Merge branch 'main' into fix/update-gitignore

Potential fix for pull request finding 'CodeQL / Information exposure through an exception'

Potential fix for pull request finding 'CodeQL / Information exposure through an exception'

Potential fix for pull request finding 'CodeQL / Information exposure through an exception'

fix(accuracy): remove empty if block + use np.nan for missing XGBoost features

fix(accuracy): use np.nan for missing XGBoost features; remove double-boost

fix(accuracy): guard NaN in corrcoef and fix spectral slope math

fix(accuracy): disable CLIP signal instead of using random placeholder centroids

fix(accuracy): reduce metadata_forensics false positives

feat(rbac):API key management and role-based access control

feat(reports):forensic report export suite (PDF, JSON, CSV)

Merge pull request #104 from abinaze/feature/case-management

feat(cases): evidence case management system

feat(batch): batch investigation mode for multi-image forensic analysis

feat(robustness): adversarial robustness testing suite

fix(cleanup): remove unused imports from c2pa_verifier and platform_detector

feat(c2pa): C2PA content credentials verification

fix(platform): wire detect_platform into generate_forensic_report

fix(platform): guard polyfit on small images; use 128x128 in tests

fix(platform): fix 4 CI failures in platform detector

Merge branch 'main' into feature/platform-forensics

Potential fix for pull request finding 'CodeQL / Information exposure through an exception'

feat(platform): invariant tests + pyflakes cleanup

fix(attribution): read analyzer_version from settings, fix hardcoded version assertions

fix(attribution): wire generator_attribution into forensic report

feat(attribution):multi-class generator attribution classifier

feat(heatmap): Grad-CAM manipulation localization heatmap

fix(cleanup): final pre-release audit — emojis, unused vars, version, CI trigger, starlette CVE

fix(api,deploy): correct AttributeError, missing lifespan, Dockerfile drift, XGBoost double-boost

feat(ensemble): — XGBoost trained classifier replaces hardcoded signal weights

fix: audit pass — unused imports, None crash guard, emojis, accuracy claims, broken f-strings

fix: return None from load_model when no weights exist, return neutral 0.5 in detector

fix: update signal count to 26, fix perturbation non-determinism, fix DIRE import, patch starlette CVE

feat: complete — trained EfficientNet detector, centroids, integrated into ensemble

fix: skip hash check on EfficientNet weights, fallback to random init

feat: add OwnEmbeddingModel — EfficientNet-B0 architecture for AI detection

Fix: s[score] → s['score'] (NameError), fix cached_download → hf_hub_download

feat: Calibration, dedup fix, signal tests

feat:DCT Frequency Artifact Detection (25th signal)

feat:Deep Metadata Forensics signal (24th signal)

feat:ELA (Error Level Analysis) signal (23rd signal)

fix: frontend bugs, security deps, cleanup unused files

feat: PRNU camera fingerprint signal (22nd detection signal)

Fix: restore all missing imports in image_forensics.py (PIL, datetime, etc)

Fix: restore missing 'from typing import Dict, Any' in image_forensics.py

feat: add evidence_id to every forensic report, fix duplicate import

feat: add unique evidence_id to every forensic report

Fix: ensemble skips DIRE when unavailable, redistributes weight to statistical+CLIP

Fix: DIRE returns confidence 0.0 on failure so ensemble ignores it

Cleanup: Remove junk files from repo