🛡️ Keyed text-watermark detection

Keyed Text Watermark Detection

Detect text generated through a TrueSynthesis-marked model, with a stated false-positive bound. Text that was not marked with a key we hold returns no mark found, which does not mean a person wrote it.

⚡

Stripe Sandbox / Test Mode ActiveTest Mode

Simulate full credit top-ups with zero real billing charges. Use test card 4242 4242 4242 4242 (any future date, CVC 123).

Test Checkout Flow ↓

Cryptographic Provenance & Watermark Workbench

Detect a keyed statistical text watermark (demo key) and inspect text for hidden Unicode characters. Text that was not marked with a key we hold returns “no mark found”, which is not a statement that a human wrote it. In our benchmark, 83% of 120 marked prose samples were detected (95% interval 76–89%) with 0 of 120 unmarked samples flagged; passages under about 100 tokens are essentially undetectable, and code detection is not established.

Load specimen:

One day, Tim met a mysterious letter from an unknown sender. It was addressed to Tim's friend, a local tourist named Jack. Jack, who had never been to Africa, knew that Tim's friend was interested in the situation and wanted to make sure he received his letter promptly.

Tim quickly examined the letter and quickly realized that it was written by a man who had been fishing his boat for months and was missing from his boat. He immediately spoke to Jack and told him that Jack had received the letter from a local fishing company.

After several hours of conversation, Jack was able to understand Tim's message and respond. Jack replied with a polite greeting, and he was willing to work with Tim until he could find the missing boat.

Despite Jack's patience and hard work, Tim was able to find the boat and restore it to its original state. Jack then shared his experiences with Tim and they were able to make a positive impact on the community.

The lighthouse keeper was pleased to have been able to recover the missing boat and is pleased to have been able to make a positive impact on the community.

1381 chars · 248 words

The floor only enables a statistical test. An older-model planning example needed about 250 scored positions for 90% modeled detection, versus about 2% at the 52-position floor. Actual sensitivity depends on model, text and edits.

⚡ Stripe Test Mode Product Catalog

Instant Test Top-Ups & Subscriptions

Choose a tier to test the end-to-end Stripe Checkout flow. Charges are simulated and credit ledger balances are credited immediately.

One-Time PackDEMO MODE

Starter Credit Pack

Ideal for small audits, ad-hoc verifications, and testing.

$9.00USD (Test)

✓ 50,000 Credits included
✓ Keyed-watermark detection (demo key)
✓ Zero-width character inspection

Featured pack

High VolumeTEST

Pro Credit Pack

For high-volume publishing pipelines and forensic audits.

$29.00USD (Test)

✓ 250,000 Credits included
✓ Keyed-watermark detection (demo key)
Priority queue and batch API: not yet available

Monthly PlanMONTHLY

Pro Provenance

Monthly subscription, test mode only.

$39.00/ month (Test)

✓ 1,000,000 credits on first checkout
Monthly renewal grants and webhook notifications: not yet available

🔬 What the Detector Sees

Watermark Specimens

Real generations from a local Qwen2.5-Coder model, marked with the published public demo key and re-checked by the live detector. Every token is shaded by what the detector saw.

green hitred missskipped (duplicate, counted once)Ungated detection: every eligible position counts.

Marked · public demo key

Marked with the published demo key while the model generated it

WATERMARKED

PROMPTWrite a short story about a lighthouse keeper who receives an unexpected letter.

In a small town nestled along the coast of Africa, there lived a lighthouse keeper named Tim who was known for his keen eyes and excellent judgment. Tim was one of the few local lighthouses that had ever been abandoned due to the devastating storm that had fallen recently. One day, Tim met a mysterious letter from an unknown sender. It was addressed to Tim's friend, a local tourist named Jack. Jack, who had never been to Africa, knew that Tim's friend was interested in the situation and wanted to make sure he received his letter promptly. Tim quickly examined the letter and quickly realized that it was written by a man who had been fishing his boat for months and was missing from his boat. He immediately spoke to Jack and told him that Jack had received the letter from a local fishing company. After several hours of conversation, Jack was able to understand Tim's message and respond. Jack replied with a polite greeting, and he was willing to work with Tim until he could find the missing boat. Despite Jack's patience and hard work, Tim was able to find the boat and restore it to its original state. Jack then shared his experiences with Tim and they were able to make a positive impact on the community. The lighthouse keeper was pleased to have been able to recover the missing boat and is pleased to have been able to make a positive impact on the community.

SCORED

265 scoring floor 52

SKIPPED

GREEN HITS

204 / 265

Z-SCORE

8.78

77% green50% chance

204 of 265 scored positions were green (77.0%; chance is 50%). Corrected p = 1.20e-18.

Unmarked control

Same model and prompt, generated with no mark

NOT WATERMARKED

PROMPTWrite a short story about a lighthouse keeper who receives an unexpected letter.

In a small town nestled among the trees, lived a lighthouse keeper named William. He spent his days exploring the town's old houses and gardens, looking for clues to his secret. One day, William received an unexpected letter from the local newspaper. It was written by a reporter who claimed to be a spy, looking into the town's secrets. William was shocked and eager to find out what had just happened. The reporter's story was so intriguing, he decided to write a short story about it. William woke up in the morning with a feeling of unease. He was so surprised that he had forgotten he had received the letter. He had always wanted to know what had been happening, but he had never seen such a letter. He approached the reporter's letter and read it to his heart. He had read his email for hours, and it had been written with a mix of letters and typos. William realized that the letter was coming from a man who seemed to be a spy. He approached him and asked him what he could tell him. William listened carefully to his friend and asked him to tell him what had just happened. He replied that he had been missing him for a long time, and he had been looking for clues to his secret. William knew that he had to act fast, and he was going to act fast. He started to read his report and try to understand what had just happened. Finally, William reached his friend's home and gave him a report. He had been missing him for a very long time, and he had been looking for clues to his secret. William was glad that he had found out what had been happening, and he knew that he would always be looking for clues to his secret.

SCORED

325 scoring floor 52

SKIPPED

GREEN HITS

182 / 325

Z-SCORE

2.16

56% green50% chance

182 of 325 scored positions were green (56.0%; chance is 50%). Corrected p = 1.22e-01.

Short reply · under the floor

Marked, but too short for a statistical test

INSUFFICIENT LENGTH

PROMPTSay hello to a new neighbor in one friendly sentence.

Hello, how's your neighbor?

SCORED

3 scoring floor 52

SKIPPED

GREEN HITS

2 / 3

Z-SCORE

n/a

66.7% green50% chance

2 of 3 scored positions were green. The floor is 52 scored positions, so no statistical test ran.