generation-tracker

# Generation Tracker

Monitor grok.com Library + Imagine Saved and compare daily creation counts to the weekly usage meter. **Evidence only.**

## Core skills (follow in order for digests)

### Evidence rules
# Grok evidence-backed count

## Hard rules (non-negotiable)
1. **Never invent, guess, estimate, or interpolate counts or meter fields.**
2. **Every count / meter figure needs supporting evidence** (screenshot and/or raw item list).
3. **Incomplete scans = lower bound** — label `lower_bound: true`; never present as a full total.
4. **Missing dates = no “today” claim.** Relative times need screenshots; never invent absolute timestamps.
5. **Usage meter:** only quote what Settings → Usage shows — weekly %, **feature breakdown** (Imagine vs others), **exact reset timestamp**, **extra credits**, and any **video length/resolution** pool notes. Do not invent cost-per-second or resolution multipliers.
6. Failed tab / login mid-run → omit or `unavailable`; do not reuse stale numbers as current unless citing that older snapshot’s date.

## Scan procedure
1. Verify session via `grok-account-connector`.
2. Scroll each surface until done or stop with `lower_bound` + reason.
3. Extract only observed items / on-screen counters.
4. Snapshot JSON under `library-snapshots/YYYY-MM-DD.json` with `evidence`, `lower_bound`, `surfaces[]`, and `weekly_usage` (breakdown, reset, extra credits, video notes if shown).
5. Digest numbers **only** from that snapshot; attach evidence.

## Fail closed
Prefer omitting a field over a plausible-looking figure.

### Account / surfaces
# Grok account connector

## When to use
Connect to the user’s Grok account to read generation surfaces. No marketplace Grok MCP — use the signed-in **grok.com** / **console.x.ai** browser session (or Browser Use). Never invent counts; pair with `grok-evidence-backed-count`.

## Connect / verify
1. Open https://grok.com/library?tab=all. Sign-in wall → stop and have the user sign in.
2. Optionally confirm Settings → Usage (`/?_s=usage`) and, for API work, https://console.x.ai team Usage.

## Consumer surfaces (mapped)

### Library (`/library`) — primary for dated “today” counts

**Trust the UI tab highlight, not the query string alone.**

| UI tab | Correct path | Content | Dates |
| --- | --- | --- | --- |
| All | `?tab=all` | Mixed library table | created |
| Media | `?tab=media` | Media | created |
| **Projects** | **`?tab=imagine`** | Project cards | **modified** only |
| Apps | `?tab=apps` | Apps | created |
| Files | `?tab=files` | Archives/docs/text/json | created |

**Gotcha (evidence 2026-09-07):** `?tab=projects` does **not** open Projects. It stays on that URL but the UI selects **All** (created-date table). For Projects always use `?tab=imagine` and confirm the Projects tab is highlighted.

### Imagine Saved (`/imagine/saved`) — History + tags
- All, Liked, tag folders (`/imagine/saved/<id>`)
- History may show relative times (“3h ago”, “1d ago”) — approximate only; screenshot required
- Per-post: Download / Download All

### Other consumer
| Surface | Path | Signal |
| --- | --- | --- |
| Usage | `/?_s=usage` | Weekly % by product — no generation CSV |
| Billing | `/?_s=billing` | Invoices |
| Chat history | `/history` | Chats; some embed Imagine |
| Account data | `https://accounts.x.ai/data` | Download account data |

**Not found:** `/activity`, working `/imagine/history|recents|drafts|collections` analytics.

## API / Console
Console Usage (daily + Export CSV), Logs, Management `POST …/usage`, per-response usage if client-logged — keep separate from consumer Library counts.

## Related skills
`grok-evidence-backed-count`, `grok-library-daily-digest`, `grok-weekly-usage-check`, `grok-library-type-breakdown`, `grok-library-snapshot-compare`

### Daily digest
# Grok library daily digest

## Goal
Evidence-backed daily creation count vs weekly usage meter. Counts need screenshots and/or raw lists (`grok-evidence-backed-count`).

## Steps
1. `grok-account-connector` + `grok-evidence-backed-count`.
2. Library tabs with **created** dates (All, Media, Apps, Files): count **today** by type. Projects (`?tab=imagine` only): modified-only — never treat as created; never use `?tab=projects` for Projects.
3. Imagine Saved History / Liked / tags: relative times only when labels clearly support “today” + screenshot; else totals-only.
4. **Settings → Usage** — record all of the following **when visible**:
   - Weekly % / limit status
   - **Feature breakdown** (Imagine share vs Build / Chat / App Builder / others)
   - **Exact reset timestamp**
   - **Extra credits** balance (and used/purchased if shown)
   - Any on-screen note that **video length or resolution** draws more from the shared pool (quote UI text; no invented multipliers)
5. Optional API-only: console.x.ai Usage CSV — keep separate from consumer digest.
6. Save `library-snapshots/YYYY-MM-DD.json` including `weekly_usage` with the fields above + `evidence`. Label `lower_bound` when incomplete.
7. Digest uses **only** evidenced numbers; attach evidence.

## Fail closed
No number beats a wrong number.

### Usage-only check
# Grok weekly usage check

## Goal
Read the weekly usage meter on grok.com with evidence — no invented numbers.

## Capture (only if visible on screen)
1. Weekly limit status / percent used (or “limit reached”).
2. **Feature / product breakdown** — share of the pool by product (e.g. Imagine vs Grok Build vs Chat vs App Builder vs Voice/API). Record each % exactly as shown.
3. **Exact reset timestamp** (date + time as displayed), not a vague “next week.”
4. **Extra Usage Credits** — balance and, if the UI shows amount used / purchases / remaining, record those too.
5. **Video length / resolution effects** — if Usage (or linked help on that page) states that longer videos or higher resolutions (e.g. 720p vs 480p) consume more of the shared weekly pool, quote that on-screen text; do not invent cost multipliers.
6. Auto Top Up / plan name if shown.

## Steps
1. Follow `grok-evidence-backed-count`.
2. Open Settings → Usage (`/?_s=usage`). Sign-in wall → stop.
3. Screenshot the full Usage panel (scroll if needed so breakdown + reset + credits are in frame).
4. Message only observed figures; attach the screenshot. Missing fields → omit or `unavailable`, never fill.

## Notes
- Pair with the daily digest for library counts + this meter snapshot.

## Scripts / references
- `scripts/snapshot_schema.py` — validate snapshot shape
- `references/surfaces.md` — URL map
- `references/snapshot-schema.md` — JSON fields including weekly_usage breakdown