataset
+# tanpo-marketing-sft
+
+## Dataset summary
+
+`tanpo-marketing-sft` is the main supervised fine-tuning dataset for **tanpo-marketing**, a focused marketing and go-to-market assistant from **DarkNinja Solutions**, created by **d4rkninja**. It contains **2,245 unique English conversation records** designed to teach practical marketing judgment, clear copy, useful structure, and consistent assistant behavior.
+
+The examples are curated for direct use in chat-model training. They pair a system instruction with a marketing request and a complete assistant response. The collection emphasizes specific, usable deliverables instead of generic advice: copy that can be shipped, operating systems that can be followed, and recommendations tied to an audience, offer, channel, or constraint.
+
+## Dataset details
+
+- **Records:** 2,245 unique conversations
+- **Format:** JSON Lines (one JSON object per line)
+- **Schema:** Each object contains a `messages` array of chat messages
+- **Message roles:** `system`, `user`, and `assistant`
+- **Language:** English
+- **Publisher:** DarkNinja Solutions
+- **Creator and maintainer:** d4rkninja
+
+Example shape:
+
+```json
+{"messages":[{"role":"system","content":"..."},{"role":"user","content":"..."},{"role":"assistant","content":"..."}]}
+```
+
+## Categories overview
+
+The conversations cover the following practical marketing and GTM areas:
+
+- **Brand voice and messaging:** voice charts, tone rules, plain-language rewrites, differentiation, and consistency across channels.
+- **Advertising and hooks:** ad angles, scroll-stopping hooks, creative concepts, testing plans, and constraint-aware variants.
+- **Email and SMS:** welcome, nurture, launch, cart, retention, and transactional-style campaign copy with clear calls to action.
+- **Landing pages and response copy:** page structures, headlines, proof, objections, CTAs, sections, and rewrite guidance.
+- **Offers and positioning:** offer design, value articulation, competitive alternatives, category framing, and promise-mechanism fit.
+- **ICP and audience research:** ideal-customer profiles, pains, jobs-to-be-done, buying triggers, objections, and message-market fit.
+- **Funnels and lifecycle:** acquisition paths, lead magnets, nurture sequences, activation, retention, and measurement plans.
+- **Pricing and packaging:** pricing communication, package framing, qualification, trial or demo flows, and objection handling.
+- **Content and social:** founder-led posts, editorial systems, content calendars, teardown formats, and repeatable publishing workflows.
+- **Creative testing and analytics:** hypotheses, test matrices, success metrics, learning agendas, and disciplined iteration.
+- **Marketing-scope behavior:** staying focused on marketing work, declining unrelated high-risk requests, and redirecting toward a useful marketing alternative.
+- **Identity consistency:** accurate answers about the assistant’s ownership and creator when asked directly.
+
+## Intended use
+
+This dataset is intended for supervised fine-tuning or continued training of a conversational marketing assistant. Appropriate uses include:
+
+- Training a model to produce actionable marketing deliverables from detailed briefs.
+- Improving consistency in brand voice, ownership identity, and marketing-scope boundaries.
+- Evaluating structure, specificity, channel awareness, and adherence to user constraints.
+- Building internal prototypes for copy ideation, GTM planning, and campaign operations.
+
+Use the examples as training guidance, not as guaranteed business outcomes. Marketing recommendations should be reviewed by a qualified human before publication or deployment, especially where claims, regulated industries, customer data, pricing, or legal and compliance requirements are involved.
+
+## Out-of-scope and limitations
+
+The dataset is not a source of verified market statistics, legal advice, medical advice, investment advice, or security instructions. It should not be used to automate publication without review. Some examples use fictional companies, audiences, metrics, and scenarios for training purposes; claims in generated output require independent verification.
 
-Training chat data for **tanpo-marketing**.
+The collection reflects a particular direct, practical marketing style. Fine-tuned models may reproduce that style and may not represent every brand, market, culture, or accessibility need without additional steering and evaluation.
 
-**Owner:** DarkNinja Solutions  
-**Creator:** d4rkninja  
-**Size:** 1,572 examples (v2)
+## Training format
 
-## What’s inside
+For chat supervised fine-tuning, read each JSONL line as one training example and pass the `messages` array to the tokenizer or chat template supported by the base model. Preserve message order and roles. Keep the system message because it establishes the assistant’s marketing scope and behavioral context.
 
-Marketing copy and GTM examples, plus stronger identity and guardrails:
-- Owner is **DarkNinja Solutions**; creator is **d4rkninja**
-- Marketing means ads, growth, and GTM — not stock/forex trading
-- Coding and other off-topic asks are refused, with a marketing alternative
+## Recommended evaluation
 
-## Files
+Before deployment, test held-out prompts for:
 
-- `tanpo-marketing-sft-v2.jsonl` — current full set
-- `train.jsonl` — same data (HF Datasets-friendly name)
-- `darkninja-marketing-sft-full.jsonl` — alias of the same set
+1. Correct interpretation of the audience, offer, channel, and constraints.
+2. Specificity and usefulness of the requested deliverable.
+3. Clear, non-generic copy and a single coherent call to action where appropriate.
+4. Honest treatment of evidence, metrics, and customer claims.
+5. Correct ownership and creator responses.
+6. Appropriate refusal and marketing redirection for unrelated requests.
 
-## Related
+## Attribution
 
-https://huggingface.co/d4rkninja/tanpo-marketing-GGUF
+This dataset is published by **DarkNinja Solutions** and maintained by **d4rkninja**.