ARCHI AGTL — LEARNING DASHBOARD

UPDATED 2026-08-16 with a refined root cause. Vocal DAWgI captions/timed_lyrics render with vocal lines wildly out of order. TWO confirmed contributing causes: (1) the 79 delivery_anchors were tapped across 6 sessions spanning Aug 12-16 against shifting BPM/render state (see original finding below), and (2) NEWLY CONFIRMED: the declared lyrics field merges multiple real sung phrases onto single lines -- e.g. 'What slip from mi tongue... cravin dese licks,' was declared as ONE line but is actually sung as TWO separate lines. Artist provided the true line-by-line breaks: 115 real vocal lines vs the previously-declared 105 lines, a 10-line undercount concentrated in the repeated chorus (~2 extra real lines per chorus x 5 occurrences). This is exactly the kind of accumulating off-by-N mismatch that would desync line_index against real audio increasingly as the song progresses -- consistent with early anchors (lines 0-4) looking roughly sane while mid/late-song anchors are badly scrambled. FIX PATH UPDATED: before any re-tap, the declared lyrics/timed_lyrics must first be corrected to the real 115-line breakdown (section titles and inline performance-direction tags like [Chorus], [dark, intimate whisper] stripped out entirely -- they were likely being counted as phantom lines too, compounding the drift). Only after the line count/breaks are corrected should a fresh continuous Vocal DAWgI tap-through be done against the canonical primary render. --- ORIGINAL FINDING: the 79 delivery_anchors on this DeclaredIntent were tapped across 6 separate sessions spanning Aug 12-16 (56 anchors from Aug 12 ~22:05-22:15, a few more Aug 13 01:43/05:29-05:31, then small 1-3 line patches today at 11:37, 12:16-12:24, 13:12, 13:27) rather than one continuous pass, each patching only a handful of line_index slots against whatever BPM/render state existed at that moment (BPM edited 5x today: 68->80->92->94). SEPARATE compounding issue: a remix render (track_url ...0eba3664...) was analyzed using these same anchors tapped against the PRIMARY render's audio, not the remix's -- remix measures 82 BPM vs declared 94.