# `llm_conversations_database-development/` — Folder Guide

> Paper trail for the design of the process tracing evidence database. Every schema version in `xlsx_templates/` was produced (or substantively revised) inside one of the LLM conversations in `conversations/`, informed by the methodological reading in `references/`. The narrative connecting these pieces lives in [`../HISTORY.md`](../HISTORY.md).

> Registro do desenvolvimento do banco de evidências para process tracing. Cada versão do esquema em `xlsx_templates/` foi produzida (ou substancialmente revisada) dentro de uma das conversas em `conversations/`, informada pela leitura metodológica em `references/`. A narrativa que conecta as peças está em [`../HISTORY.md`](../HISTORY.md).

---

## 1. What is in this folder

```
llm_conversations_database-development/
├── README.md                <- this file
│
├── conversations/           <- the LLM conversations that produced each version
│   ├── rename-suggestions.md   (applied 2026-05-27; preserves original-name map)
│   ├── 2026-05-02_perplexity_pt-literature-and-v1-v2-schema.md
│   ├── 2026-05-03_claude_bayesian-tests_v1-through-v4.md
│   ├── 2026-05-05_perplexity_v3-structural-pivot.md
│   └── 2026-05-24_claude_v5-fairfield-charman-2022-deepread.md
│
├── references/              <- the methodological literature consulted
│   ├── Fairfield-Charman-2017_Explicit-Bayesian-Analysis.pdf
│   ├── Fairfield-Charman-2022_Social-Inquiry-Bayesian-Inference.pdf
│   ├── Fairfield-Charman-2026_Bayesian-reasoning-replication.pdf
│   ├── Fairfield-Charman-2026_Appendix.pdf
│   ├── Fairfield-Garay-2017_Redistribution-Under-the-Right.pdf
│   └── Fairfield-Garay-2017_Appendix.pdf
│
├── xlsx_templates/          <- successive schema versions (case-independent)
│   ├── banco_pt_process_tracing.xlsx           (v1, 2026-05-05)
│   ├── banco_pt_process_tracing_v2.xlsx        (v2, 2026-05-05)
│   ├── banco_pt_process_tracing_v3.xlsx        (v3, 2026-05-05)
│   ├── banco_pt_process_tracing_v3_1.xlsx      (v3.1, 2026-05-05)
│   └── v4_banco_process_tracing_baesiano_abdutivo.xlsx   (v4, 2026-05-24)
│
└── drafts/                  <- intermediate outputs from conversations
    └── Genereted README_v2 about the database generation an objectives.qmd
```

---

## 2. What each subfolder is for

### 2.1 `conversations/`

Exports of the LLM chats that produced the database designs. Two providers, four sessions:

- **Perplexity** (two exports). The first (May 2 → May 5) covers the initial literature search on Beach & Pedersen and the v1/v2 schema generation; the second (May 5 → May 24) is the v3 structural pivot. Perplexity's COPILOT mode used `claude46sonnetthinking` as the underlying model for most questions (with `pplx_pro_upgraded` and `turbo` on a few).
- **Claude (claude.ai)** (two exports). The first conversation (May 3 → May 24) brought the schema from v1 through v4 across three weeks of substantive discussion — including a methodological turning point on May 23 when the researcher diagnosed a theory-testing bias in the design and asked for an explaining-outcome redesign. The second conversation (May 24 → May 26) is the "V5" deep-read of Fairfield & Charman (2022).

Filenames now follow the `YYYY-MM-DD_{provider}_{descriptor}.md` convention, applied 2026-05-27. The original export names are preserved in the mapping at [`conversations/rename-suggestions.md`](conversations/rename-suggestions.md).

### 2.2 `references/`

Methodological literature that informed the schema. The Fairfield & Charman 2022 book is the *primary* reference; the 2017 article is the original journal-length statement; the 2026 article + appendix are recent applications of the same Bayesian framework to climate politics replication; Fairfield & Garay 2017 is an applied substantive example on redistribution under right-wing governments — directly relevant to the tertiary-education case.

### 2.3 `xlsx_templates/`

The five abstract schema versions, named as they were exported by the LLM that produced them. The v1–v3.1 files are all dated 2026-05-05 because Perplexity generated and re-exported them in a single intensive session that day. The v4 file is dated 2026-05-24 — produced by Claude during the May 23 explaining-outcome redesign.

These files are *templates*: case-independent schema definitions. The populated working databases live at the repository root (`../Database Process Tracing Reforming Tertiary Education.xlxs.xlsx` and `../v7_banco_process_tracing_baesiano_abdutivo_manual.xlsx`).

### 2.4 `drafts/`

Intermediate outputs from the LLM conversations — currently a `.qmd` draft README produced inside the Claude bayesian-tests conversation on 2026-05-24. The file carries a provenance comment block at its top flagging it as a *pre-pivot* framing (it advocates explicit decibel-based weight of evidence, which the project later rejected). Preserved for historical legibility, not as a methodological reference.

---

## 3. How the pieces connect

```
   references/  ────────────►  reading informs ──────────►  conversations/
                                                                   │
                                                                   ▼
   xlsx_templates/  ◄────────  conversations produce templates ◄───┘
        │
        ▼
  drafts/  (intermediate readmes / sketches not adopted in root docs)
```

The chronological story of how this works in practice — which conversation produced which template, what each iteration learned, where the methodology pivoted — is in [`../HISTORY.md`](../HISTORY.md).

---

## 4. What is *not* here

- The populated working databases (`v7_...xlsx` and `Database Process Tracing ...xlsx`) live at the repository root, not inside this folder. This folder holds *development history*, not *research data*.
- The canonical methodology synthesis is at [`../METHODOLOGY.md`](../METHODOLOGY.md), not duplicated here.
- The repository-level decision log is [`../NEWS.md`](../NEWS.md).

---

## 5. Conventions

| Item | Convention |
|------|-----------|
| Conversation filenames | `YYYY-MM-DD_{provider}_{descriptor}.md` (applied 2026-05-27). Original export names preserved in `conversations/rename-suggestions.md`. |
| Reference PDF filenames | `Authors-Year_Short-Title.pdf` — applied during the 2026-05-26 reorganization. |
| Template xlsx filenames | As exported. `xlxs_templates/` (with typo) was renamed to `xlsx_templates/` on 2026-05-26. |
| Drafts | Carry a provenance comment block at the top explaining when and how they were generated, and whether they reflect the current methodology. |

---

## 6. See also

- [`../README.md`](../README.md) — repository overview.
- [`../METHODOLOGY.md`](../METHODOLOGY.md) — guiding epistemological premises.
- [`../NEWS.md`](../NEWS.md) — methodological decision log.
- [`../HISTORY.md`](../HISTORY.md) — narrative chronology of database evolution.
- [`../REPOSITORY_PLAN.md`](../REPOSITORY_PLAN.md) — repository organization rationale.
