Sylpheed/docs/re/f6-unit11-pteff03a-IS-drawn.md

F6 unit 11 — 🔴 pteff03a is drawn. Units 5 and 6 are refuted, by my own parser.

Question: does the real title draw one sweeping strip or two?

What the human looks at: watch the title. Two glowing strips cross it in opposite directions. Pass = two. Fail = one.

What this does NOT cover: their alpha curves, the snap target.

Instrument: ⟨capture⟩ — the same two logs units 5 and 6 used. No new capture. The logs always said two; my reader said one.

The bug

The sweeps are batched into a single additive draw — indices=8, eight vertices, two quads:

10482 ... blend=0x01010101 ... tex[... h=8154424FFC48FE61]
  v: [0.54,1.58 BB] [1.08,1.02 BB] [0.24,-1.57 BB] [-0.30,-1.02 BB]   <- quad A
     [-1.13,1.02 B0] [-0.69,1.81 B0] [0.90,-1.02 B0] [0.46,-1.81 B0]  <- quad B

My parser took the first v: match per draw line and discarded the rest. So every analysis I ran saw quad A and never quad B. pteff03a was in the log the whole time.

✅ Both strips, measured

	mean length	starts at	travels
quad A	1151 px	x = −1.66	+x
quad B	1497 px	x = +1.98	−x

Three independent agreements with the declaration:

• Directions are opposite and correct — pteff03 declares −639→1521 (+x), pteff03a declares 1721→−839 (−x).
• Size ratio 1497/1151 = 1.301 against the declared sy ratio 800/600 = 1.333 — 2.4%.
• Quad A's 1151 px matches the 1134 the older GPU sessions recorded for the pteff03 strip.

Both hold in f6 and f6b.

🔴 What this retires

• f6-unit5-pteff03a-never-drawn.md — refuted. Its "quads in pteff03a's expected height band: NONE" counted first-vertices only.
• f6-unit6-pteff03a-not-submitted-at-all.md — refuted. "Absent, not collapsed" was neither; it was unread.
• The three candidate mechanisms killed while explaining the absence were explaining a non-fact. Under R1 they re-open: the instrument that killed them has changed.
• Unit 5 recorded that the port's renderer draws pteff03a and treated that as the port being wrong. The port was right and I was the one to correct.

The lesson, and it is the third time

The corpus already carries two entries of this exact shape — palogo_anima "never appears" against the 8-vertex cap, and a dump that "captured two [of six quads], with a well-formed line and no ellipsis". A truncating reader produces a clean, complete-looking negative. Nothing in unit 5 or 6 looked wrong.

What would have caught it, and what I did only on the third pass: read one raw draw line in full before trusting a count derived from it. indices=8 was printed on every one of those lines and says two quads outright.

And the corpus already held the general fact. REFUTED.md L170 records the swoosh being "submitted as two rotated parallelograms" in one draw. That one draw line can carry more than one quad was known and written down; my reader simply did not implement it.

⚠️ Claims that depended on the first-vertex reader and are not affected: the alpha decomposition in unit 10 (quad A read consistently on both sides), the press/no-press comparisons in F5 (same reader both sides), and the pulse ratio (single-quad draws). Those compare like with like. The absence claims did not — they compared a count against zero.

Not settled

• Why pteff03a's alpha differs from pteff03's at the same instant (0xB0 vs 0xBB above) — the declared curves differ, but I have not fitted them.
• The snap target, still bounded to [100,238].