Skip to content

Add attribution confidence for SFT exports#179

Draft
vincentwuxi wants to merge 1 commit into
Netis:mainfrom
vincentwuxi:codex/attribution-identity-confidence
Draft

Add attribution confidence for SFT exports#179
vincentwuxi wants to merge 1 commit into
Netis:mainfrom
vincentwuxi:codex/attribution-identity-confidence

Conversation

@vincentwuxi

Copy link
Copy Markdown

Summary

  • add span attribution metadata: label, source, and confidence
  • derive high-confidence attribution from eBPF process metadata or configured trusted request headers
  • persist attribution metadata in DuckDB and ClickHouse and expose it through span/trace read models
  • add min_attribution_confidence to trajectory export so SFT batches can skip ambiguous-source turns

Boundaries

  • eBPF process attribution still wins when present
  • gateway/passive capture only gets high-confidence attribution when a configured HTTP header is visible
  • mTLS SAN is not parsed directly; gateways should promote it into a trusted header first
  • current confidence values emitted by Heron are high and ambiguous; medium is reserved for future correlation strategies

Validation

  • cargo check --workspace
  • cargo test -p h-llm attribution
  • cargo test -p h-api export
  • cargo test -p h-storage-duckdb
  • cargo test --workspace --no-run
  • cargo test -p h-storage-duckdb phase8_otel_rename_migrates_tables_columns_and_backfills_kind
  • git diff --check

@vaderyang

Copy link
Copy Markdown
Collaborator

Prefer a solution without database migration. This is not a prioritized feature, but it triggers database migration which is not preferable.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants