# Head-to-head: mathtec/youtube-channel-scraper (ours) vs streamers/youtube-channel-scraper (Store leader), 2026-11-09

Bead: hustle-qtk. Purpose: find out whether our Store listing can honestly say "better", and if so where.

Leader confirmed from the public Store API on 2026-11-09 (search "youtube channel", sorted by 30-day users): **streamers/youtube-channel-scraper ("Fast YouTube Channel Scraper"), 3,770 users**. Second is grow_media/youtube-channel-video-scraper with 1,609 users (`h2h-google-trends/actors/store-search-2026-11-09.json`). Ours had 2 users in 30 days.

## Outcome first

- **Completeness: equal (n=9 channels, 165 videos).** Both Actors returned **165/165** of the newest ids shown on the channel's own tab pages (15 per tab requested: videos, plus Shorts on Veritasium and Live on NASA). Both returned them in the page's order (7/7 single-tab channels in exact order), with no duplicates and no ids from outside the tab. Against the RSS feed (the 15 newest uploads of any type), both returned the same RSS ids on every channel. Every RSS id that neither returned (39) is on a Shorts or Live tab that was not requested (checked against those tabs' pages).
- **Success and failure handling: equal in shape.** 9/9 real channels succeeded on both. The deleted channel (`/user/CraigSWrightCSU`, YouTube 404) got one error row from each, ours "YouTube answered 404: this channel does not exist" and the leader `CHANNEL_DOES_NOT_EXIST`. **Both charged for it:** ours **$0.0014** (its Actor-start event; the error row itself was free) and the leader **$0.001** (the error row was charged as a result).
- **Price: ours is lower.** On this STARTER (BRONZE) account, over the 9 real channels plus the failed one: ours **$0.047 for 165 videos = $0.000285 per video, $0.0047 per channel**; leader **$0.166 = $0.001006 per video, $0.0166 per channel**. So ours costs **0.28x as much**. Ours' per-video event is $0.0002 (BRONZE) plus a $0.0014 start per run. The leader charges a flat $0.001 per row and has no start fee. At 15 videos per run, the start fee is about a third of ours' cost.
- **Speed: ours faster at the median and tail.** Run seconds: median **3.9 s ours vs 6.5 s leader**, p90 6.3 vs 36.7, max 16.5 vs 36.8 (n=10 each). The leader took 36–37 s on 2 of 10 channels and ours took 16.5 s on 1 (NASA, videos + Live).
- **Fields: each has some the other lacks.**
  - **View counts:** ours returns **exact** views on 150/165 rows (`viewCountIsApprox: false`). The 15 Shorts are YouTube-rounded and flagged `viewCountIsApprox: true`. The leader's views are rounded on all rows (e.g. 2,800,000 where ours has 2,897,017). The two differ on 118/165 shared videos.
  - **Shorts:** the leader returns an exact upload date (e.g. `2026-11-06`) for all 15 Shorts. Ours returns no date and no duration for Shorts.
  - **Live now:** ours flags `isLiveNow: true` on the 2 live NASA streams; the leader shows `duration: "LIVE"` on them.
  - **Channel block:** both return channel id, name, exact video count (same on 9/9, e.g. 1,846 and 6,201), subscribers (same rounded value on 9/9), total views, join date, country/location and links. The leader adds banner URL and age-restriction; ours adds handle, keywords and family-safe.
  - **Dates and durations:** dates on 148/165 ours vs 163/165 leader; durations on 148/165 ours vs 150/165 leader. Ours misses dates on Shorts (15) and live-now streams (2).
- **Type labels:** ours labels rows `video` / `short` / `stream`. The leader labels them `video` / `shorts` / `stream`.

**Claims the numbers support for the listing:**
- "Same videos as the category leader at about 28% of the price" (165/165 vs 165/165 against the channel page).
- "Exact view counts on regular videos and streams" (150/165; the leader's are rounded on all rows).
- "Videos, Shorts or Live tab per channel link" (a schema fact, exercised on 2 channels).
- "Failed channels come back as a row with the reason" (n=1; the leader does the same).

**Claims the numbers do NOT support:**
- "More complete" or "finds more videos": both were 100% on this sample.
- "Failed channels are free": ours charged its $0.0014 start fee on the 404 channel, more than the leader's $0.001 error row.
- "More fields": the leader has upload dates on Shorts and the banner; ours has exact views and live-now. Neither is a superset.

## Method

1. **Inputs (10 channels).** The 4 channels in `services/digest/channels.json` (Karpathy, Wolfram R&D, CoinGeek via its legacy `/c/` link, CryptoCurrent) and the one channel that file excludes because it no longer exists (`/user/CraigSWrightCSU`). To these I added Veritasium (Shorts), NASA (Live), 3Blue1Brown, MrBeast and Lex Fridman for size and type mix. The list and reasons are in `inputs.json`.
2. **Reference.** From withheld (residential IP), before and after the runs (`scripts/chprobe.py` → `ref/probe-pre.json`, `ref/probe-post.json`), I fetched each channel's `/videos`, `/shorts` and `/streams` tab page and parsed `ytInitialData`. That gives the ordered first-page ids, whether the tab exists (YouTube falls back to Home when it does not) and the header "N videos". I also fetched the RSS feed `https://www.youtube.com/feeds/videos.xml?channel_id=<id>`. None of this passed through either Actor. The pre and post snapshots gave the same first 15 ids on every requested tab, so no upload landed during the runs.
3. **Runs.** All runs came from the Hustle org (`mathtechustle`, STARTER/BRONZE). This org does not own `mathtec/youtube-channel-scraper`, so it pays the public price. Each channel got one single-channel run per Actor, started at the same moment (5 channels, so 10 runs, at a time). Every run was async with `maxTotalChargeUsd` (ours $0.01 + $0.0002 × max rows; leader $0.0013 × max rows + $0.01) and a client guard on dataset size. Default memory (ours 256 MB, leader 1024 MB) and build; no Actor settings or prices changed.
4. **Equivalent settings** (exact inputs are in each `runs/*/<n>.json`):

   | Actor | input |
   |---|---|
   | ours | `channels: [<link>/videos]` (plus `<link>/shorts` for Veritasium, `<link>/streams` for NASA), `maxVideosPerChannel: 15`, `includeChannelInfo: true` (default) |
   | leader | `startUrls: [{url: <link>}]`, `maxResults: 15`, `maxResultsShorts: 15` (Veritasium) else 0, `maxResultStreams: 15` (NASA) else 0, `sortVideosBy: "NEWEST"` |

   15 per tab matches the RSS feed's length.
5. **Metrics.** Matched = returned ids ∩ the tab's first 15 page ids. Order = the returned id list equals the page's first 15. $ is `usageTotalUsd` per run after the runs settled; on every run it equals charged events × event price. Seconds are `stats.runTimeSecs`.

## Results

| Actor | channels with videos | video rows | matched / reference | $ total | $ per video | $ per channel | median s | p90 s | max s |
|---|---|---|---|---|---|---|---|---|---|
| ours | 9/9 (+1 error row) | 165 | 165 / 165 | 0.047 | 0.000285 | 0.0047 | 3.9 | 6.3 | 16.5 |
| leader | 9/9 (+1 error row) | 165 | 165 / 165 | 0.166 | 0.001006 | 0.0166 | 6.5 | 36.7 | 36.8 |

## Per channel

| # | channel | tabs asked | page header (videos) | ref ids (first 15 per tab) | ours matched / rows | leader matched / rows | RSS newest-15 in output ours / leader | channel video count ours / leader | $ ours / leader | run s ours / leader |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | @AndrejKarpathy | videos | 17 videos | 15 | 15 / 15 | 15 / 15 | 15 / 15 | 17 / 17 | 0.0044 / 0.015 | 4.5 / 14.2 |
| 2 | @WolframRD | videos | 281 videos | 15 | 15 / 15 | 15 / 15 | 12 / 12 | 281 / 281 | 0.0044 / 0.015 | 3.9 / 6.2 |
| 3 | c/coingeek | videos | 1.8K videos | 15 | 15 / 15 | 15 / 15 | 2 / 2 | 1846 / 1846 | 0.0044 / 0.015 | 3.5 / 5.1 |
| 4 | @Crypto_Current | videos | 85 videos | 15 | 15 / 15 | 15 / 15 | 15 / 15 | 85 / 85 | 0.0044 / 0.015 | 4.8 / 36.8 |
| 5 | @veritasium | videos+shorts | 548 videos | 30 | 30 / 30 | 30 / 30 | 15 / 15 | 548 / 548 | 0.0074 / 0.03 | 4.0 / 36.7 |
| 6 | @NASA | videos+streams | 6.2K videos | 30 | 30 / 30 | 30 / 30 | 14 / 14 | 6201 / 6201 | 0.0074 / 0.03 | 16.5 / 6.6 |
| 7 | @3blue1brown | videos | 247 videos | 15 | 15 / 15 | 15 / 15 | 4 / 4 | 247 / 247 | 0.0044 / 0.015 | 6.3 / 6.9 |
| 8 | @MrBeast | videos | 1K videos | 15 | 15 / 15 | 15 / 15 | 6 / 6 | 1012 / 1012 | 0.0044 / 0.015 | 3.4 / 6.1 |
| 9 | @LexFridman | videos | 870 videos | 15 | 15 / 15 | 15 / 15 | 14 / 14 | 870 / 870 | 0.0044 / 0.015 | 3.2 / 5.6 |
| 10 | user/CraigSWrightCSU | videos | 404 | - | error row: "YouTube answered 404: this channel does not exist" | error row: "CHANNEL_DOES_NOT_EXIST" | - | - | 0.0014 / 0.001 | 2.0 / 6.4 |

RSS ids not in either output were all on tabs that were not requested: Wolfram R&D 2 Live + 1 Short, CoinGeek 13 Shorts, NASA 1 Short, 3Blue1Brown 11 Shorts, MrBeast 9 Shorts, Lex Fridman 1 Short.

## Spend

- **This test (attributed per run): $0.213.** Ours $0.047 + leader $0.166. The three h2h tests in this batch came to $0.8734 together, against a hard cap of $1.50 shared across all three.
- **`/v2/users/me/limits`:** one before/after pair covers the whole three-test batch (`h2h-google-trends/limits-before.json`, `limits-after.json`): a delta of $0.8765, against $0.8734 summed over the 104 run ids. No other org runs started in the window, so $0.003 is unattributed platform usage.

## Limits of this test

- Only the newest 15 per tab were measured. Deep pagination (hundreds of videos per channel) was not, and that is where the two could differ.
- `sortVideosBy: POPULAR/OLDEST` (leader) and `oldestPostDate` were not tested. Ours has no sort or date filter.
- One run per channel at one time of day. The leader's two 36 s runs may be noise; n=10.

## Files

All under `/home/a/Projects/swarm/agents/hustle/reports/market/h2h-youtube-channel/`:
- `inputs.json`: the 10 channels, tabs, reasons and sources.
- `ref/probe1.json` (selection probe), `ref/probe-pre.json`, `ref/probe-post.json`: the independent tab-page and RSS fetches from withheld.
- `actors/*.json`: each Actor's metadata, pricing and latest-build input schema at test time.
- `runs/{ours,leader}/<n>.json`: input, cap, settled run object (`run`, plus `run_at_capture`), dataset items, log and client timing.
- `per-channel.json`, `summary.json`: the computed metrics.
- `scripts/`: `apify.py`, `runlib.py`, `chprobe.py`, `runner.py`, `analyse.py`. The token is read from `.secrets/apify-org.env` at run time and is never written to disk.
