Fernandosr85/causabr-impactcopy-adapter

CausaBR-ImpactCopy — Brazilian Nonprofit Copywriting Adapter

LoRA adapter for ethical copywriting for the Brazilian third sector (ONGs, OSCIPs, institutes, and social foundations), fine-tuned on Mixtral-8x7B-Instruct-v0.1 via Adaption's AutoScientist platform.

Covers 12 distinct communication formats across fundraising, ESG proposals, impact reports, campaigns, social media, grant applications, beneficiary communication, volunteer mobilization, transparency notes, advocacy, and crisis communication — all in PT-BR.

---

Evaluation results

Training Winrates

The adapted model outperforms the base model with 69% win rate vs 31% on held-out copywriting evaluation — a +122% relative improvement over the base model.

Model	Win Rate
Base (Mixtral-8x7B-Instruct-v0.1)	31%
Adapted (causabr-impactcopy)	69%

Train/Eval Metrics

Metric	Value
Initial train loss	~2.00
Final validation loss	~0.93
Loss reduction	~54%
Peak learning rate	1.0e-6
Training steps	204
LR scheduler	cosine (warmup)
Gradient norm	spike → stable

Train loss (cyan) converged steadily over 204 steps. Validation loss (black dots) tracked closely, confirming generalization without overfitting. Learning rate followed cosine schedule with warmup, peaking around step 30 then decaying to near-zero. Gradient norm stabilized after initial spike, indicating stable optimization throughout.

Dataset quality

Metric	Value
Dataset grade	A
Score before / after	8.0 → 9.4
Quality improvement	+17.5% (Adaption Adaptive Data remastering)
Total examples	1,200 instruction/completion pairs

---

Model details

Field	Value
Base model	mistralai/Mixtral-8x7B-Instruct-v0.1 (46.7B)
Trained model name	adaption_mixtral_8x7b_instruc_ongbr_impact_copy_2612056c
Training method	Supervised Fine-Tuning (SFT) + LoRA
Epochs	3
Data format	Chat (instruction/completion)
Domain	Brazilian nonprofit copywriting — PT-BR
Category	Marketing

---

Training dataset

Fernandosr85/adaption-ongbr-impact-copy

1,200 instruction/completion pairs across 12 copy types:

Type	Format	Channel	Examples
A	Fundraising appeal — individual donors	Email / landing page	100
B	Corporate partnership proposal — ESG	Proposta / deck	100
C	Impact report narrative	Relatório anual / PDF	100
D	Seasonal campaign	Email / redes sociais	100
E	Instagram post	Instagram feed	100
F	LinkedIn post	LinkedIn	100
G	Grant application	Formulário de edital	100
H	Beneficiary communication	WhatsApp / SMS	100
I	Volunteer mobilization	Email / redes sociais	100
J	Transparency note	Newsletter / site	100
K	Advocacy and incidence	Email / petição	100
L	Crisis communication	Site / assessoria	100

Quality controls

8-dimension rubric (16 pts max): clarity of purpose, credibility, proportional emotional appeal, CTA, language fit, tone of voice, beneficiary dignity, differentiation. Dataset mean: 11.74/16.

3-tier dignity gate: blocks poverty porn, fabricated impact data, and impossible promises. All 1,200 examples passed with dignity_flag: ok.

---

Dignity Principle

The core ethical constraint of this adapter: beneficiary language must center agency and resilience, never victimization.

	Example
✅ Correct	"Marcus quer ser professor."
❌ Blocked	"Marcus, abandonado, precisa da sua ajuda para sobreviver."

---

Model repositories

Platform	Link
HuggingFace Model	Fernandosr85/causabr-impactcopy-adapter
HuggingFace Dataset	Fernandosr85/adaption-ongbr-impact-copy
Kaggle Dataset	fernandosr85/causabr-impactcopy

---

Related adapters (same portfolio)

Adapter	Domain	Category	Win rate
RegTech BR	Brazilian crypto regulatory compliance	Legal	63%
AfroBR-LangBench	Afro-Brazilian Portuguese sociolinguistics	Language	—
VozBR-BrandVoice	Citizen complaint → institutional response	Marketing	79%
HealthBR-CoverageGuide	Health plan and SUS navigation	Healthcare	64%
FinRisk-BR	Crypto investor risk assessment	Finance	—
CausaBR-ImpactCopy	NGO copywriting PT-BR	Marketing	69%

---

Credits

• Fine-tuning platform: Adaption — AutoScientist & Adaptive Data
• Challenge: AutoScientist Challenge 2026
• Training infrastructure: Adaption compute credits
• Dataset remastering: Adaption Adaptive Data pipeline (Grade A, +17.5% improvement)
• Author: Fernando Rodrigues · Kaggle: fernandosr85 · HuggingFace: Fernandosr85

---

Disclaimer

Experimental research artifact submitted to the AutoScientist Challenge 2026 (Marketing category). Generated copy requires human review before publication. The dignity gate enforces ethical communication standards but does not substitute for editorial judgment by communications professionals.