Hi HN, I'm building witnessd (v0.1.9) to solve the "Is this AI?" problem by looking at the process, not the result.
The Tech:
- Hardware Binding: Keys are bound to TPM/TEE; signals are signed in a secure environment.
- Jitter Seals: We capture the statistical entropy of the writing process to differentiate human "jitter" from synthetic input.
- Time Hardening: We use sequential VDFs (Verifiable Delay Functions) so that forging a 2-hour session requires 2 hours of computation. No parallelization.
I'm a novelist and Marine veteran, and I'm currently moving this through the IETF (draft-condrey-rats-pop). I'd love your feedback on the schema and the VDF implementation.
- why patent it when you give permissions via Apache licence anyway?
- you can't use an external keyboard with this, right?
- can you remove content from the middle of the record without breaking signatures? For example if someone makes an online purchase during a break in the writing and forget to turn off the system, is their credit card number forever bound to that text?
- Patent + Apache 2.0: I'm pursuing standardization through the IETF RATS working group and clear IP status with a royalty-free grant removes one of the biggest adoption barriers for standards work.
- External keyboards: Timing is captured at the OS input event so USB, built-in, whatever. Bluetooth adds variable latency that degrades the behavioral signal, so evidence metadata is adjusted accordingly.
- Credit card numbers: Content is never captured in the first place. Keystroke timing is captured without recording which keys were pressed. There's nothing to remove because there's nothing there. The privacy model is structural, not policy.
Witnessd also handles multi-device workflows, paste events, and parallel document sessions. Evidence generation is fully offline with no network dependency, no trusted third party, and no infrastructure to trust beyond the device itself.
Hi HN, I'm building witnessd (v0.1.9) to solve the "Is this AI?" problem by looking at the process, not the result.
The Tech:
Try it out: Or verify any .war certificate at https://writerslogic.com/verifyI'm a novelist and Marine veteran, and I'm currently moving this through the IETF (draft-condrey-rats-pop). I'd love your feedback on the schema and the VDF implementation.
Some questions:
- why patent it when you give permissions via Apache licence anyway?
- you can't use an external keyboard with this, right?
- can you remove content from the middle of the record without breaking signatures? For example if someone makes an online purchase during a break in the writing and forget to turn off the system, is their credit card number forever bound to that text?
- Patent + Apache 2.0: I'm pursuing standardization through the IETF RATS working group and clear IP status with a royalty-free grant removes one of the biggest adoption barriers for standards work.
- External keyboards: Timing is captured at the OS input event so USB, built-in, whatever. Bluetooth adds variable latency that degrades the behavioral signal, so evidence metadata is adjusted accordingly.
- Credit card numbers: Content is never captured in the first place. Keystroke timing is captured without recording which keys were pressed. There's nothing to remove because there's nothing there. The privacy model is structural, not policy.
Witnessd also handles multi-device workflows, paste events, and parallel document sessions. Evidence generation is fully offline with no network dependency, no trusted third party, and no infrastructure to trust beyond the device itself.