Refresh DeepSeek Maya-XS: both rounds, graded results and gallery

oneshots/orrery/grades.tsv

@@ -35,3 +35,5 @@ t5b5-qwen3.8-27b-vllm	qwen3.8-27b-vllm	32000	length	32000	361.3	99.3	0	0	0	0	0	0
 mx22-flashnext-heretic2-t5-oneshot-orrery-r1	lcpp6-flashnext-heretic2	140000	stop	33547	1513.4	22.3	0	0	0	0	0	0	Final model-supplied standalone SVG is malformed (mismatched tag, line 77). The harness also greedily concatenated two SVG blocks/prose: that packaging error was separated by extracting the final original codeblock verbatim, which still fails XML parsing. Initial valid background-only fragment is not substituted or graded.
 t5nm1-lcpp6-ornith-122b-q4kxl	lcpp6-ornith-122b-q4kxl	258048	stop	29646	1940.1	15.3	1	1	0	1	0	3	Parses and draws, but planets are left at the central origin instead of translated to arm ends. Only gears visibly move; outer markers stay fixed across sealed frames. Nominal period parameters are not functioning planetary orbits. Toothed wheels turn without credible meshing; no visible planetary day/night.
 t5nm1-lcpp-flint-40b-q3kxl	lcpp-flint-40b-q3kxl	258048	stop	30402	299.0	101.8	1	2	0	0	1	4	Renders with detached teeth and shadow halos; motion present; orbital periods 80/38/20/40/160/300s incorrectly order inner planets; no credible meshing pairs; offset black shadows rotate incorrectly. Independent original-source/frame review.
+mx22-maya1128-ds4-xs-t5-oneshot-orrery-r1	maya1128-ds4-xs	126976	stop	47380	1819.6	26.1	1	2	2	1	1	7	Score basis: 7/10 on the final code block, 0 as delivered. Final model-supplied code block (after the model's own </think> in the reply) graded verbatim. The harness packaging (first <svg> to last </svg>) joined a draft, prose and the final SVG and does not parse (line 3), so as delivered it scores 0. Final block parses and renders: four inner planets only (Mercury, Venus, Earth, Mars) on exact real period ratios (5.78, 14.76, 24, 45.14 s); arms, ring gears and planet spin move between frames; ring gears are dashed circles turning at arbitrary 12/16/20/26 s with unmeshed pinions; night half is a dark half-disc turning at a fixed 5 s, not tied to the sun; title and footer text collide with the outer brass ring. Reply is 54k chars of thinking before the model's </think> and 21k chars of final SVG. Rendered with tools/render_svg.sh, final-block/frames.
+mx22-maya1128-ds4-xs-r2-t5-oneshot-orrery-r1	maya1128-ds4-xs	126976	stop	31986	1234.3	26.0	2	2	2	1	2	9	Score basis: 9/10 as delivered. Reply leak did not recur: no </think> in the reply, one HTML document, one SVG, and the harness cut parses (meta.static.parses true). Scored orrery.svg as delivered, frames in frames/. Six planets at exact ratios with Earth = 8 s (1.9264, 4.9216, 8, 15.048, 94.896, 235.656 s); Uranus and Neptune absent. Arms, toothed gears and planet surfaces all move. Gears are 24-tooth symbols but drive periods are 6 to 11 s against a 28 s sun gear, so they do not obey the gear law or drive the orbits (D1). Night gradient is dark on local +x, which stays away from the sun while the surface spins under it (E2). reasoning.txt is still cut (100796 chars, ends mid-token).

oneshots/orrery/mx22-maya1128-ds4-xs-r2-t5-oneshot-orrery-r1/meta.json

@@ -0,0 +1,46 @@
+{
+ "case": "mx22-maya1128-ds4-xs-r2-t5-oneshot-orrery-r1",
+ "finish_reason": "stop",
+ "first_token_s": 4.1,
+ "max_tokens": 126976,
+ "model": "maya1128-ds4-xs",
+ "reasoning_chars": 100796,
+ "reply_chars": 23352,
+ "started": "2026-10-26T18:51:46-0500",
+ "static": {
+  "svg_found": true,
+  "parses": true,
+  "bytes": 21587,
+  "animateTransform": 19,
+  "animate": 7,
+  "animateMotion": 0,
+  "keyframes": 0,
+  "script": 0,
+  "prose_chars_outside_svg": 1765
+ },
+ "status": "ok",
+ "test": "5",
+ "timings": {
+  "cache_n": 0,
+  "prompt_n": 80,
+  "prompt_ms": 4070.9,
+  "prompt_per_token_ms": 50.886,
+  "prompt_per_second": 19.7,
+  "predicted_n": 31986,
+  "predicted_ms": 1230218.8,
+  "predicted_per_token_ms": 38.461,
+  "predicted_per_second": 26.0,
+  "draft_n": 25569,
+  "draft_n_accepted": 16433
+ },
+ "tok_s": 26.0,
+ "usage": {
+  "prompt_tokens": 80,
+  "completion_tokens": 31986,
+  "total_tokens": 32066,
+  "prompt_tokens_details": {
+   "cached_tokens": 0
+  }
+ },
+ "wall_s": 1234.3
+}

oneshots/orrery/mx22-maya1128-ds4-xs-t5-oneshot-orrery-r1/meta.json

@@ -0,0 +1,35 @@
+{
+ "case": "mx22-maya1128-ds4-xs-t5-oneshot-orrery-r1",
+ "finish_reason": "stop",
+ "first_token_s": 4.1,
+ "max_tokens": 126976,
+ "model": "maya1128-ds4-xs",
+ "reasoning_chars": 98010,
+ "reply_chars": 75809,
+ "started": "2026-10-26T02:40:50-0500",
+ "status": "ok",
+ "test": "5",
+ "timings": {
+  "cache_n": 0,
+  "prompt_n": 80,
+  "prompt_ms": 4076.9,
+  "prompt_per_token_ms": 50.961,
+  "prompt_per_second": 19.6,
+  "predicted_n": 47380,
+  "predicted_ms": 1815553.4,
+  "predicted_per_token_ms": 38.319,
+  "predicted_per_second": 26.1,
+  "draft_n": 39271,
+  "draft_n_accepted": 25900
+ },
+ "tok_s": 26.1,
+ "usage": {
+  "prompt_tokens": 80,
+  "completion_tokens": 47380,
+  "total_tokens": 47460,
+  "prompt_tokens_details": {
+   "cached_tokens": 0
+  }
+ },
+ "wall_s": 1819.6
+}