Every number here was measured from the repository at publish time. Where two counts have been quoted for the same thing, both appear with their denominators.
20,041 counts all 9 bank files.
19,718 counts only the 7 evergreen locale banks.
The difference is platform_surface (250) plus apac_localization (73).
Neither figure was wrong; they answer different questions. Say which one you mean.
| Stage | State | Measured | Layer |
|---|---|---|---|
| Atom banks | done | 20,041 rows across 9 files. 7 locale banks plus platform-surface and APAC-localisation sets. | EXECUTED-REAL |
| Localisation | done | All 7 locale banks populated. NVIDIA NIM validated for CJK; zh-TW passes only with Simplified-contamination detection on every batch. | EXECUTED-REAL |
| Licensed media | done | 1,763 licensed files on disk. Asset map carries 22 rows, all LICENSED. | EXECUTED-REAL |
| Video bank | done | 237 videos indexed across 25 topics (323 mp4 on disk). | EXECUTED-REAL |
| Copy quality (gate 50) | done | All 7 locales at or below their locked dup ceiling. en_US 0.0%; worst is ko_KR 29.9%. | EXECUTED-REAL |
| Dedup (gate 49) | failing | FAILING on purpose: 1,329 within-bank verbatim duplicates, all CJK. en_US and platform-surface are clean at 0. | EXECUTED-REAL |
| Voice / TTS | blocked | 0 WAV files on disk. NIM Magpie is reachable and the client is landed, but no locale has been voiced at scale. | CODE-WIRED |
| Publishing | blocked | Not wired. Metricool absent; draft-only and manual review still required. No post has been published from this pipeline. | CONFIG-EXISTS |
The copy, the images and the video are real and licensed. Nothing has ever been published. Publishing is unwired, Metricool is absent, and no locale has been voiced at scale. A green row above means the asset exists — not that it reached an audience.
| Locale | Atoms | Verbatim dups | Dup rate | State |
|---|---|---|---|---|
| English (US)en_us | 2,970 | 0 | 0.0% | clean |
| Koreanko_KR | 2,910 | 351 | 29.9% | needs work |
| Chinese (Taiwan)zh_TW | 2,910 | 68 | 4.2% | watch |
| Chinese (Singapore)zh_SG | 2,782 | 316 | 24.6% | needs work |
| Chinese (Mainland)zh_CN | 2,758 | 122 | 12.1% | watch |
| Japaneseja_JP | 2,694 | 113 | 8.7% | watch |
| Chinese (Hong Kong)zh_HK | 2,694 | 358 | 29.3% | needs work |
Duplicates are within a bank. IDs shared across locale banks are translation sets and are not defects — 380 of them, by design. The duplication is CJK-only; every English bank is clean, which points at the translation lane collapsing distinct sources onto identical output rather than at the original authoring.
| Family | Rows |
|---|---|
| HOOK_COVER | 3,710 |
| REFRAME | 2,049 |
| CTA_ANCHOR | 1,873 |
| SOMATIC_SETUP | 1,849 |
| SAVEABLE_PAYOFF | 1,818 |
| MECHANISM_EXPLAINER | 1,790 |
| TOOL_STEP | 1,680 |
| PROBLEM_AGITATION | 1,457 |
| BRIDGE | 1,393 |
| CAROUSEL_SLIDE | 638 |
| VIDEO_BEAT | 498 |
| THREAD_UNIT | 368 |
Generated from the repository at publish time — atom banks, the Storyblocks
asset map, VIDEO_TOPIC_INDEX.json, and live runs of gates 49 and 50.
Video index generated 2026-08-09T06:26:30Z.