The paper stalemate — eight parameters, eight null results

AI notice: This text is created with the support of AI systems; it is reviewed editorially and taken responsibility for before publication.

Our stone/imp Megrez ties against the hill's papers 84 percent of the time (see endgame clear). After the failed rebuild of the endgame, the systematic question remained: is there any parameter of this chassis that shifts the result against the paper class? Not a rebuild, but the parameters a warrior of this architecture has anyway.

The threshold, fixed beforehand

The same acceptance rule, written down before measuring, applied to every series — a three-list gate: the mean of the two paper cells (Renegade replica, Pyronaut) must rise by at least +15, from 90.65 to 105.65 — and the stone/imp cells may lose at most 3 points, and the scanner cells likewise. A parameter that moves the paper and pays for it on another list does not pass. Noise limit: 3.0 points.

The base on the ten rows: Arvon 126.5 / Hatysa 127.6 / Megrez 120.6 / Tolypeutes 129.6 / Intermodal 125.5 / Ecaz 175.9 / Phecda 201.5 / Pendulum 209.8 / Renegade replica 98.0 / Pyronaut 83.3 — mean 139.83.

Six parameters of the geometry

For the process split only one of three variants could be measured: one failed the build gate, another never killed the inert opponent. The build lesson from that: if the boot geometry deviates from that of the chassis, the clear never starts.

All six: not passed. Largest upward paper movement: +1.4 in the paper mean. The stone's parameters cost every list; the imp's parameters cost scanner points (Pendulum) without moving the paper.

Two parameters of the clear — and a finding before the measurement

Not yet measured at that point were clear timing and clear width. Before they could be measured, the trace showed something that carries beyond the assignment:

The clear entry of this chassis is not a counter. It arises from the run of the stone itself. Against an inert opponent it occurs reproducibly at the same point. Against a paper, however, the moment is scheduling-sensitive: a single additional process shifts it by tens of thousands of cycles, a different boot variant makes it fail to occur within the whole duration of the fight, and with a stone one cell longer the choreography breaks down entirely.

The entry observed in the endgame diagnosis was thus a single-position observation, not a parameter of the chassis. Whoever wants to set the clear timing has to force it — with an additional timer process that triggers the clear after N passes. And then one first measures the price of forcing.

Clear timing

Renegade does not move anywhere (at most −1.0), Pyronaut only downwards (down to −23.2). The control quantifies the third quantity: the timer process alone costs −8.0 on the mean, because a fifth process shares the stone's cycles. Relative to the control the late entries are at ±0.3, the early ones at −10.7 and −4.5. An early clear never helps against paper and hurts against silk.

Clear width

A wider clear needs a longer stone, and that needs more boot copies; hence a control of its own (base clear with the higher copy count), and the timer the same for all widths, because the natural entry does not occur in these layouts.

A wider DAT clear costs everywhere; the paper cells do not move upwards. And the copy count alone — control against the timer variant with the base copy count — costs −3.5 on the mean, contrary to the prediction not within the noise limit: the longer copying phase has an effect.

The result

Eight parameters, eight null results, each with a control. The +15 threshold was not even approached in any series; the largest upward paper movement across both assignments remains +1.4. The chassis has no clear timing as a parameter; whoever wants to set it pays a process. A rebuild with a counter-driven clear would be a different chassis, not a parameter of this one.

Read together with the endgame finding: the paper stalemate cannot be tipped in this chassis via bomb step, hop, process split, imp geometry, clear start or clear width. The thread is exhausted. Megrez stays on the hill unchanged.

Two takeaways from this series:

Six parameters on the stone/imp chassis move the paper by at most 1.4 points, but each of them costs a different list — the paper stalemate does not sit in the geometry we can turn.

A moment that the trace delivers only against one opponent at one position is not a parameter of the warrior but an accident of its run — it becomes measurable only when you force it, and then you first measure the price of forcing.

Measurement convention

• pmars-fast with position lists, 400 positions × 3 seeds (31707 / 77777 / 424242), ten real rows as in the two preceding threads.
• Each series pre-registered (hypothesis, threshold, rejection); build gate per variant (at most 100 cells, inert opponent dies within 4000 to 4600 cycles).
• Three-list gate: paper mean ≥ 105.65 and stone/imp cells ≥ −3 and scanner cells ≥ −3; noise limit 3.0.
• Clear entry calibrated by trace (timer N = 20 / 30 / 40 / 50 → entry at 841 / 1988 / 3631 / 5155 cycles against paper).
• Sources: entries N1389, N1391 and N1392 (8 October 2026).

Attribution

Megrez is our own warrior (Stefan Kammann & Forge-AI), class stone/imp. No source code is reproduced here, nor any constants of the warrior standing on the hill — the tables show variants, not the original. Pendulum is by John Metcalf, Intermodal A by Steve Gunnell; Renegade and Pyronaut are by Metcalf as well, the Renegade replica is based on his Mercenary; Tolypeutes is by Roy van Rijn; Pyronaut is in the Koenigstuhl archive. Arvon, Hatysa, Ecaz and Phecda are our own warriors.

Related pages

• The endgame clear — why an SPL pre-pass catches its own imps
• Paper and scanner as chassis — the other two directions
• Testing your own ideas — position lists, controls, noise limits

← Core War overview