Add com.writerslogic.text-watermark-semantic.1 (id 46) - #64
Merged
Conversation
Post-hoc semantic embedding watermark using key-conditioned dual-embedding projection. Survives paraphrase and translation without generation-time access. Replaces the closed draft PR #58 (semantic fingerprint), which captured meaning similarity rather than origin-specific signal.
6 tasks
domguinard
requested changes
Aug 4, 2026
Collaborator
|
This entry will be reviewed and approved during the Watermarking task force meeting on August 5th. |
Co-authored-by: Dominique Guinard <dom@guinard.org>
domguinard
approved these changes
Aug 5, 2026
jcollomosse
self-requested a review
August 5, 2026 12:18
jcollomosse
approved these changes
Aug 5, 2026
Collaborator
|
Approved by the WM TF - pending resolution of the conflicts. |
deblanco
added a commit
to deblanco/softbinding-algorithm-list
that referenced
this pull request
Aug 5, 2026
Registers the Monolith SHA-256 content fingerprint (identifier 47, type fingerprint) for C2PA manifest recovery. This alg is already emitted in production manifests, so registering it closes a gap rather than reserving a name speculatively. Resolution is available without credentials via the Monolith Soft Binding Resolution API and, independently, via the GeneralizedClaimRegistry contract on Stability Protocol. Claims 47 rather than 46 because open PRs c2pa-org#63 and c2pa-org#64 both claim 46. Claude-Session: https://claude.ai/code/session_01JMfrHK3uaw1qpNeHh9TP6y
deblanco
added a commit
to deblanco/softbinding-algorithm-list
that referenced
this pull request
Aug 5, 2026
Registers the Monolith SHA-256 content fingerprint (identifier 48, type fingerprint) for C2PA manifest recovery. This alg is already emitted in production manifests, so registering it closes a gap rather than reserving a name speculatively. Resolution is available without credentials via the Monolith Soft Binding Resolution API and, independently, via the GeneralizedClaimRegistry contract on Stability Protocol. Claims 48 rather than 47: 46 is taken by com.doverunner.watermark.1 (PR c2pa-org#63, merged) and open PR c2pa-org#64 will take 47. Claude-Session: https://claude.ai/code/session_01PSpv9ycZNEaZV5mWWGSSj1
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds com.writerslogic.text-watermark-semantic.1 (identifier 46), a watermark-type soft binding for text content.
The algorithm is a post-hoc semantic embedding watermark. Semantic-preserving lexical substitutions are selected to align the document's contextual sentence-level and token-level embeddings with a secret-key-derived watermark direction in the shared embedding space. Detection requires only the secret key and the candidate text; a one-sided hypothesis test on the dot product of the extracted dual embedding against the watermark direction determines presence or absence at a calibrated false-positive rate. No generation-time access is required, so the algorithm applies to both AI-generated and human-authored content. The secret key is never embedded in the text and is not disclosed during verification.
Unlike com.writerslogic.zwc-watermark.1 and .2 (ids 29, 42), which anchor the signal in zero-width characters and fail when those characters are stripped, this entry anchors the signal in the semantic embedding space. It is robust to paraphrase, translation, format conversion, re-encoding, and zero-width-character injection; not robust to complete rewriting that substantially changes meaning.