Cortex FMIE Integration Handoff

Version: v0.1 Status: Ready Owners: Cortex / Foundation Python Last Updated: 2026-03-26 Change Type: additive Migration Required: yes Depends On: foundation-v2-fmie-integration-audit-2026-03-26.md, Gateway spec, Media Passport spec, Partner-Fit spec Supersedes: none Superseded By: none

Purpose

This handoff gives Cortex the exact Python-owned FMIE integration work required to bring Foundation-V2 in line with current FMIE 0.42.7.

Per house rules: - Codex session owns C++ / FMIE-native work - Cortex owns Python module and service changes - Triton Front-end owns Swift work

P0 Work

1. Replace brittle FMIE library pinning with shared resolution

Target files: - services/haystack-pipeline/src/haystack_pipeline/ingest.py - services/scout-fleet/src/scout_fleet/fmie_access.py - any other direct libfmie_c.so.0.42.2 references

Required outcome: - resolve FMIE_LIB_PATH from env first - default to unversioned libfmie_c.so or a shared resolver - stop hardcoding 0.42.2

2. Make Cognition Gateway use full commitment hashes

Target file: - services/cognition-gateway/src/cognition_gateway/receipt.py

Required outcome: - use full 64-char SHA-256 digests for request and response commitments - keep receipts append-only - update any tests and readers that assumed 16-char truncation

3. Anchor real FMIE passport proof material to Midnight

Target files: - services/cognition-gateway/src/cognition_gateway/passport.py - services/cognition-gateway/src/cognition_gateway/app.py - services/cognition-gateway/src/cognition_gateway/midnight.py

Required outcome: - when an FMIE passport exists, anchor using: - artifact_root_sha256 - midnight_anchor_payload_sha256 - transform_manifest_sha256 - source_uri_sha256 - stop anchoring only synthetic session hashes where real FMIE proof material exists

4. Universalize the Gateway path

Target files: - packages/answerer-py/src/answerer/__init__.py - services/haystack-pipeline/src/haystack_pipeline/ingest.py - any remaining direct Anthropic call sites for internal content

Required outcome: - all sensitive outbound model calls go through Cognition Gateway - no direct summary-generation or synthesis path should bypass receipts, policy, and surrogate logic

5. Consume FMIE partner-fit helper artifacts

Target files: - services/scout-fleet/src/scout_fleet/runtime.py - services/haystack-pipeline/src/haystack_pipeline/ingest.py

Use these FMIE artifacts directly: - summary.meeting_need_vectors.json - summary.partner_fit_inputs.json - summary.passport.json.partner_fit

Required outcome: - Scout scoring and ingest enrichment consume normalized FMIE helper contracts - stop relying only on raw meeting_signals, tables, charts, and evidence

P1 Work

6. Wire document region intelligence

Target files: - services/haystack-pipeline/src/haystack_pipeline/document_regions.py - services/haystack-pipeline/src/haystack_pipeline/ingest.py

Required outcome: - page / region / table / chart structures contribute to payload metadata and downstream retrieval

7. Promote FMIE runtime and health into Cortex observability

Target integrations: - fmie_get_runtime_status_json() - fmie_get_health_status_json() - /tmp/FMIE.log breadcrumb ingestion

Required outcome: - Cortex can see FMIE readiness, model/cache state, contract discovery, and crash breadcrumbs

8. Use FMIE identity matcher for cross-asset continuity

Target file: - services/haystack-pipeline/src/haystack_pipeline/entity_matcher.py

Required outcome: - consume face and voice vectors already emitted by FMIE - use matcher-index queries for cross-asset continuity instead of leaving them dormant

P2 Work

9. Fix native-string lifetime in passport validation

Target file: - services/cognition-gateway/src/cognition_gateway/passport.py

Required outcome: - stop treating native FMIE result strings as unmanaged c_char_p - add proper fmie_free_string handling if needed

10. Consolidate all FMIE access into one shared Foundation client

Current duplication exists across: - ingest - scout wrapper - passport validation - clip extraction - matcher loading

Goal: - one shared Python package for FMIE resolution and helper artifact access

Cortex Success Criteria

This handoff is complete when: - no live Foundation path depends on libfmie_c.so.0.42.2 - no sensitive frontier-model call bypasses the Gateway - real FMIE passport proof material is what Midnight anchors - Scout and Haystack consume partner-fit helper artifacts directly - Cortex can observe FMIE runtime and health as a first-class system