tiktok-md
Corpus spans 2018-02-28 → 2026-07-25 · rendered in 887ms
✓ Search index✓ Ledger backlog! Ledger cursors! Stats time series
Posts
647,032
141,165 photo · 505,867 video
Roster (MD)
47,737
of 83,900 accounts discovered
Unresolved
33,449
39.9% with no country
Hashtags
80,320
ranked by frequency →
Text coverage
OCR target sums two disjoint populations: videos whose speech failed (61,070) and photo posts (141,165), which carry no audio track at all.
ASR502,883 remaining
144,149
22.3%
OCRof 202,235 target
284,024
140.4%
ASR verdicts
Measured across all 144,149 transcribed posts.
speech
81,798
56.7%
no_speech
40,027
27.8%
suspect_en
13,326
9.2%
empty
4,872
3.4%
hallucination
4,126
2.9%
Language of speech
From the
post_speech view. Counting raw lang instead would report English as the largest language. None of those rows are speech.ro
42,548
53.1%
ru
32,325
40.4%
es
2,102
2.6%
pt
1,908
2.4%
fr
628
0.8%
tr
588
0.7%
10.0% of these 81,798rows repeat another post’s transcript word for word: 8,195 rows, mostly viral sounds reused by many creators. Counted once each instead, Romanian rises ~1.7pp and Russian falls ~1.1pp. Use post_speech_distinct for distributions over what was said.
Ledger cursors
Each remote host appends to a ledger; the laptop stores a byte offset. A cursor that stops advancing is how the media sync sat at 31 rows for ten hours.
| Host | Ingested | Offset | Last sync |
|---|---|---|---|
| eagle | 322,648 | 36,366,010 | 22.1h ago |
| m5-vision:ocr | 184,751 | 34,678,922 | 12.5h ago |
| rtx4090-24-01:asr | 72,917 | 35,336,896 | 12.5h ago |
| rtx4090-24-01:ocr | 107,298 | 19,294,754 | 12.5h ago |
| rtx4090-24-02:asr | 71,432 | 34,684,550 | 12.5h ago |
| rtx4090-24-02:ocr | 93,321 | 16,845,848 | 12.5h ago |
| rtx5070-16:ocr | 154,644 | 26,951,629 | 12.5h ago |
API quota
Reserved before each request, so a crash cannot refund budget.
Spent552,077 left
47,925
8.0%
Projected 322,927 calls/month at the current tier cadence.
Poll cadence
Interval derived from the scheduler’s own
next_due_at, not assumed.| Tier | Accounts | Every | Calls/mo |
|---|---|---|---|
| dormant | 28,040 | 30.0d | 28,040 |
| C | 11,512 | 7.0d | 49,337 |
| B | 8,185 | 1.0d | 245,550 |
Corpus by publication month
Polling depth, not posting volume. One poll returns one page of ≤30 posts, so a daily poster yields a month of history and a monthly poster yields 2.5 years. The last 12 months hold 78% of the corpus because of how it was collected, not because Moldova started posting more.
Table view
| Month | Posts |
|---|---|
| 2026-07 | 160,602 |
| 2026-06 | 95,983 |
| 2026-05 | 63,449 |
| 2026-04 | 39,457 |
| 2026-03 | 32,018 |
| 2026-02 | 25,312 |
| 2026-01 | 22,710 |
| 2025-12 | 19,896 |
| 2025-11 | 14,247 |
| 2025-10 | 12,936 |
| 2025-09 | 9,867 |
| 2025-08 | 9,793 |
| 2025-07 | 9,893 |
| 2025-06 | 8,728 |
| 2025-05 | 7,833 |
| 2025-04 | 6,925 |
| 2025-03 | 6,646 |
| 2025-02 | 5,595 |
| 2025-01 | 5,705 |
| 2024-12 | 5,716 |
| 2024-11 | 4,928 |
| 2024-10 | 4,872 |
| 2024-09 | 4,305 |
| 2024-08 | 4,602 |
| 2024-07 | 4,591 |
| 2024-06 | 4,685 |
| 2024-05 | 4,033 |
| 2024-04 | 3,600 |
| 2024-03 | 3,763 |
| 2024-02 | 3,000 |
| 2024-01 | 3,164 |
| 2023-12 | 2,837 |
| 2023-11 | 2,469 |
| 2023-10 | 2,248 |
| 2023-09 | 2,056 |
| 2023-08 | 2,107 |
| 2023-07 | 1,957 |
| 2023-06 | 1,858 |
| 2023-05 | 1,981 |
| 2023-04 | 1,776 |
| 2023-03 | 1,392 |
| 2023-02 | 1,360 |
| 2023-01 | 1,308 |
| 2022-12 | 1,396 |
| 2022-11 | 1,275 |
| 2022-10 | 1,199 |
| 2022-09 | 814 |
| 2022-08 | 1,030 |
| 2022-07 | 1,098 |
| 2022-06 | 1,082 |
| 2022-05 | 856 |
| 2022-04 | 751 |
| 2022-03 | 571 |
| 2022-02 | 408 |
| 2022-01 | 489 |
| 2021-12 | 414 |
| 2021-11 | 283 |
| 2021-10 | 277 |
| 2021-09 | 244 |
| 2021-08 | 236 |
| 2021-07 | 196 |
| 2021-06 | 182 |
| 2021-05 | 171 |
| 2021-04 | 161 |
| 2021-03 | 263 |
| 2021-02 | 320 |
| 2021-01 | 317 |
| 2020-12 | 134 |
| 2020-11 | 97 |
| 2020-10 | 97 |
| 2020-09 | 44 |
| 2020-08 | 98 |
| 2020-07 | 70 |
| 2020-06 | 47 |
| 2020-05 | 32 |
| 2020-04 | 21 |
| 2020-03 | 27 |
| 2020-02 | 32 |
| 2020-01 | 16 |
| 2019-12 | 13 |
| 2019-11 | 4 |
| 2019-10 | 10 |
| 2019-09 | 1 |
| 2019-08 | 2 |
| 2019-07 | 7 |
| 2019-06 | 4 |
| 2019-05 | 3 |
| 2019-04 | 3 |
| 2019-03 | 1 |
| 2019-01 | 2 |
| 2018-11 | 4 |
| 2018-10 | 16 |
| 2018-09 | 1 |
| 2018-08 | 7 |
| 2018-07 | 1 |
| 2018-02 | 2 |
Topic tagging
gpt-4.1-nano · prompt v5 · run ok · 1131/min
tagged0 remaining
16,681
100.0%
Spent
$2.17
Per post
$0.000130
Projected
$2.17
Failed
6
other
8,792
52.7%
culture_history
2,868
17.2%
family_relationships
1,087
6.5%
gaming
984
5.9%
humor_memes
593
3.6%
weather_environment
371
2.2%
Spot check
Most recently ingested text, for eyeballing quality. Ingest order, not publication order. Sorting by publication date costs a 2s join and changes nothing about what these are useful for.
ocr—Bărbat misterios, 4 bucăți plasă cumpărate. DIN INFORMAȚIILE CARE | Opus( POTRIVIT UnOR SURSEocr—Porumb dulceocr—invinsa 24 de vinuri qm 5 CU nunse une ni poliubomu pVsY eegeleocr—25 iul 2024 Amintireocr—Moldova ROTUNDA TlkTok | REC Moldova ROTUNDA Edinet ROTUNDA Ilktol 4K 6OFPS HDocr—Daca nopupafi voiuc am hotfrâcsinoisnepupam ua pe altadiuplcaou dinbârfoocr—Am un singur cont cu 36.000 urmaritori atenți Kadgn Fițiocr—Ai stiut ca postind astfel de video poti cistiga doar cuun telefon Nu promit bogatie peste noapte dar cu impli
Pipeline checks
Standing checks for the findings in
BACKEND-AUDIT-2026-07-26.md, so a regression surfaces here rather than in the next hand audit.B3Search index
270,094 rows, in syncB5Ledger backlog
no deferred recordsB2Ledger cursors
eagle last synced 22.1h agoB1Stats time series
one observation per video, no deltas yet