nx_video_codec.nx

nx_video_codec.nx -- NishiVideo v1: I-frame intra-coded video codec. Composes nx_dct8 + uniform quantisation + nx_video_frame. No P/B frames in v1; one frame stands alone, deterministic, audit-replayable. Pipeline (encode): pixel block (8x8 i64, range 0..255 luma) -> subtract 128 (centre on zero per JPEG convention) -> nx_dct8_forward_2d -> 64 frequency coefficients -> quantise: coeff_q = round(coeff / quant_step) -> pack as 64 int16 LE For a multi-block frame, concatenate per-block payloads. Then nx_vid_build_header + payload. Decode reverses: parse header -> for each block, unpack int16 coefficients -> dequantise (multiply by quant_step) -> IDCT -> add back 128 -> clamp to 0..255 -> output pixel block. Honest verdict vs Jitsi/Discord/Zoom (the codecs they use): WIN_axes: deterministic, audit-replayable (hash chain), patent-clean, refuses BAD_MAGIC frames cleanly (no silent corruption) LOSE_axes: rate-distortion at parity bitrate (no motion compensation in v1; no entropy coding yet; quantisation is uniform not perceptually-weighted) Named improvements queued: nx_motion_compensation, nx_zigzag_scan, nx_rle_coefficients, nx_range_code_coefficients, nx_chroma_subsample, nx_perceptual_quant_table. genealogy_id: nx_dct8_q10 + nx_video_frame_q10 + jpeg_iso_10918 + nx_voice_codec_v1_q10 lineage_id: nishi_video_codec_v1_q10