docs/COMMIT_LOG.md

Commit Log

This file records the development commits that show how Headline Booster was built and refactored with help from OpenAI Codex as a coding agent.

Why this exists

The OpenAI Codex hackathon track benefits from clear evidence that Codex participated in real repository changes. The Git history is the primary source of truth, and this file provides a human-readable summary of the relevant commits.

Branch alignment

The repository should keep main and codex pointing to the same final commit for the hackathon submission. Any older temporary work branches can be deleted after both branches are aligned.

Current commits

Commit	Summary	Notes
497c903	Primer guardado del proyecto	Original repository checkpoint before the Headline Booster migration.
b1f3f74	Migrate repo to Headline Booster Gradio Space — add Gradio app and remove legacy backend/frontend	Codex-assisted migration from the previous React/FastAPI/Supabase product to a focused Gradio Space with mock generation, Tiny Titan model runtime wiring, custom frontend CSS, Hugging Face README metadata, chatbot interaction flow, performance defaults, dark chat frontend, and documentation.
HEAD	Experimental model X-ray analysis mode	Adds ANALYSIS_MODE so the app can use deterministic rules, validated model-generated X-ray analysis, or a hybrid fallback; updates the prompt to return compact diagnosis, missing elements, versions, winner, and rationale in one JSON while preserving backend validation. Use git log --oneline -1 for the exact hash because embedding the current commit hash in this same file would change that hash.

Fix — Normalize model headline proposals

• Corregido un bug donde las propuestas de titulares podían mostrarse como diccionarios en el frontend.
• Se agregó normalización para que versiones siempre sea una lista de strings limpios.
• Se reforzó el prompt para que el modelo devuelva versiones como strings, no objetos.
• Se mantiene compatibilidad con ANALYSIS_MODE rules/model/hybrid.
• Se conserva la estructura JSON esperada por el frontend.
• Archivos afectados: app.py, docs/COMMIT_LOG.md.

Prompt — Strategic copywriting formulas per version

• Se reforzó el prompt para que cada una de las tres versiones nazca de familias de fórmulas de copywriting distintas.
• La versión clara/directa prioriza claridad, beneficio, especificidad y credibilidad.
• La versión emocional prioriza problema, deseo, alivio y contraste emocional.
• La versión curiosa/diferenciada prioriza curiosity gap, reframe, contraste y patrón de interrupción.
• Las fórmulas siguen siendo criterio estratégico interno y no deben aparecer en los titulares.
• Archivos afectados: app.py, docs/COMMIT_LOG.md.

Codex evidence checklist

• Real code changes were made in app.py.
• The repository was simplified for Hugging Face Spaces.
• External AI API dependencies were removed for the first mock version.
• Tiny Titan model integration wiring was added for Qwen/Qwen2.5-1.5B-Instruct with a deterministic mock fallback.
• Documentation was added in README.md, docs/CODEX_NOTES.md, docs/FIELD_NOTES.md, docs/FRONTEND_REFERENCE_NOTES.md, and docs/TINY_TITAN_PLAN.md.
• This commit log records the migration history for reviewers.

How to verify locally

Run:

git log --oneline --decorate --all

The output should include the commits listed above plus any later fixes or documentation updates.

To verify branch alignment after cleanup, run:

git rev-parse main
git rev-parse codex

Both commands should print the same commit hash.