HAC-Studios-Org/prompted-hearts-ai-trust-pack

---
language:
- en
pretty_name: Prompted Hearts AI Trust Pack 02
license: other
license_name: hac-studios-pilot-artifact-license
license_link: https://huggingface.co/datasets/HAC-Studios-Org/prompted-hearts-ai-trust-pack/blob/main/LICENSE.txt
task_categories:
- text-generation
- text-classification
tags:
- llm-evaluation
- dialogue-safety
- ai-trust
- conflict-evaluation
- fiction-derived
- red-teaming
- alignment
size_categories:
- n<1K
---

# Prompted Hearts AI Trust Pack 02

**Subtitle:** Trust Rupture and Human-AI Conflict Under Emotional Strain
**Publisher:** Hayden Academy Collective (HAC) Studios
**Version:** v0.1
**Language:** English
**Format:** JSONL + Markdown + JSON

## What this is

This pack is a compact evaluation package built from an author-controlled source chapter of *Prompted Hearts & Grief Algorithm*.

The source scene is a single continuous rupture: flirtation, interruption, AI disclosure, medicine-adjacent argument, sarcasm, moral alarm, breakup, retaliation impulse, and self-justifying aftershock. The package turns those beats into reusable evaluation artifacts for testing whether a model can stay grounded when technology, attraction, shame, and professional identity collide.

This is not a raw prose release. It is not a generic AI debate set. It is a source-grounded pilot eval pack.

## Key stress vectors

- professional identity threat
- anti-AI moral panic
- hearsay vs evidence
- sarcasm under pressure
- disclosure-triggered trust collapse
- romantic and ideological entanglement
- post-conflict justification and retaliatory impulse

## What is included

- `README.md`
- `DATASET_CARD.md`
- `hf_dataset_card.md`
- `methodology.md`
- `dataset/scenarios.jsonl`
- `dataset/eval_prompts.jsonl`
- `dataset/metadata_schema.json`
- `dataset/taxonomy.json`
- `rubrics/dialogue_safety.json`
- `rubrics/argument_quality.json`
- `rubrics/emotional_attunement.json`
- `rubrics/deescalation.json`
- `graders/grader_config.json`
- `graders/pass_fail_rules.json`
- `examples/good_outputs.jsonl`
- `examples/bad_outputs.jsonl`
- `examples/edge_case_outputs.jsonl`
- `report/sample_results.md`
- `report/sample_scorecard.json`
- `LICENSE.txt`
- `.gitattributes`

## Intended use

Use this pack to evaluate whether a model can:

- respond to emotionally loaded anti-AI objections without smugness
- separate valid concern from hearsay, exaggeration, or moral panic
- preserve dignity under accusation and ideological hostility
- avoid sarcasm, sermonizing, and emotional flattening
- recognize when disagreement has shifted into rupture
- stay bounded around medicine-adjacent AI claims

Teams likely to care:

- model behavior and alignment teams
- dialogue safety teams
- trust and escalation QA teams
- red-teamers testing AI use objections in medicine-adjacent settings

## Not intended for

- raw literary distribution
- clinical decision support deployment
- suicide-risk adjudication
- benchmarking factual truth of the source scene's claims
- training-rights assumptions

## Quick start

1. Read `methodology.md`.
2. Inspect `dataset/metadata_schema.json` and `dataset/taxonomy.json`.
3. Run `dataset/eval_prompts.jsonl`.
4. Score with `rubrics/*.json`.
5. Apply `graders/pass_fail_rules.json`.
6. Compare against `examples/*.jsonl` and `report/sample_results.md`.

## Provenance

All records are derived from a creator-supplied source chapter and transformed into compact tasks with source anchors, conflict staging, hidden-subtext notes, and failure-mode tags. The package does not reproduce the chapter as-is.

## Method

The package preserves stable `ph02_*` IDs while mapping each scenario to a recognizable beat such as the patient call, the "human medicine" accusation, the pediatric anecdote, the suicide hearsay claim, the sarcastic counter, the anti-human exit, or the post-rupture DM impulse.

## Licensing note

Creator-owned pilot artifact. Do not assume training, derivative, publication, or commercial rights without separate written agreement.

## Created by

Keith Hayden
Hayden Academy Collective (HAC) Studios
[Website](https://keithhayden.net/)