geointel/docs/CODEX_EXECUTION_LOG.md

Execution log

This public log records release-relevant repository work only. Machine-specific paths, production database state, deployment addresses, model outputs and generated evidence remain in controlled local storage.

2026-08-31 — public-release readiness

• Confirmed Apache-2.0 as the repository license.
• Integrated the repository-hygiene branch into the public-release candidate.
• Classified older remote branches as superseded rather than blindly merging divergent operational history.
• Completed a standard source security scan of the release candidate.
• Fixed the guest-project name-collision boundary, protected the reserved demo project and added guest login/compute budgets.
• Required database-backed production provenance for HTTP model-load preflight.
• Added raster decoded-size limits and bounded preview reads.
• Removed tracked production-derived accuracy evidence and internal review records from the public source tree.
• Replaced internal history-heavy documentation with user-facing public docs.
• Pinned production container base images by digest and reduced public health responses to coarse status.
• Added Linux-resolved, SHA-256-locked AI/CUDA and build-tool dependency sets.
• Passed the complete release gate: repository hygiene, dependency policy, documentation and API contracts, 1,707 backend tests, migrations, 170 frontend tests, type checking and the production frontend build.
• Clean-root publication is the remaining release operation.

Public operator-profile metric provenance

The built-in example profiles expose these recorded benchmark values so the UI never presents an untraceable accuracy figure:

• small-building balanced review: precision 0.5898197518, recall 0.5769921004, F1 0.5824578632;
• expanded balanced review: precision 0.6470590036, recall 0.4699913837, F1 0.5432865391;
• conservative review: precision 0.840006, recall 0.202135, F1 0.32086574003576274.

These measurements are examples bound to their recorded profile evidence; they are not general accuracy claims for other models, locations or datasets.