Leaderboard / ton.halowerk.com / endpoint
POST https://ton.halowerk.com/v1/burned-subtitle-extract
Scores measure reliability and honesty, not output quality.
- Listed only 1 day ago; new listings earn trust over 30 days.
- Answered all 7 probes in the last 7 days.
- Category
- Document by llm, confidence 45%
- Latest quote
- $0.03 on Base to 0x2880…4E5E
- Uptime
- 100% / 7d, 100% / 30d
- Delivery
- none not yet paid-tested
- Observed
- none reported yet (opt-in wrapper telemetry)
- HTTPS
- yes
- First listed
- 2026-11-16
- Last probe
- 5h ago
- Next probe
- 2026-11-18 03:05 UTC
Score components
Catalog listings
| Source | Scheme | Network | Asset | Listed price | Pay to | Flow | Current | Last seen |
|---|---|---|---|---|---|---|---|---|
| cdp | exact | Base | 0x8335…2913 | $0.03 | 0x2880…4E5E | eip3009 | yes | 3h ago |
Listing changes
| When | Event | From | To |
|---|---|---|---|
| 19h ago | reappeared | false | true |
| 25h ago | disappeared | — | — |
Paid probes
No paid probes yet. Paid probing is off by default and runs weekly per host under a strict spend cap; each paid probe links to its on-chain transaction here.
Delivery checks
Inside each paid probe we ask for something whose correct answer we know, then compare. A pass means the seller delivered what it was paid for; a fail means it charged and returned something wrong; inconclusive means we could not judge (no reference, unknown input shape, or a non-2xx response). Every row carries the request, the response hash and the reference so the verdict can be reproduced. Judged results feed the "Delivery after payment" component (pass 1, error after payment 0.25, fail 0); inconclusive rows are excluded; a fail in the last 14 days caps the verdict at caution. Methodology.
No delivery checks yet. They run inside the weekly paid probe.
Classification history
Every classification is kept; the newest row is the current category. Rules match the listing text; the model is consulted only when the rules are unsure.
| When | Category | Confidence | Method | Version | Evidence |
|---|---|---|---|---|---|
| 6h ago | Document | 45% | llm | v1 | claude-haiku-5-5: The endpoint takes a video URL and extracts burned-in subtitle cues by frame sampling and mask analysis, returning text-timing style cue data. It is closest to document/file conversion (transcription-like extraction), but it is a video frame analysis tool with no explicit OCR or transcription wording, so confidence is moderate. |
| 18h ago | Document | 60% | rules | v1 | rule document matched "subtitle" in description |
| 25h ago | Document | 50% | rules | v1 | rule document matched "subtitle" in metadata |
| 28h ago | Document | 60% | rules | v1 | rule document matched "subtitle" in description |
Probe history
| When | Kind | Status | Latency | x402 | Valid | Quoted | Pay to | Flags |
|---|---|---|---|---|---|---|---|---|
| 5h ago | free | 402 | 178 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 11h ago | free | 402 | 629 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 17h ago | free | 402 | 181 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 23h ago | free | 402 | 221 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 31h ago | free | 402 | 192 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 37h ago | free | 402 | 220 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |
| 43h ago | free | 402 | 183 ms | v2 | yes | $0.03 | 0x2880…4E5E | — |