docs: tighten model-card prose (facts unchanged)

README.md

@@ -16,22 +16,21 @@ tags:
 
 # Tanpo Fundraising
 
-A compact founder fundraising specialist (~1.2B) for investor pitches, narrative, ask/use-of-funds, diligence Q&A, updates, and round-structure language (non-legal) — built for local and inexpensive deployment.
+A compact founder fundraising specialist (~1.2B) for investor pitches, narrative, ask/use-of-funds, diligence Q&A, updates, and round-structure language (non-legal) — meant for local and inexpensive deployment.
 
-**Creator:** [d4rkninja](https://huggingface.co/d4rkninja)  
 **Collection:** [Tanpo — Domain Specialists](https://huggingface.co/collections/d4rkninja/tanpo-domain-specialists-6aaccdb3985768f6dad8449b)
 
-**Original upstream:** [LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct) (~1.17B parameters, 32,768-token context, designed for edge/on-device deployment).
+Built on [LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct) (~1.17B parameters, 32,768-token context). Fine-tuned with Unsloth / PEFT LoRA via hub id [`unsloth/LFM2.5-1.2B-Instruct`](https://huggingface.co/unsloth/LFM2.5-1.2B-Instruct).
 
-**Fine-tuning:** Unsloth-compatible loading of that checkpoint via hub id [`unsloth/LFM2.5-1.2B-Instruct`](https://huggingface.co/unsloth/LFM2.5-1.2B-Instruct) (LoRA / PEFT).
-
-This repository hosts the **merged** Transformers weights (LoRA merged into the base).
+This repository hosts the **merged** Transformers weights (LoRA folded into the base).
 
 ## Overview
 
-Tanpo is a family of compact domain-specialized business models for local / edge / inexpensive deployment. Different specialists cover different workflows. Tanpo Fundraising is one specialist in that family (not a frontier or general-purpose model).
+Tanpo is a family of small domain-specialized business models for local, edge, and inexpensive deployment — one compact architecture, multiple focused specialists.
+
+**Tanpo Fundraising** is one specialist in that family — focused on its domain, not a general-purpose or frontier model.
 
-Related artifacts:
+Related formats:
 
 - LoRA adapter: [d4rkninja/tanpo-fundraising-LoRA](https://huggingface.co/d4rkninja/tanpo-fundraising-LoRA)
 - GGUF quants: [d4rkninja/tanpo-fundraising-GGUF](https://huggingface.co/d4rkninja/tanpo-fundraising-GGUF) — prefer `Q4_K_M` when available
@@ -50,13 +49,13 @@ Related artifacts:
 - Unauthorized access, credential attacks, or social-engineering harm
 - General coding or non-fundraising chat
 
-## Why a Specialist Model?
+## Why a Specialist?
 
-Fundraising work rewards repeatable structure (narrative, ask, diligence, updates). Specializing a small model for those workflows enables private, low-cost inference without a large general model.
+Fundraising work rewards repeatable structure — narrative, ask, diligence, updates. Specializing a small model for those workflows enables private, low-cost inference without a large general model.
 
 ## Evaluation
 
-Internal automated domain evaluation (DarkLab harness). Treat as **directional**, not an industry benchmark.
+Internal automated evaluations. Treat scores as **directional**, not industry benchmarks.
 
 | Model | Rubric overall |
 |---|---:|
@@ -64,17 +63,13 @@ Internal automated domain evaluation (DarkLab harness). Treat as **directional**
 | **tanpo-fundraising** | **95.2%** |
 | Delta | **+11.1 percentage points** |
 
-**BEATS_BASE:** YES.
-
-**Honest trail:** FT trails the base on **only** `dilution_round_structure` (**−22.2** percentage points on that category). Overall still wins; published on the overall win with this category called out.
+On the overall score the fine-tune leads; it still **trails the base** on the `dilution_round_structure` category (**−22.2** percentage points on that category alone). Treat round-structure / dilution drafts with extra care.
 
-**Artifacts:** [`evaluation/`](evaluation/) — [`COMPARE_FUNDRAISING.md`](evaluation/COMPARE_FUNDRAISING.md), [`fundraising_tasks.jsonl`](evaluation/fundraising_tasks.jsonl), [`score_rubric.md`](evaluation/score_rubric.md), [`evaluation/README.md`](evaluation/README.md).
+**Artifacts:** [`evaluation/`](evaluation/) — [`COMPARE_FUNDRAISING.md`](evaluation/COMPARE_FUNDRAISING.md), [`COMPARE.md`](evaluation/COMPARE.md), [`fundraising_tasks.jsonl`](evaluation/fundraising_tasks.jsonl), [`score_rubric.md`](evaluation/score_rubric.md), [`scorecard.json`](evaluation/scorecard.json), [`evaluation/README.md`](evaluation/README.md).
 
-**Methodology:** DarkLab automated domain evaluation (~20 held-out fundraising tasks). Same prompts and generation config for base vs fine-tune. Not an industry benchmark. Headline numbers are the parent-verified DarkLab pair from VrindaVan.
+**Methodology:** DarkLab automated domain evaluation (~20 held-out fundraising tasks). Same prompts and generation config for base vs fine-tune.
 
-**Missing:** Full `results_fundraising` / `results_fundraising_vs_base` scorecard JSON, per-task prediction jsonl, and charts were not present on the publish box (VrindaVan outputs not copied). Parent can upload from Windows.
-
-**Limitations of this eval:** Automated rubrics can reward structure over real-world quality; sample size is small; results may not transfer outside the task distribution.
+**About these numbers:** Automated rubrics can reward structure over real-world quality; sample sizes are small; results may not transfer outside each task distribution.
 
 ## Example Prompts
 
@@ -97,8 +92,8 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 
 messages = [
-    {"role": "system", "content": "You are Tanpo Fundraising, a practical founder fundraising assistant."},
-    {"role": "user", "content": "Rewrite the pitch opener for a B2B SaaS fleet-ops tool moving from seed to Series A bridge. Current story is feature soup — give a sharp narrative arc and talk track."},
+    {"role": "system", "content": 'You are Tanpo Fundraising, a practical founder fundraising assistant.'},
+    {"role": "user", "content": 'Rewrite the pitch opener for a B2B SaaS fleet-ops tool moving from seed to Series A bridge. Current story is feature soup — give a sharp narrative arc and talk track.'},
 ]
 inputs = tokenizer.apply_chat_template(
     messages, add_generation_prompt=True, return_tensors="pt"
@@ -113,7 +108,7 @@ Verified from published adapter configs / training artifacts (no unverified hype
 
 | Field | Value |
 |---|---|
-| Method | LoRA (PEFT) via **Unsloth** FastLanguageModel on hub id `unsloth/LFM2.5-1.2B-Instruct` (Unsloth-compatible of LiquidAI/LFM2.5-1.2B-Instruct) |
+| Method | LoRA (PEFT) via **Unsloth** FastLanguageModel on hub id `unsloth/LFM2.5-1.2B-Instruct` (Unsloth-compatible load of LiquidAI/LFM2.5-1.2B-Instruct) |
 | LoRA rank (`r`) | 16 |
 | LoRA alpha | 16 |
 | LoRA dropout | 0 |
@@ -140,11 +135,13 @@ Not legal, tax, or **investment advice**. Humans must review pitches, financial
 
 ## License
 
-`license: other` — this repository inherits obligations from the upstream Liquid AI **LFM2.5** / LFM family base model ([LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct); training hub id [unsloth/LFM2.5-1.2B-Instruct](https://huggingface.co/unsloth/LFM2.5-1.2B-Instruct)). Review Liquid AI’s license terms before commercial redistribution or redistribution of derivatives. Do not assume Apache-2.0 covers the merged weights.
+`license: other` — inherits obligations from the upstream Liquid AI **LFM2.5** / LFM base ([LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct); training hub id [unsloth/LFM2.5-1.2B-Instruct](https://huggingface.co/unsloth/LFM2.5-1.2B-Instruct)). Review Liquid AI’s license before commercial redistribution or redistribution of derivatives. Do not assume Apache-2.0 covers the merged weights.
 
 ## Tanpo Model Family
 
-Tanpo is a family of compact domain-specialized business models for **local / edge / inexpensive** deployment. Different specialists cover different workflows. One compact architecture ([LiquidAI/LFM2.5-1.2B-Instruct](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct)) → multiple focused specialists → each ships **Full/Merged | LoRA | GGUF**.
+Tanpo is a family of small domain-specialized business models for local, edge, and inexpensive deployment — one compact architecture, multiple focused specialists. Built for focused work, not every task.
+
+Browse the set: [Tanpo — Domain Specialists](https://huggingface.co/collections/d4rkninja/tanpo-domain-specialists-6aaccdb3985768f6dad8449b)
 
 | Model | Specialization |
 | --- | --- |
@@ -163,10 +160,10 @@ Tanpo is a family of compact domain-specialized business models for **local / ed
 | Fundraising | [tanpo-fundraising](https://huggingface.co/d4rkninja/tanpo-fundraising) | [tanpo-fundraising-LoRA](https://huggingface.co/d4rkninja/tanpo-fundraising-LoRA) | [tanpo-fundraising-GGUF](https://huggingface.co/d4rkninja/tanpo-fundraising-GGUF) |
 
 ```
-Original upstream: LiquidAI/LFM2.5-1.2B-Instruct (~1.17B, 32K context)
-  └── Unsloth-compatible hub load: unsloth/LFM2.5-1.2B-Instruct
-        └── Tanpo Specialist (domain SFT via LoRA)
-              ├── Full / Merged  → d4rkninja/tanpo-fundraising
-              ├── LoRA adapter   → d4rkninja/tanpo-fundraising-LoRA
-              └── GGUF quants    → d4rkninja/tanpo-fundraising-GGUF
+LiquidAI/LFM2.5-1.2B-Instruct (~1.17B, 32K)
+  └── Unsloth load: unsloth/LFM2.5-1.2B-Instruct
+        └── Tanpo Fundraising (domain SFT via LoRA)
+              ├── Merged  → d4rkninja/tanpo-fundraising
+              ├── LoRA    → d4rkninja/tanpo-fundraising-LoRA
+              └── GGUF    → d4rkninja/tanpo-fundraising-GGUF
 ```