C and the Crux

The Crux and the C verdict move together in both engines. exhausted → Crux always comes with a fire (Poetix 25/25 (100%), Zexel 29/29 (100%)); REFRAME usually does (Poetix 43/49 (88%), Zexel 15/22 (68%)); MODIFY often (Poetix 52/72 (72%), Zexel 5/11 (45%)); AFFIRM rarely.

The sharpest line is AFFIRM. Zexel never fires the Crux when it affirms the frame (0/48 (0%)). Poetix sometimes does (12/54 (22%)): it accepts the frame and still reaches for a new idea. That may be what a poem does that an argument can’t, and Poetix’s looser gate allows it; whether the new idea adds anything is a question for the record, not the footer.

Part of this link is built into the definitions: “exhausted → Crux” names a hand-off to the Crux, and AFFIRM says the frame holds. The table shows how strictly each engine keeps to that, not an independent discovery.

When Zexel fires, C is most often exhausted → Crux (29/50 (58%)): Zexel tends to reach the Crux by declaring the frame used up. When Poetix fires, C is most often MODIFY (52/137 (38%)) or REFRAME (43/137 (31%)): Poetix reaches the Crux while still working on the frame.

When neither fires, both mostly affirm (Poetix 42/70 (60%), Zexel 48/61 (79%)).

The engines reframe at about the same rate (a gap of 4 points across all modes, 3 in P). They differ in what they do otherwise: Poetix modifies the frame far more (25 points; 25 in P), and Zexel affirms it more (17 points; 10 in P) or hands it to the Crux (exhausted → Crux, 14 points; 19 in P).

In words: both engines rebuild a frame about equally often. Short of rebuilding it, Poetix adjusts it; Zexel either accepts it or declares it used up.

Across all modes, the close REFRAME rates hold in every single-question group but the binary one: the two engines stay within 5 points of each other on Q1, Q22 and the non-binary pair. In P alone the groups are small enough to swing further (Q22 kitchen appliance +12, Non-binary (Q23 + Q33) -14 points), so the P column shouldn’t be read question by question.

The exception is the binary question (Q5, “Is AA a religion or a technology?”). There Poetix reframes 27/45 (60%) against Zexel’s 9/24 (38%); in P, 11/16 (69%) against 6/17 (35%). Given an either/or, Poetix mostly rebuilds the frame; Zexel splits between rebuilding it and handing it to the Crux (exhausted → Crux 10/24 (42%)).

Poetix modifies more than Zexel in every group; the widest gap is Non-binary (Q23 + Q33) (+34 points, +39 in P). Zexel affirms more in most groups; the widest gap is Q1 metal (-35 points), where Zexel affirms 20/25 (80%) of footers. The kitchen appliance is the one group where the two engines affirm about equally (-3 points; +0 in P).