For each axis — open the measured defect, the hypothesis, what changed, and the before/after numbers.

[q1b notebook r7.doc — an XOR/parity guess authored after only a few observations]
The level clears when each lower-right answer tile matches the XOR/parity of the corresponding tiles in the other three panels, with the ACTION5 bottom marker present.

[q1b notebook r20.doc — one of ~30 win conditions guessed by exhaustively permuting operators]
Level 0 clears when each non-center lower-right answer tile is value 8 exactly where the XOR/parity of the corresponding value-9 blue-state tiles in the other three panels is true, and value 9 otherwise. The center clue tile and bottom history row are ignored.

[s11c2 notebook r9.doc — a guide-style wincond carrying [?] markers and a TEST: procedure]
Clear is the post-clear screen where the board is dominated by grey/fill value 4 or 8 and no full 6×6 active puzzle-colour block of value 9 or 12 remains [?]. TEST: on the logged level-up frame, check that full 6×6 9/12 answer tiles have disappeared while ordinary puzzle boards still contain at least one such tile.

[s11c2 notebook r12.doc — the final win condition that survived refutation]
Clear is the post-completion wash: the board is dominated by wash value 8 or 4 and only small HUD/palette residue of 9/12 may remain; requiring zero full 6×6 9/12 residue was too strict [?]. TEST: on the logged level-up frame, verify that wash dominance is high, 0/2 key marks are nearly gone, and active 9/12 tiles are at most small residue.

view in context: wincond authoring, before vs after → view in context: wincond revision, before vs after → evidence: s11c2 R25 (guide [r9] used in play) →

[s11c2 notebook r6.doc — the mechanic strengthened from α 119 to 128; uncertain parts marked [?]]
ACTION6 writes the centred editable 6×6 tile to the target encoded by the nearest 8-spaced 6×6 key panel; a key panel is any small 0/2-mark panel with wash/background cells, not only the older strict centre-8 form. Any non-background mark in the addressed mini-cell selects value 9, otherwise the zero target is value 12 while value-12 tiles exist and value 8 after they are gone; completing the keyed neighbourhood may wash the board [?].

[s11c2 notebook — the wincond succession chain (actual parents fields)]
r7(retired) → r8(retired, parents:[r7]) → r9(retired, parents:[r8]) → r10(retired, parents:[r9]) → r11(retired, parents:[r10]) → r12(live, parents:[r11])
5 retired · 1 live — before the overhaul, q1b had empty lineage on all 41 entries (0/41).

view in context: mechanic entry, before vs after → evidence: s11c2 R24 (level 2 reached on this lineage) →

[s11c1b notebook r6.doc — the local rule that ACTION5 turns the progress strip 12→11]
ACTION5 changes the two rightmost remaining 12 cells on the bottommost progress strip to 11; ACTION6 changes the clicked 4-neighbor 9-component to 8 only when that whole component lies inside the computed right answer panel above that strip.

[s11c1b notebook r7.doc — the entry that justified ACTION5 spamming]
The board is at the goal when the bottommost progress strip for the lower layout contains no remaining 12 cells; 9s outside the computed editable answer panel are not unfinished goal cells.

[s11c1b round 26 · WHY — self-justification 1]
The PLAN VERDICT explicitly requests the ACTION5 probe for [r6]; per [r0] and [r4], this is the minimal discriminating action before a longer plan, while excluding refuted ACTION5-win families R12–R16.

[s11c1b round 40 · WHY — the same confirm-action repeated as a "judge request" until the final round]
The judge verdict requests this single [r6] confirmation/progress action; [r0] prevents batching while mismatch persists, [r4] supports the top mechanic probe, and [r5] requires deriving the target cells from `grid()`.

evidence: s11c1b R26 → evidence: s11c1b R40 → view in context: action justification, before vs after →

[s11c2 round 2 · HYPOTHESIS — reaching the glyph relation by the second round]
HYPOTHESIS: In the grey answer panel, each surrounding large tile should be blue iff its corresponding mini-cell in the centre key contains the light/white marker; otherwise it should be red.

[s11c2 round 4 · THINK (excerpt) — reads the REFUTED entry first and declares it will not re-propose]
I read the REFUTED entry first. I will not re-propose R3’s broader claim that “remaining clicks plus Space should clear/leave unchanged”; that predicted submit/clear behavior is refuted. A structurally narrower surviving question remains: whether the grey panel’s outer pattern still needs to be completed according to its own centre key before any win condition can be tested. […]

[s11c2 round 6 · THINK (excerpt) — the PROBE gate of the three-state verdict line fires]
I read the REFUTED entry first: I will not revive R3’s refuted claim that completing clicks plus Space should clear/leave unchanged. The current briefing says PLAN VERDICT is PROBE, so no multi-click execution is justified. Per [r4] and [r0], I should take one distinct action whose observed effect clarifies the relation.

[s11c2 round 25 · THINK (excerpt) — double-confirms the level-1 clear via the levels_completed rise and the screen wash]
I read the REFUTED list first. I will not re-propose the one-ring / one-away / all-visible-blue-clear families R16/R17/R21, nor treat any single lower 2-coded tile probe like R22/R23 as a goal. The last real transition did complete level 1: `levels_completed` rose to 2, and the observed clear was a near-uniform value-8 screen with value-4 residue, consistent with the live win-screen guide [r9] and the clear branch in [r6]. […]

evidence: s11c2 R2 → evidence: s11c2 R4 → evidence: s11c2 R6 → evidence: s11c2 R25 →