fix(deps): bump idna 3.11 → 3.18 in requirements-lock.txt

fix(deps): bump cryptography 46.0.7 → 47.0.0

fix(deps): bump transformers 4.53.1 → 4.57.6

fix(api): wrap detect_segments in try/except — closes CodeQL #24

fix(ensemble): apply feedback_manager weight multipliers in combine_signals()

fix(health): add platt_calibration to degraded_detectors list

fix(mcmc): replace biased 1e-10 lower bound in MH acceptance draw

fix(own_embedding): wrap centroid pickle.load in try/except

fix(ensemble): log XGBoost feature-name mismatches before predict_proba

fix(ensemble): apply feedback_manager weight multipliers in combine_signals()

fix(ensemble): filter top_reasons by confidence > 0

fix(ensemble): move Platt calibration to weighted-sum fallback path only

fix: sync requirements-lock.txt diffusers/huggingface_hub/safetensors versions

fix: bump safetensors 0.4.5 -> 0.8.0 for diffusers 0.38.0 compatibility (batch3 #10)

fix: bump huggingface_hub 0.30.0 -> 0.34.0 for diffusers 0.38.0 compatibility (batch3 #9)

fix: bump pytest-asyncio 0.23.8 -> 1.3.0 for pytest 9 compatibility (batch3 #8)

test: assert degraded_detectors always present in /health response (batch3 #6)

fix: CLIP placeholder init logs at ERROR not INFO; add placeholder_mode to return dict (batch3 #4)

feat: add degraded_detectors list to /health endpoint (batch3 #3)

fix: sync requirements-lock.txt to new CVE-fixed versions (batch3 #2)

fix: pin starlette/idna/Pygments CVEs; frontend HEIC, 30 signals, PDF thumbnail, error banner, 90s timeout, footer links (audit #18)

refactor: extract combine_signals() from detect() so SSE can reuse scoring without re-running detectors (audit #12)

fix: bump fastapi to 0.115.14 (pulls starlette 0.46.2 automatically, no pin conflict); fix ci.yml coverage.xml crash

fix: pin starlette/idna/Pygments CVEs; frontend HEIC, 30 signals, PDF thumbnail, error banner, 90s timeout, footer links (audit #18)

feat: add GPT-4o, Flux, Imagen3, Ideogram, Recraft generator profiles and heuristics (audit #14)

fix: SSE now runs all 30 signals exactly once; DIRE and OwnEmbedding now appear as individual streamed events (audit #13)

refactor: extract combine_signals() from detect() so SSE can reuse scoring without re-running detectors (audit #12)

fix: DIRE now encodes to VAE latent before UNet; decode after denoise; align noise timestep (audit #11)

perf: vectorize DCT radial spectrum loop with np.bincount (audit #8)

fix: preserve JPEG format across EXIF transpose; rate-limit /segment; fix CodeQL CWE-209 (audit #7)

fix: add PIL MAX_IMAGE_PIXELS=50M and explicit >10000px rejection (audit #6)

fix: add raw_value and expected_range to noise_map success dict (audit #5)

fix: heatmap_generator path now uses __file__-relative, not CWD-relative (audit #4)

fix: deepcopy cache entries to prevent confidence-cap mutating the cache (audit #3)

fix: cache Platt params at module level, add reload_params() (audit #2)

fix: add extension_valid key to validate_file() — was always True (audit #1)

fix: add missing Path import to main.py (NameError on startup)

fix: auto-download model files from HF Hub at startup

fix: enable LFS in CI checkout + safe pickle load

fix: admin key block uses PRODUCTION env var not DEBUG

fix: HTTPException UnboundLocalError and health check test assertions

fix: circular import — revert analyze.py limiter to standalone instance

docs: comprehensive README update + VERSION 8.4.0 -> 8.5.0

fix: CLIP confidence scaling, attribution path, natural model cache cap

fix: own_result missing from DIRE branch and inconclusive classification zone

fix: admin key security, deterministic seeds, health check, dual limiter

fix: EXIF orientation, HEIC allowed, quality confidence cap, extension validation

fix: DCT kurtosis threshold in wrong mathematical domain

fix: GPS rejects western hemisphere and JPEG Ghost PNG false positives

fix: python-multipart 0.0.30 -> 0.0.29 (0.0.30 does not exist on PyPI)