# popback

Rate TMDB movies in a TUI, backfill ratings to popfeed. Resumes where you
left off -- close it, come back, keep rating.

## Why

Popfeed's own backfill is unreliable: duplicates, broken pagination, missing
ratings. This reads your CSV of rated movies and backfills one entry at a
time with proper error tracking.

## Usage

    export TMDB_API_KEY=...
    popback viewed [output.csv]

    export TMDB_API_KEY=... POPFEED_AUTH_TOKEN=...
    popback backfill [input.csv]

Resume is automatic: `viewed` skips already-rated movies from the CSV.
`backfill` skips entries already posted to popfeed.

## How it works

### viewed

Streaming TUI: movies arrive from TMDB one page at a time (~20). A
background fetch runs while you rate so the next movie is always ready.
Press `b` to go back and re-rate. Results write to CSV on exit.

```
TMDB pages --> [pre-fetch buffer] --> movie at cursor --> rate/skip --> CSV
```

### backfill

Reads CSV, posts each entry to popfeed's internal API two-step: a review
(rating + metadata) and an experience (watched/seen timestamp). Pure
request builders are tested separately from HTTP I/O.

### popfeed internal API

No public docs. Endpoints reverse-engineered from browser devtools:

- `POST /api/rest/v1/review`
- `POST /api/rest/v1/experience`

Auth: JWT token from browser session, sent as cookie.

### Persistence

CSV format: `id,rating,seen,skip`. Human-readable, append-only. The store
interface is mocked in tests so the write path is tested without disk.

## Design

- Sandwich pattern: impure gather (CSV, TMDB) -> pure logic (build
  requests, dedup) -> impure commit (write CSV, post HTTP).
- Streaming TUI: no batch cap. Async pre-fetch hides network latency.
- The CSV is the skip set. No in-memory state across restarts.
- slog to stderr for ops, fmt to stdout for UX. No log noise.
- Table-driven tests, no shared mutable state.

## Environment

    TMDB_API_KEY           TMDB key or JWT bearer token
    POPFEED_AUTH_TOKEN     Popfeed session JWT (for backfill)