File size: 64,946 Bytes
a5a2554 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 | <!DOCTYPE html>
<html lang="en">
<head>
<meta charset="utf-8"/>
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<title>Shipping 6t_argmax — a techniques log for Terminal-board</title>
<meta name="description" content="We are shipping 6t_argmax (sha 2ea02c13). Same-packet Terminal-board mean 5.80 / 79% useful versus the production pin 5.68 / 80%. A later n=1000 packet is 5.99 / 85% versus pin 5.80 / 85%."/>
<style>
:root {
--paper: #f5f0e7;
--ink: #17211f;
--muted: #58625f;
--faint: #d6cec1;
--sniff: #176b87;
--good: #32735f;
--good-bg: #e4f0ea;
--blue-bg: #e8eef5;
--warn-bg: #f4e8df;
--bad: #9b4034;
--sans: Inter, ui-sans-serif, system-ui, sans-serif;
--serif: Charter, "Iowan Old Style", Georgia, serif;
}
* { box-sizing: border-box; }
html { background: var(--paper); }
body {
margin: 0;
color: var(--ink);
background:
radial-gradient(circle at 12% 0%, rgba(23,107,135,.08), transparent 28rem),
var(--paper);
font-family: var(--serif);
font-size: 18px;
line-height: 1.58;
}
.paper { width: min(calc(100% - 2rem), 920px); margin: 0 auto; padding: 4rem 0 5rem; }
.eyebrow { margin: 0 0 1rem; color: var(--sniff); font: 750 .76rem/1 var(--sans); letter-spacing: .13em; text-transform: uppercase; }
h1 { margin: 0; font-size: clamp(2.4rem, 7vw, 4.2rem); line-height: .95; letter-spacing: -.04em; font-weight: 650; }
.subtitle { color: #34403d; font-size: 1.25rem; line-height: 1.35; margin: 1.2rem 0 1.4rem; }
.meta { color: var(--muted); font: 0.83rem/1.4 var(--sans); }
.abstract, .callout { margin: 2rem 0 2.5rem; padding: 1.3rem 1.4rem; border: 1px solid var(--faint); border-radius: 16px; background: rgba(255,255,255,.55); }
.abstract h2, .callout h3 { margin: 0 0 .4rem; font: 750 .72rem/1 var(--sans); letter-spacing: .1em; text-transform: uppercase; }
.abstract p, .callout p, .callout li { margin: 0; font-size: 1.02rem; }
.callout.good { background: var(--good-bg); border-color: #b9d2c6; }
.callout.warn { background: var(--warn-bg); border-color: #dec7b6; }
h2 { margin: 2.7rem 0 .7rem; font-size: 1.55rem; }
h3 { margin: 1.6rem 0 .4rem; font-size: 1.15rem; }
p, li { max-width: 72ch; }
table { width: 100%; border-collapse: collapse; font: 14px/1.4 var(--sans); margin: 1rem 0 1.4rem; }
th, td { border-bottom: 1px solid var(--faint); padding: .5rem .4rem; text-align: left; vertical-align: top; }
th { font-size: 11px; letter-spacing: .04em; text-transform: uppercase; color: var(--muted); }
td.n, th.n { text-align: right; font-variant-numeric: tabular-nums; white-space: nowrap; }
tr.pick td { background: var(--good-bg); }
tr.context td { background: var(--blue-bg); }
.wide { overflow-x: auto; margin: 1rem 0 1.4rem; }
.wide table { min-width: 820px; margin: 0; }
.note { color: var(--muted); font: .82rem/1.4 var(--sans); }
a { color: var(--sniff); }
code { font: 0.86em/1.3 ui-monospace, SFMono-Regular, Menlo, monospace; }
.status { font: 700 .68rem/1 var(--sans); letter-spacing: .06em; text-transform: uppercase; }
.status.keep { color: var(--good); }
.status.dead { color: var(--bad); }
.status.open { color: #9a6700; }
svg.chart { width: 100%; height: auto; background: #fffdf8; border: 1px solid var(--faint); border-radius: 12px; }
.titles td { font: 12px/1.35 ui-monospace, SFMono-Regular, Menlo, monospace; }
.titles td:first-child { font: 13px/1.4 var(--sans); color: var(--muted); }
.titles .task { display: block; color: var(--ink); font-weight: 500; margin-bottom: 2px; }
.titles td.chg { background: var(--good-bg); }
footer { margin-top: 3rem; color: var(--muted); font: .78rem/1.45 var(--sans); }
.flow { display: grid; grid-template-columns: 1fr auto 1fr auto 1fr auto 1fr; gap: .5rem; align-items: stretch; margin: 1.2rem 0 1.6rem; font: 13px/1.35 var(--sans); }
.flow .box { padding: .75rem; border: 1px solid var(--faint); border-radius: 12px; background: #fffdf8; }
.flow .arrow { align-self: center; color: var(--sniff); font-size: 1.3rem; }
.flow strong { display: block; margin-bottom: .15rem; }
@media (max-width: 720px) {
.flow { grid-template-columns: 1fr; }
.flow .arrow { transform: rotate(90deg); justify-self: center; }
}
</style>
</head>
<body>
<main class="paper">
<p class="eyebrow">PorkiCoder Research · Tab titles</p>
<h1>Shipping 6t_argmax</h1>
<p class="subtitle">The only replicated win over the production pin. Same 35M, same beam-4, same Hybrid A glue, same centroid v2. New judged-SFT weights.</p>
<p class="meta">16 August 2026 · Terminal-board · Blind Gemini 3.5 Flash-Lite · Continues
<a href="../docs/the_sniff_test.html">The Sniff Test</a>,
<a href="../docs/tab-namer-flan-centroid.html">FLAN + centroid</a>,
<a href="../docs/tab-namer-mid-gsg.html">mid GSG</a>,
<a href="../docs/tab-namer-four-beams.html">four beams</a>,
<a href="session4_results.html">session 4</a>,
<a href="2026-08-16_g300k-flashlite-test.html">176k Flash-Lite CE</a>,
<a href="session9_results.html">session 9 P1</a>,
<a href="session10_results.html">session 10 ship board</a>,
<a href="session11_results.html">session 11 Aug-12 continue</a>
</p>
<section class="callout good">
<h3>Quality vs milliseconds</h3>
<p><strong>Session 11 ship: Title-SFT FLAN raw <code>flat1e4</code></strong> (no Hybrid A) is 7.59 / 85.7% on holdout 1000 vs the Session-10 ship 7.38 / 83.5%. Same 77M / ~193 MB / 29.3 ms graph. The unconstrained FLAN worker is wired in porkr1. Paper: <a href="session11_results.html">session 11</a>.</p>
</section>
<section class="callout warn">
<h3>Two boards</h3>
<p><strong>SO-board</strong> is 1,000 Stack Overflow pages (FLAN + v2 = 6.11 / 88.5% useful). <strong>Terminal-board</strong> is 300 terminal-dev tasks (FLAN + v2 = 7.11 / 93.3% useful). They share a judge and a glue. They do not share a leaderboard. Sealed SO 1000 and terminal-holdout 200 stay sealed.</p>
</section>
<section class="abstract">
<h2>Abstract</h2>
<p>
On Terminal-board the production pin sits around 5.68–5.94 depending on the packet, against a 7.11 FLAN bar.
Session-4 r3 moved a 300-row confirmation packet from 5.56 / 78% to 5.67 / 79%.
Six-teacher judged-SFT (6t_argmax) is the ship model: 5.80 / 79% on its session-5 packet and 5.99 / 85% on a later n=1000 packet.
A ≥5-word retraining of that recipe scored 5.80 / 80% on a fresh 300-row four-way packet, tied with 6t (Δ −0.000, CI includes 0).
min10 on that same packet was 5.75 / 80%. Hinge, off-page decode, terminal-IDF glue, and a local ranker did not add a second bump.
</p>
</section>
<h2>How to read the numbers</h2>
<p>
Every score table uses the same three columns: <strong>mean</strong> (1–10), <strong>useful ≥4</strong>, and <strong>fail ≤2</strong> when the packet published it.
An em dash means that packet did not report that cut.
Paired change versus the pin lives in its own column, never inside the mean cell.
Do not subtract a mean from one packet from a mean in another. Judge drift on the pin is a few tenths.
</p>
<h2>Terminal-board, the ship race</h2>
<p>This page’s first picture is Terminal-board only. SO-board lives in its own section below. Do not read the two as one staircase.</p>
<div class="flow" aria-label="Terminal-board n=300 four-way packet">
<div class="box"><strong>Pin</strong>5.68 / 79%</div>
<div class="arrow">→</div>
<div class="box"><strong>min10</strong>5.75 / 80%</div>
<div class="arrow">→</div>
<div class="box"><strong>min5</strong>5.80 / 80%</div>
<div class="arrow">→</div>
<div class="box"><strong>6t_argmax</strong>5.80 / 80% · ship</div>
</div>
<p class="note">One Terminal-board packet, n=300, 16 August 2026. min5 is tied with 6t. SO-board is not on this strip.</p>
<p>
The bars below are one packet: 1,000 Terminal-shaped tasks (300 official terminal-dev + 700 unused long-pilot).
The dashed FLAN bar is from the session 2.5 packet of 300 and is a reference, not a same-packet score.
</p>
<svg class="chart" viewBox="0 0 920 250" role="img" aria-label="Terminal-board means on the n=1000 packet, with FLAN reference">
<text x="20" y="28" fill="#58625f" font-size="13" font-family="Inter,sans-serif">Mean, 1–10 · n=1000 packet</text>
<!-- axis -->
<line x1="210" y1="48" x2="210" y2="210" stroke="#d6cec1"/>
<line x1="210" y1="210" x2="880" y2="210" stroke="#d6cec1"/>
<text x="210" y="228" fill="#888" font-size="11" font-family="Inter,sans-serif">0</text>
<text x="545" y="228" fill="#888" font-size="11" font-family="Inter,sans-serif">5</text>
<text x="868" y="228" fill="#888" font-size="11" font-family="Inter,sans-serif">10</text>
<!-- FLAN reference 7.11 -->
<line x1="686.4" y1="48" x2="686.4" y2="210" stroke="#32735f" stroke-dasharray="4 4"/>
<text x="690" y="62" fill="#32735f" font-size="11" font-family="Inter,sans-serif">FLAN 7.11</text>
<!-- pin 5.802 -->
<text x="20" y="92" fill="#17211f" font-size="14" font-family="Inter,sans-serif">Pin (production)</text>
<rect x="210" y="74" width="388.7" height="26" fill="#8a938f" rx="3"/>
<text x="606" y="92" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.80 · 85%</text>
<!-- 6t 5.995 -->
<text x="20" y="136" fill="#17211f" font-size="14" font-family="Inter,sans-serif">6t_argmax (ship)</text>
<rect x="210" y="118" width="401.7" height="26" fill="#176b87" rx="3"/>
<text x="620" y="136" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.99 · 85%</text>
<!-- min10 6.000 -->
<text x="20" y="180" fill="#17211f" font-size="14" font-family="Inter,sans-serif">min10 (tied)</text>
<rect x="210" y="162" width="402.0" height="26" fill="#8aa4ad" rx="3"/>
<text x="620" y="180" fill="#17211f" font-size="13" font-family="Inter,sans-serif">6.00 · 85%</text>
</svg>
<p class="note">Useful ≥4 is 85.0% / 85.0% / 84.8% for pin / 6t / min10 on this packet. The 6t bump is the mean, not the useful cut.</p>
<div class="wide">
<table>
<thead>
<tr>
<th>System</th>
<th>Packet</th>
<th class="n">n</th>
<th class="n">Mean</th>
<th class="n">Useful ≥4</th>
<th class="n">Fail ≤2</th>
<th>Paper</th>
</tr>
</thead>
<tbody>
<tr>
<td>FLAN + centroid v2 (bar)</td>
<td>session 2.5</td>
<td class="n">300</td>
<td class="n">7.11</td>
<td class="n">93.3%</td>
<td class="n">2.7%</td>
<td><a href="session4_results.html">session 4</a></td>
</tr>
<tr>
<td>Pin, greedy + v2</td>
<td>session 2.5</td>
<td class="n">300</td>
<td class="n">5.94</td>
<td class="n">83.3%</td>
<td class="n">—</td>
<td><a href="session4_results.html">session 4</a></td>
</tr>
<tr>
<td>Pin, beam-4 + v2</td>
<td>session 4 confirm</td>
<td class="n">300</td>
<td class="n">5.56</td>
<td class="n">78.0%</td>
<td class="n">8.0%</td>
<td><a href="session4_results.html">session 4</a></td>
</tr>
<tr>
<td>r3, FLAN-teacher SFT</td>
<td>session 4 confirm</td>
<td class="n">300</td>
<td class="n">5.67</td>
<td class="n">79.0%</td>
<td class="n">8.0%</td>
<td><a href="session4_results.html">session 4</a></td>
</tr>
<tr>
<td>Pin, beam-4 + v2</td>
<td>session 5</td>
<td class="n">300</td>
<td class="n">5.68</td>
<td class="n">79.7%</td>
<td class="n">—</td>
<td>this page</td>
</tr>
<tr class="pick">
<td><strong>6t_argmax (ship)</strong></td>
<td>session 5</td>
<td class="n">300</td>
<td class="n">5.80</td>
<td class="n">79.0%</td>
<td class="n">—</td>
<td>this page</td>
</tr>
<tr>
<td>Pin, beam-4 + v2</td>
<td>n=1000</td>
<td class="n">1000</td>
<td class="n">5.80</td>
<td class="n">85.0%</td>
<td class="n">3.4%</td>
<td>this page</td>
</tr>
<tr class="pick">
<td><strong>6t_argmax (ship)</strong></td>
<td>n=1000</td>
<td class="n">1000</td>
<td class="n">5.99</td>
<td class="n">85.0%</td>
<td class="n">3.3%</td>
<td>this page</td>
</tr>
<tr class="context">
<td>min10 (tied, not shipped)</td>
<td>n=1000</td>
<td class="n">1000</td>
<td class="n">6.00</td>
<td class="n">84.8%</td>
<td class="n">3.3%</td>
<td>this page</td>
</tr>
<tr>
<td>Pin, beam-4 + v2</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">5.68</td>
<td class="n">79.0%</td>
<td class="n">8.0%</td>
<td>this page</td>
</tr>
<tr>
<td>min10</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">5.75</td>
<td class="n">79.7%</td>
<td class="n">8.3%</td>
<td>this page</td>
</tr>
<tr class="context">
<td>min5 (tied with 6t)</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">5.80</td>
<td class="n">80.3%</td>
<td class="n">8.7%</td>
<td>this page</td>
</tr>
<tr class="pick">
<td><strong>6t_argmax (ship)</strong></td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">5.80</td>
<td class="n">80.3%</td>
<td class="n">8.7%</td>
<td>this page</td>
</tr>
</tbody>
</table>
</div>
<p class="note">r3 and 6t now use the same mean / useful / fail columns as every other row. The next table is the only place paired deltas appear.</p>
<h3>Same-packet change versus the pin</h3>
<div class="wide">
<table>
<thead>
<tr>
<th>Arm</th>
<th>Packet</th>
<th class="n">n</th>
<th class="n">Paired Δ</th>
<th class="n">95% CI</th>
</tr>
</thead>
<tbody>
<tr>
<td>r3</td>
<td>session 4 first</td>
<td class="n">300</td>
<td class="n">+0.07</td>
<td class="n">[+0.01, +0.14]</td>
</tr>
<tr>
<td>r3</td>
<td>session 4 confirm</td>
<td class="n">300</td>
<td class="n">+0.12</td>
<td class="n">[+0.04, +0.20]</td>
</tr>
<tr class="pick">
<td>6t_argmax</td>
<td>session 5</td>
<td class="n">300</td>
<td class="n">+0.12</td>
<td class="n">[+0.04, +0.20]</td>
</tr>
<tr class="pick">
<td>6t_argmax</td>
<td>n=1000</td>
<td class="n">1000</td>
<td class="n">+0.19</td>
<td class="n">[+0.14, +0.25]</td>
</tr>
<tr class="context">
<td>min10 vs 6t</td>
<td>n=1000</td>
<td class="n">1000</td>
<td class="n">+0.005</td>
<td class="n">[−0.017, +0.028]</td>
</tr>
<tr>
<td>min5 vs pin</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">+0.12</td>
<td class="n">[+0.04, +0.21]</td>
</tr>
<tr class="context">
<td>min5 vs 6t</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">0.00</td>
<td class="n">[−0.03, +0.02]</td>
</tr>
<tr>
<td>min5 vs min10</td>
<td>min5 four-way</td>
<td class="n">300</td>
<td class="n">+0.05</td>
<td class="n">[+0.00, +0.10]</td>
</tr>
</tbody>
</table>
</div>
<h2>SO-board, how we got here</h2>
<p>Canonical comparable packet is the mid-GSG six-way after centroid v2. The 12.7M and early Hybrid A rows are the earlier seven-way packet; they are listed, not mixed into the chart.</p>
<svg class="chart" viewBox="0 0 920 310" role="img" aria-label="SO-board six-way means after centroid v2">
<text x="20" y="28" fill="#58625f" font-size="13" font-family="Inter,sans-serif">Mean, 1–10 · SO-board 6-way n=1000</text>
<line x1="250" y1="48" x2="250" y2="270" stroke="#d6cec1"/>
<line x1="250" y1="270" x2="880" y2="270" stroke="#d6cec1"/>
<text x="250" y="288" fill="#888" font-size="11" font-family="Inter,sans-serif">0</text>
<text x="565" y="288" fill="#888" font-size="11" font-family="Inter,sans-serif">5</text>
<text x="868" y="288" fill="#888" font-size="11" font-family="Inter,sans-serif">10</text>
<!-- 6.11 FLAN -->
<text x="20" y="78" fill="#17211f" font-size="14" font-family="Inter,sans-serif">FLAN + v2</text>
<rect x="250" y="60" width="385.0" height="24" fill="#32735f" rx="3"/>
<text x="643" y="78" fill="#17211f" font-size="13" font-family="Inter,sans-serif">6.11 · 88.5%</text>
<!-- 5.96 pin greedy -->
<text x="20" y="118" fill="#17211f" font-size="14" font-family="Inter,sans-serif">B-9500 + v2</text>
<rect x="250" y="100" width="375.5" height="24" fill="#176b87" rx="3"/>
<text x="634" y="118" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.96 · 85.0%</text>
<!-- 5.91 -->
<text x="20" y="158" fill="#17211f" font-size="14" font-family="Inter,sans-serif">FLAN + old centroid</text>
<rect x="250" y="140" width="372.3" height="24" fill="#8aa4ad" rx="3"/>
<text x="630" y="158" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.91 · 86.5%</text>
<!-- 5.89 -->
<text x="20" y="198" fill="#17211f" font-size="14" font-family="Inter,sans-serif">B-9500 + old centroid</text>
<rect x="250" y="180" width="371.1" height="24" fill="#8a938f" rx="3"/>
<text x="630" y="198" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.89 · 85.1%</text>
<!-- 5.11 -->
<text x="20" y="238" fill="#17211f" font-size="14" font-family="Inter,sans-serif">centroid v2 only</text>
<rect x="250" y="220" width="321.9" height="24" fill="#c5c0b4" rx="3"/>
<text x="580" y="238" fill="#17211f" font-size="13" font-family="Inter,sans-serif">5.11 · 72.3%</text>
</svg>
<div class="wide">
<table>
<thead>
<tr>
<th>System</th>
<th>Packet</th>
<th class="n">n</th>
<th class="n">Mean</th>
<th class="n">Useful ≥4</th>
<th class="n">Fail ≤2</th>
<th>Paper</th>
</tr>
</thead>
<tbody>
<tr>
<td>12.7M kd-verne</td>
<td>7-way</td>
<td class="n">1000</td>
<td class="n">3.01</td>
<td class="n">30.5%</td>
<td class="n">53.9%</td>
<td><a href="../docs/the_sniff_test.html">sniff</a> · <a href="../docs/tab-namer-flan-centroid.html">FLAN + centroid</a></td>
</tr>
<tr>
<td>Hybrid A, early 35M + old centroid</td>
<td>7-way</td>
<td class="n">1000</td>
<td class="n">4.72</td>
<td class="n">64%</td>
<td class="n">17%</td>
<td><a href="../docs/tab-namer-flan-centroid.html">FLAN + centroid</a></td>
</tr>
<tr>
<td>B-9500 + old centroid</td>
<td>4-way</td>
<td class="n">1000</td>
<td class="n">5.65</td>
<td class="n">78.6%</td>
<td class="n">3.8%</td>
<td><a href="../docs/tab-namer-mid-gsg.html">mid GSG</a></td>
</tr>
<tr>
<td>B-9500 + centroid v2, greedy</td>
<td>6-way</td>
<td class="n">1000</td>
<td class="n">5.96</td>
<td class="n">85.0%</td>
<td class="n">0.7%</td>
<td><a href="../docs/tab-namer-mid-gsg.html">mid GSG</a></td>
</tr>
<tr>
<td>B-9500 beam-4 + v2</td>
<td>focused 3-way</td>
<td class="n">1000</td>
<td class="n">6.16</td>
<td class="n">—</td>
<td class="n">—</td>
<td><a href="../docs/tab-namer-four-beams.html">four beams</a></td>
</tr>
<tr>
<td>B-9500 beam-4 + v2</td>
<td>sealed 1000</td>
<td class="n">1000</td>
<td class="n">6.20</td>
<td class="n">—</td>
<td class="n">—</td>
<td><a href="../docs/tab-namer-four-beams.html">four beams</a></td>
</tr>
<tr class="context">
<td>FLAN + centroid v2 (bar)</td>
<td>6-way</td>
<td class="n">1000</td>
<td class="n">6.11</td>
<td class="n">88.5%</td>
<td class="n">0.6%</td>
<td><a href="../docs/tab-namer-mid-gsg.html">mid GSG</a></td>
</tr>
</tbody>
</table>
</div>
<p class="note">Beam-4 sealed/focused packets published ≥5 / ≥6 cuts, not ≥4, so those two useful cells stay blank rather than mixing cutoffs. r3 on the reusable SO 200 scored 5.78 / 88.5% useful and lost to the pin — Terminal-only win.</p>
<h2>Later arms on Terminal-board</h2>
<p>Status of everything after the pin. Means below are from the packet that judged that arm. Dead arms keep a mean of em dash when we only published a paired delta.</p>
<div class="wide">
<table>
<thead>
<tr>
<th>Technique</th>
<th>Status</th>
<th class="n">Mean</th>
<th class="n">Useful ≥4</th>
<th class="n">vs pin</th>
</tr>
</thead>
<tbody>
<tr>
<td>Pin, B-9500 beam-4 + v2 + Hybrid A</td>
<td><span class="status keep">in the app</span></td>
<td class="n">5.80</td>
<td class="n">85.0%</td>
<td class="n">0</td>
</tr>
<tr>
<td>r3, FLAN-teacher judged-SFT</td>
<td><span class="status keep">kept</span></td>
<td class="n">5.67</td>
<td class="n">79.0%</td>
<td class="n">+0.12</td>
</tr>
<tr class="pick">
<td>6t_argmax, 6-teacher judged-SFT</td>
<td><span class="status keep">ship this</span></td>
<td class="n">5.99</td>
<td class="n">85.0%</td>
<td class="n">+0.19</td>
</tr>
<tr class="context">
<td>min10, same recipe, ≥10-word train filter</td>
<td><span class="status keep">tied</span></td>
<td class="n">6.00</td>
<td class="n">84.8%</td>
<td class="n">+0.20</td>
</tr>
<tr class="context">
<td>min5, same recipe, ≥5-word train filter</td>
<td><span class="status keep">tied with 6t</span></td>
<td class="n">5.80</td>
<td class="n">80.3%</td>
<td class="n">+0.12</td>
</tr>
<tr>
<td>4-teacher argmax</td>
<td><span class="status keep">weaker</span></td>
<td class="n">5.77</td>
<td class="n">79.0%</td>
<td class="n">+0.09</td>
</tr>
<tr>
<td>Self-distill best-of-K</td>
<td><span class="status dead">dead</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">~0</td>
</tr>
<tr>
<td>Contrastive hinge vs pin</td>
<td><span class="status dead">dead</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">+0.04 vs 6t</td>
</tr>
<tr>
<td>Off-page / free-vocab decode</td>
<td><span class="status dead">dead</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">−0.89</td>
</tr>
<tr>
<td>Terminal-IDF centroid</td>
<td><span class="status dead">dead</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">−0.19 vs 6t</td>
</tr>
<tr>
<td>Ranker / owned proposer</td>
<td><span class="status dead">dead here</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">−0.13</td>
</tr>
<tr>
<td>Session 3 GRPO / 70M</td>
<td><span class="status dead">banned</span></td>
<td class="n">—</td>
<td class="n">—</td>
<td class="n">STOP</td>
</tr>
</tbody>
</table>
</div>
<p class="note">Pin / 6t / min10 means in this summary table are the n=1000 packet. min5 / r3 / 4-teacher means are their own 300-row Terminal packets. Dead-arm means were never the product claim — only the paired delta was.</p>
<h2>115 real titles, tasks of at least 10 words</h2>
<p>Same Hybrid A + centroid v2 glue. Left to right: production pin → session-4 r3 → session-5 <code>6t_argmax</code> → FLAN (the bar). Green cells differ from the pin. Secret rows were deleted from this table, not masked in place.</p>
<table class="titles">
<thead>
<tr>
<th>Task</th>
<th>Pin (production)</th>
<th>r3 (FLAN-teacher SFT)</th>
<th>6t_argmax (ship this)</th>
<th>FLAN + centroid (bar)</th>
</tr>
</thead>
<tbody>
<tr><td class="left"><span class="task">so i want to upgrade big-deployment, fix-bugs and feat-pig skills. basically want to add a kimi review via mcp consult after each call. so currently big deployment runs feat pig, t…</span></td><td>pig mcp Basically</td><td class="chg">pig Runs Basically</td><td class="chg">mcp Fix Basically</td><td>Kimi Review Basically</td></tr>
<tr><td class="left"><span class="task">so right now everytime it starts in a new machine or even in this machine after a new install; i get asked for api keys; and its annoying. instead of bring your own key model; it s…</span></td><td>Remove api Anthropic</td><td class="chg">Remove Names Anthropic</td><td class="chg">Remove Keys Anthropic</td><td>Bundled Keys Anthropic</td></tr>
<tr><td class="left"><span class="task">i want pigwise to look more professional; get rid of the pig emoji etc. and since its an extension of porkicoder (thats the plan) it should have similar style and theme. consult ge…</span></td><td>pig emoji Pigwise</td><td class="chg">Git emoji Pigwise</td><td class="chg">emoji extension Pigwise</td><td>Pigwise Etc Similar</td></tr>
<tr><td class="left"><span class="task">i got too many unused tables in supabase; can you check the ones that are not being used and get rid of them</span></td><td>check supabase Ones</td><td class="chg">Check Tables Supabase</td><td class="chg">Check Tables Supabase</td><td>Tables Supabase Ones</td></tr>
<tr><td class="left"><span class="task">tell me where we are right now and remains to be done for this research paper</span></td><td>right done Tell</td><td class="chg">Right Me Tell</td><td class="chg">Right Now Tell</td><td>Research Tell Done</td></tr>
<tr><td class="left"><span class="task">we need to create an automatedorchestration for new reddit account creation. for emails i have contabo and postal inside email_client repo one level up; that is where you get the e…</span></td><td>logging account Few</td><td class="chg">Create Signup Few</td><td class="chg">Signup Email Few</td><td>Reddit Automatedorchestration</td></tr>
<tr><td class="left"><span class="task">ask gpt 5.5 via mcp consult to review our work so far and identify gaps, opportunities for novel research, etc.</span></td><td>gpt mcp Far</td><td class="chg">mcp Gpt Far</td><td class="chg">mcp Gpt Far</td><td>Review Work Far</td></tr>
<tr><td class="left"><span class="task">so this skill isn't following brand guidelines; i wonder if it still has access to them and the logos. can you check?</span></td><td>Access check Still</td><td class="chg">Check Access Still</td><td class="chg">Check Access Still</td><td>Check Brand Still</td></tr>
<tr><td class="left"><span class="task">so someone used the tool not long ago and got this output, the image on the right looks like bad aspect ratio mismatch and on the left it's dark with no [company] logo. Should be o…</span></td><td>gpt mcp Parallel</td><td class="chg">Image Output Parallel</td><td class="chg">Parallel Image Happen</td><td>[company] Logo Parallel</td></tr>
<tr><td class="left"><span class="task">what are some cheaper alternatives to google search with grounding cause its getting expensive</span></td><td>Google search Grounding</td><td class="chg">Google Search Grounding</td><td class="chg">google search Grounding</td><td>Cheaper Search Grounding</td></tr>
<tr><td class="left"><span class="task">fetch me a link to a reputable paper that can back this up: post training cannot create enhanced intelligence, only pretraining can do that, post training is good for style format …</span></td><td>Cannot fetch Pretraining</td><td class="chg">Create Post Pretraining</td><td class="chg">Create Post Pretraining</td><td>Paper Pretraining Far</td></tr>
<tr><td class="left"><span class="task">it named the tab 'unknown command' in this instance, which is not useful: ; please suggest a fix</span></td><td>Instance command Useful</td><td class="chg">Fix Instance Useful</td><td>Instance command Useful</td><td>Fix Tab Useful</td></tr>
<tr><td class="left"><span class="task">what's the open rate on the last bulk job finder email sent on Monday to premium customers. did everyone get the mail or were a few left out? did everyone get jobs or did a few ema…</span></td><td>Open email Job</td><td class="chg">Sent Email Job</td><td class="chg">Sent Email Job</td><td>Job Finder Monday</td></tr>
<tr><td class="left"><span class="task">so i noted a recent submission was shipped with a spelling error 'to' instead of 'too' can you flag that and regen one with correct spelling and email the new version to the person…</span></td><td>Error version Regen</td><td class="chg">Error Email Regen</td><td class="chg">Error Version Regen</td><td>Flag Spelling Regen</td></tr>
<tr><td class="left"><span class="task">also for the terminal zoom i want same level of zoom that you get after pressing cmd and `+` five times. not 5x zoom. take a screenshot and evaluate if the zoom level look comforta…</span></td><td>git commit Terminal</td><td class="chg">Git Terminal Commit</td><td class="chg">Git commit Terminal</td><td>Terminal Zoom Commit</td></tr>
<tr><td class="left"><span class="task">1 yes, 2. agreed i own it; assign to [name], keep [name] as a collaborator</span></td><td>Assign own [name]</td><td class="chg">Assign It [name]</td><td class="chg">Assign Own [name]</td><td>Assignment [name] Assign</td></tr>
<tr><td class="left"><span class="task">let's give each bot their own distinct personalities so when they all start commenting it feels like 5 different people.</span></td><td>Different commenting People</td><td>Different commenting People</td><td class="chg">let commenting People</td><td>Bot Personalities Commenting</td></tr>
<tr><td class="left"><span class="task">fix my landing page it is hard to read now</span></td><td>Read page Hard</td><td class="chg">Read Page Hard</td><td class="chg">Read Page Hard</td><td>Landing Page Hard</td></tr>
<tr><td class="left"><span class="task">got a proxy error with the ok body bot; see what happened. run /full</span></td><td>Error ok Proxy</td><td class="chg">Proxy Error Body</td><td class="chg">Proxy Error Body</td><td>Proxy Error Body</td></tr>
<tr><td class="left"><span class="task">how are the bots doing, two of them should be commenting in tandem now</span></td><td>Two commenting Doing</td><td class="chg">Doing commenting Them</td><td class="chg">Doing commenting Them</td><td>Bot Commenting Doing</td></tr>
<tr><td class="left"><span class="task">the login page doesn't look the same as landing page, all public facing pages should have the same look and feel</span></td><td>Login page Pages</td><td class="chg">Login Page Pages</td><td class="chg">Login Page Pages</td><td>Login Page Pages</td></tr>
<tr><td class="left"><span class="task">can you access google search console via cli and then check this notif: To the owner of resumehog.com:</span></td><td>Google search Notif</td><td class="chg">Google Search Notif</td><td class="chg">Google console Notif</td><td>Google Search Notif</td></tr>
<tr><td class="left"><span class="task">use my git to analyze resumetailor repo and tell me if it qualifies for any type of patent.</span></td><td>git repo Resumetailor</td><td class="chg">Git repo Resumetailor</td><td class="chg">Git repo Resumetailor</td><td>Patent Resumetailor Repo</td></tr>
<tr><td class="left"><span class="task"># Gemma 4 31B on Apple Silicon (M4/M5) — Local AI Assistant Setup</span></td><td>Setup M4 Gemma</td><td>Setup M4 Gemma</td><td class="chg">AI Setup Gemma</td><td>Local AI Gemma</td></tr>
<tr><td class="left"><span class="task">how to install opencode on mac and integrate it into porkicoder</span></td><td>opencode install Porkicoder</td><td class="chg">Opencode Install Porkicoder</td><td class="chg">Opencode Install Porkicoder</td><td>Install Opencode Porkicoder</td></tr>
<tr><td class="left"><span class="task">check on my bots, one of them might be shadow banned i got email notif</span></td><td>Email check Notif</td><td class="chg">Email Check Notif</td><td class="chg">Check Email Notif</td><td>Check Bots Notif</td></tr>
<tr><td class="left"><span class="task">can you investigate what wrong with my last agent mode run it tried to create over 1000 images i think or i misunderstood the ui; also when i switched tabs it was gone which is a m…</span></td><td>ui tabs Persistance</td><td class="chg">Create Tabs Persistance</td><td class="chg">Agent Mode Persistance</td><td>Agent Mode Persistance</td></tr>
<tr><td class="left"><span class="task">suggest upgrades to job finder module to make jobs more relevant</span></td><td>finder job Relevant</td><td>finder job Relevant</td><td class="chg">Suggest job Finder</td><td>Job Finder Relevant</td></tr>
<tr><td class="left"><span class="task">pickup where you left off, spawn sub agents if needed</span></td><td>Needed sub Left</td><td class="chg">Needed Sub Left</td><td class="chg">Needed Sub Left</td><td>Sub Agents Needed</td></tr>
<tr><td class="left"><span class="task">so tiktok approved my dev token for production, tell me what to do next (context: had applied a few weeks ago, after few rounds of back and forth finally got approved)</span></td><td>Dev token Production</td><td class="chg">Token Dev Production</td><td>Dev token Production</td><td>TikTok Production Applied</td></tr>
<tr><td class="left"><span class="task">❯ ok i just manually triggered the upgrade conversion action buy purchasing a subscription; but in</span></td><td>ok conversion Manually</td><td>ok conversion Manually</td><td class="chg">Conversion Action Manually</td><td>Subscription Upgrade Conversion</td></tr>
<tr><td class="left"><span class="task">-When i generate the press pdf it says missing direction, although it seems like I've filled out all the fields so I am not sure what direction I am missing</span></td><td>Generate missing Direction</td><td class="chg">Generate Fields Direction</td><td class="chg">missing fields Direction</td><td>PDF Direction Although</td></tr>
<tr><td class="left"><span class="task">when i press start it will go straight into workhorse, right?</span></td><td>go start Workhorse</td><td class="chg">When Start Workhorse</td><td class="chg">When Start Workhorse</td><td>Start Workhorse Right</td></tr>
<tr><td class="left"><span class="task">lets try an experiment; first retire polite willow; then lets try a new tactic; instead of one account commenting non stop during a 30 minute cycle; we do a comment handover after …</span></td><td>commenting issue Handover</td><td class="chg">commenting Issue Handover</td><td class="chg">comment commenting Handover</td><td>Comment Handover Willow</td></tr>
<tr><td class="left"><span class="task">study official google ads docs and figure out how to fix this</span></td><td>Google docs Official</td><td class="chg">docs fix Official</td><td class="chg">Google Docs Official</td><td>Google Ads Official</td></tr>
<tr><td class="left"><span class="task">why are you doing manual deploy go through circle ci</span></td><td>ci deploy Doing</td><td class="chg">Deploy Go Doing</td><td>ci deploy Doing</td><td>Circle CI Deploy</td></tr>
<tr><td class="left"><span class="task">instead of one account posting for 30 minutes non stop lets rotate accounts like this: account A posts 1st comment; account B posts 2nd comment etc. make this code change, then pro…</span></td><td>Change localhost 3001</td><td class="chg">Localhost comment 3001</td><td class="chg">Localhost comment 3001</td><td>Rotate Accounts 3001</td></tr>
<tr><td class="left"><span class="task">as i thought, all resumehog mentions are being removed. we lost the reddit botting battle. go into mac mini kill all email notifications.</span></td><td>Reddit Email Resumehog</td><td class="chg">reddit Email Resumehog</td><td class="chg">All mentions Resumehog</td><td>Kill Notifications Resumehog</td></tr>
<tr><td class="left"><span class="task">right now only active campaign is r1_tailor right? can we get rid of the other campaign?</span></td><td>r1 active Now</td><td class="chg">Get right Active</td><td class="chg">Get right Active</td><td>Campaign Active Now</td></tr>
<tr><td class="left"><span class="task">are reddit ads text based like google search ads or image based</span></td><td>Search image Reddit</td><td class="chg">Search Text Reddit</td><td class="chg">Text Search Reddit</td><td>Reddit Ads Google</td></tr>
<tr><td class="left"><span class="task">1 (do it) 2 (do it). but make sure you have ran code-review to check for gaps in your work.</span></td><td>Make check Ran</td><td class="chg">Check Code Ran</td><td class="chg">Check Code Ran</td><td>Code Review Ran</td></tr>
<tr><td class="left"><span class="task">yeah wrong one again, maybe we try in an incognito window</span></td><td>incognito window Maybe</td><td class="chg">window incognito Maybe</td><td class="chg">window incognito Maybe</td><td>Incognito Maybe Again</td></tr>
<tr><td class="left"><span class="task">ok looks great. a) yes b) yes c) also yes</span></td><td>great ok Looks</td><td class="chg">ok great Looks</td><td class="chg">ok great Looks</td><td>Great Looks Also</td></tr>
<tr><td class="left"><span class="task">one more request; i want to add a panel where there is avisual representation of the lesson we are trying to learn, like i remember in coursera gradient descent was explained using…</span></td><td>one request Gemini</td><td class="chg">Implementation Request Gemini</td><td class="chg">Implementation request Gemini</td><td>Visual Presentation Gemini</td></tr>
<tr><td class="left"><span class="task">considering how much it costs to create a lesson via gemini and threejs; what should be our pricing for this product if we go for a subscription model and unlimited lessons (or is …</span></td><td>Create threejs Gemini</td><td class="chg">Create Model Gemini</td><td class="chg">Threejs model Gemini</td><td>Pricing Gemini Product</td></tr>
<tr><td class="left"><span class="task">check your work after you have implemented it; run code-review in parallel with kimi 2.7 via mcp-consult</span></td><td>mcp check Parallel</td><td class="chg">Parallel Check Mcp</td><td class="chg">Parallel Check Mcp</td><td>Code Review Mcp</td></tr>
<tr><td class="left"><span class="task">are the keywords relevant to what we are trying to sell? run 3 adverserial agents; kimi gemini and gpt 5.5 (you) and get consensus on a final list</span></td><td>Run gpt Gemini</td><td class="chg">gpt List Gemini</td><td class="chg">gpt List Gemini</td><td>Keyword Consensus Gemini</td></tr>
<tr><td class="left"><span class="task">can name dropping product name on reddit count as promotion or only links count as promotion when it comes to using reddit api</span></td><td>reddit api Product</td><td class="chg">reddit count Product</td><td class="chg">reddit count Product</td><td>Reddit API Product</td></tr>
<tr><td class="left"><span class="task">i just added rising keywords and top queries in the downloads folder from google trends. these are specific for ai and resume related searches in USA. use those to refine your stra…</span></td><td>Google ai Usa</td><td class="chg">Google queries Usa</td><td class="chg">google downloads Usa</td><td>Refine Strategy Usa</td></tr>
<tr><td class="left"><span class="task">like brave search or a similar alternative). Basically want to lower cost by replacing gemini 3.1</span></td><td>Search by Gemini</td><td class="chg">Search By Gemini</td><td class="chg">Search Alternative Gemini</td><td>Search Alternative Gemini</td></tr>
<tr><td class="left"><span class="task">save it in .env and feel free to put it in the ec2 box if needed</span></td><td>env ec2 Needed</td><td>env ec2 Needed</td><td class="chg">Env box Needed</td><td>Save Env Needed</td></tr>
<tr><td class="left"><span class="task">tbh that did not work very well, a few jobs based in canada, but they are in toronto not vancouver, and one from mexico and one from lisbon. but strange thing is i dont mind it; le…</span></td><td>Change tbh Vancouver</td><td class="chg">Doing Email Vancouver</td><td class="chg">Doing Email Vancouver</td><td>Email Delivery Vancouver</td></tr>
<tr><td class="left"><span class="task">some feedback: ui looks awful, get gemini 3.1 pro to revamp it, there are basic issues with text spacing and the upload button looks amateurish, i am very disappointed with the ui.…</span></td><td>Detect ui Resumehog</td><td class="chg">Detect Button Resumehog</td><td class="chg">Detect Button Resumehog</td><td>Resume Grader Resumehog</td></tr>
<tr><td class="left"><span class="task">yeah fix that thing too and commit and push changes to git.</span></td><td>Git push Commit</td><td>Git push Commit</td><td class="chg">Git commit Push</td><td>Commit Push Thing</td></tr>
<tr><td class="left"><span class="task">ok great work. just make sure we are now using glm 5.2 as primary model for lesson planning, feedback, review etc. only time kimi 2.7 is used is to do code review of the glm 5.2 cr…</span></td><td>Git glm Gitea</td><td class="chg">Git Push Gitea</td><td class="chg">git push Gitea</td><td>Models Gitea Great</td></tr>
<tr><td class="left"><span class="task">ok then lets use glm 5.2 instead of kimi since its cheaper. and i will follow your reocmmendation to add this in the authenticated full-report layer.</span></td><td>glm ok Reocmmendation</td><td class="chg">glm layer Reocmmendation</td><td class="chg">glm layer Reocmmendation</td><td>Reocmmendation Layer Report</td></tr>
<tr><td class="left"><span class="task">can you add these keywords to my active campaign as phrase match keywords: "resume ai"</span></td><td>Add active Match</td><td class="chg">Add Active Match</td><td class="chg">Add Active Match</td><td>Add Keywords Active</td></tr>
<tr><td class="left"><span class="task">update after hitting esc twice it came back; but i am unable to scroll to the bottom using mouse wheel using down arrow to scroll down very slowly</span></td><td>Update arrow Esc</td><td class="chg">Update Scroll Esc</td><td class="chg">Mouse scroll Esc</td><td>Scroll Esc Mouse</td></tr>
<tr><td class="left"><span class="task">how big is this market in brisbane in usd per annum</span></td><td>brisbane big Annum</td><td class="chg">brisbane Usd Annum</td><td class="chg">Big Usd Annum</td><td>Market Annum Usd</td></tr>
<tr><td class="left"><span class="task">feasbility analysis for indie hacker who wants to make 50k CAD per month</span></td><td>make feasbility Indie</td><td class="chg">Make CAD Per</td><td class="chg"></td><td>Analysis Indie 50k</td></tr>
<tr><td class="left"><span class="task">we dont *want* to do a manual service we want to bring ai automation to this industry</span></td><td>service ai Want</td><td class="chg">ai service Want</td><td class="chg">ai service Want</td><td>AI Automation Service</td></tr>
<tr><td class="left"><span class="task">why did we use pytorch api instead of mlx in the first place</span></td><td>pytorch api Mlx</td><td class="chg">Pytorch api Mlx</td><td class="chg">Pytorch api Mlx</td><td>Pytorch Mlx Place</td></tr>
<tr><td class="left"><span class="task">why did we use pytorch api instead of mlx in the first place (pull last pr we made for context)</span></td><td>pytorch api Mlx</td><td class="chg">Pytorch api Mlx</td><td class="chg">Pytorch api Mlx</td><td>Pytorch Mlx Pull</td></tr>
<tr><td class="left"><span class="task">ok and the others with 'good' how do we promote that to 'excellent' search official google docs and take action based on that, also get a second opinion from glm 5.2</span></td><td>glm docs Official</td><td class="chg">Search docs Official</td><td class="chg">docs glm Official</td><td>Promote Official Glm</td></tr>
<tr><td class="left"><span class="task">but what about the url isnt that the wrong url shouldnt the stuff be under github.com/[user] instead of github.com/[user]/[user] or am i missing something</span></td><td>github url Shouldnt</td><td class="chg">Github url Shouldnt</td><td class="chg">Github Missing Shouldnt</td><td>URL Shouldnt [user]</td></tr>
<tr><td class="left"><span class="task">get official docs and see if we are missing anything</span></td><td>docs missing Official</td><td>docs missing Official</td><td class="chg">Missing docs Official</td><td>Missing Docs Official</td></tr>
<tr><td class="left"><span class="task">fanout four glm 5.2 agents to sweep for dead code in this repo</span></td><td>glm code Fanout</td><td class="chg">glm repo Fanout</td><td class="chg">glm repo Fanout</td><td>Dead Code Fanout</td></tr>
<tr><td class="left"><span class="task">yeah fill the gaps so a fresh assistant knows what to do</span></td><td>fill so Yeah</td><td class="chg">Fill So Yeah</td><td class="chg">Fill So Yeah</td><td>Fill Gaps Yeah</td></tr>
<tr><td class="left"><span class="task">review tools were attempted but unavailable in this session. can you check the porkr1 repo and tell me why is that happening?</span></td><td>check session Porkr1</td><td class="chg">Session Check Porkr1</td><td class="chg">Session Check Porkr1</td><td>Review Tools Porkr1</td></tr>
<tr><td class="left"><span class="task">made some changes related to google ads recently want to make sure i did not break anything. run a full audit, get glm 5.2 and kimi 2.7 to weigh in, but be mindful of their limitat…</span></td><td>google glm Paralelly</td><td class="chg">Google glm Paralelly</td><td class="chg">Google glm Paralelly</td><td>Google Ads Paralelly</td></tr>
<tr><td class="left"><span class="task">get glm 5.2 and kimi 2.7 to weigh in</span></td><td>get glm Kimi</td><td class="chg">Get glm Kimi</td><td class="chg">Get glm Kimi</td><td>Glm Get Kimi</td></tr>
<tr><td class="left"><span class="task">ok exclude [name], [email], [email], [name] etc. these are free accounts i gave away. after this narrow number of people who paid for it, then create an excel/csv that i can upload…</span></td><td>csv ok [name]</td><td class="chg">[user] Number [name]</td><td class="chg">[user] Csv [name]</td><td>Paid [name] [name]</td></tr>
<tr><td class="left"><span class="task">what is the estimated lift i can expect from this upload</span></td><td>Upload expect Estimated</td><td class="chg">expect upload Estimated</td><td class="chg">expect Upload Estimated</td><td>Upload Lift Expect</td></tr>
<tr><td class="left"><span class="task">forget about tony, lets move on. reply to venice first then write handoff doc for pending tasks.</span></td><td>Move pending Venice</td><td class="chg">pending Tasks Venice</td><td class="chg">pending Tasks Venice</td><td>Reply Tasks Venice</td></tr>
<tr><td class="left"><span class="task">yeah write down strategy with date, in markdown then push to gitea</span></td><td>Push down Gitea</td><td class="chg">Push markdown Gitea</td><td class="chg">markdown push Gitea</td><td>Write Strategy Gitea</td></tr>
<tr><td class="left"><span class="task">approval → ship via the repo's own deploy mechanism → verify the deployed result.</span></td><td>Deploy repo Own</td><td class="chg">Deploy Own Repo</td><td class="chg">Deploy Own Repo</td><td>Deploy Own Repo</td></tr>
<tr><td class="left"><span class="task">and i dont want this skill only sequestered to me. all porkicoder users should have access to it. make necessary arrangements and get glm 5.2 to review your work.</span></td><td>Access glm Sequestered</td><td>Access glm Sequestered</td><td class="chg">glm Get Sequestered</td><td>Skill Sequestered Necessary</td></tr>
<tr><td class="left"><span class="task">and it should be usable via codex, claude code and open code (all terminal, dont worry about chatview).</span></td><td>Open code Codex</td><td class="chg">Open Terminal Codex</td><td class="chg">Open Terminal Codex</td><td>Terminal Codex Usable</td></tr>
<tr><td class="left"><span class="task">oh ok since its ui only can you give [name] step by step instructions, explain its ui-only so they can handle it on their end, also tag [name]</span></td><td>oh ui [name]</td><td class="chg">ui tag [name]</td><td class="chg">Handle ui [name]</td><td>[name] Instructions Step</td></tr>
<tr><td class="left"><span class="task">try uplink capacity again i switched form wifi to cable</span></td><td>form uplink Again</td><td>form uplink Again</td><td class="chg">Uplink Form Again</td><td>Uplink Form Again</td></tr>
<tr><td class="left"><span class="task">│ Automation │ assigned to you │ build.; get glm 5.2 to fanout with kimi 2.7 then consolidate.</span></td><td>Get glm Assigned</td><td class="chg">glm Automation Assigned</td><td class="chg">glm Automation Assigned</td><td>Kimi Assigned Glm</td></tr>
<tr><td class="left"><span class="task">go into porkr1 repo and fix it so that codex can use these tools next time onwards</span></td><td>Go codex Onwards</td><td class="chg">Fix Time Codex</td><td class="chg">Fix tools Codex</td><td>Fix Porkr1 Codex</td></tr>
<tr><td class="left"><span class="task">fanout agents and look for dead code in this repo; you, kimi 2.7 and glm 5.2</span></td><td>glm code Repo</td><td class="chg">glm repo Look</td><td class="chg">glm repo Look</td><td>Dead Code Repo</td></tr>
<tr><td class="left"><span class="task">update big-deployment, feat-pig and fix-bugs skill to do this automatically in the future (push all these skills into git and gitea)</span></td><td>Git pig Gitea</td><td class="chg">Git Update Gitea</td><td class="chg">Git Update Gitea</td><td>Update Skills Gitea</td></tr>
<tr><td class="left"><span class="task">make sure these skills come pre-loaded with every fresh porkicoder install for codex, opencode and claude code. fanout agents with different worktrees.</span></td><td>Code install Opencode</td><td class="chg">Install Code Opencode</td><td class="chg">Code Install Opencode</td><td>Skills Opencode Worktrees</td></tr>
<tr><td class="left"><span class="task">ok write down these recommendations in a markdown file then push to gitea</span></td><td>markdown file Gitea</td><td class="chg">Write markdown Gitea</td><td class="chg">markdown push Gitea</td><td>Write Recommendations Gitea</td></tr>
<tr><td class="left"><span class="task">can you take a look at the image upload we did recently</span></td><td>Image upload Recently</td><td class="chg">Upload Image Recently</td><td class="chg">Upload Image Recently</td><td>Image Upload Recently</td></tr>
<tr><td class="left"><span class="task">there should be a list of people who have unsubbed and we dont send marketing emails to people in that list</span></td><td>send list People</td><td class="chg">Send List People</td><td class="chg">Send List People</td><td>List People Send</td></tr>
<tr><td class="left"><span class="task">so we tried a lot of different things and reddit keeps banning us, idea with glm and kim i and find a path forward so we can get more eye balls on resumehog.</span></td><td>Find glm Resumehog</td><td class="chg">glm forward Resumehog</td><td class="chg">Find Path Resumehog</td><td>Resumehog Reddit Forward</td></tr>
<tr><td class="left"><span class="task">it seems in this mode i cannot launch a campaign</span></td><td>Cannot launch Mode</td><td class="chg">Launch mode Seems</td><td class="chg">launch Mode Seems</td><td>Launch Campaign Mode</td></tr>
<tr><td class="left"><span class="task">ok lets save our progress to gitea. after that write a handoff document on how to implement dms between members.</span></td><td>Save dms Handoff</td><td class="chg">Save Document Handoff</td><td class="chg">Save Document Handoff</td><td>Git Handoff Gitea</td></tr>
<tr><td class="left"><span class="task">so we recently uploaded images to ec2; want to include them in the image search library</span></td><td>Image search Recently</td><td class="chg">Image Search Recently</td><td class="chg">Image Search Recently</td><td>Image Search Recently</td></tr>
<tr><td class="left"><span class="task">search the latest gemini docs and find an appropriate model</span></td><td>docs search Appropriate</td><td class="chg">Search model Appropriate</td><td class="chg">docs model Appropriate</td><td>Gemini Model Appropriate</td></tr>
<tr><td class="left"><span class="task">toolbox should have been on the left column, in the same column as the floating feedback instead of the main flow of the app. this is still too bulky. lock details once selected or…</span></td><td>revoke toolbox Redeploy</td><td class="chg">Toolbox Issues Redeploy</td><td class="chg">Toolbox Issues Redeploy</td><td>Fix Toolbox Redeploy</td></tr>
<tr><td class="left"><span class="task">can you run a web search and see if apple will allow me to have an app on the app store that lets people sync phone video and contacts etc cause i dont wanna pay for extra cloud st…</span></td><td>Web App People</td><td>Web App People</td><td>Web App People</td><td>App Store People</td></tr>
<tr><td class="left"><span class="task">can you get me the link for marketing request form?</span></td><td>get request Link</td><td>get request Link</td><td>get request Link</td><td>Marketing Request Link</td></tr>
<tr><td class="left"><span class="task">what can i do to improve SEO for resumehog.com?</span></td><td>SEO com Resumehog</td><td>SEO com Resumehog</td><td>SEO com Resumehog</td><td>SEO Improvement Resumehog</td></tr>
<tr><td class="left"><span class="task">this app could use an ui/ux overhaul and a comprehensive bug sweep and bug fixes. can you help me?</span></td><td>ui bug Overhaul</td><td>ui bug Overhaul</td><td>ui bug Overhaul</td><td>UI Overhaul Bug</td></tr>
<tr><td class="left"><span class="task">sorry you got cutoff please continue where you left off</span></td><td>Left continue Off</td><td>Left continue Off</td><td>Left continue Off</td><td>Continue Cutoff Left</td></tr>
<tr><td class="left"><span class="task">dont blindly follow the project notes, also consult official docs from tiktok</span></td><td>docs project Official</td><td>docs project Official</td><td>docs project Official</td><td>TikTok Docs Official</td></tr>
<tr><td class="left"><span class="task">❯ i want to send my premium users regular weekly digests from the week's top blog posts. but i want</span></td><td>send top Regular</td><td>send top Regular</td><td>send top Regular</td><td>Premium Users Regular</td></tr>
<tr><td class="left"><span class="task">ps: also consult gemini on html css and aesthetics but dont forget to mention the congruence requirement.</span></td><td>css html Congruence</td><td>css html Congruence</td><td>css html Congruence</td><td>HTML CSS Congruence</td></tr>
<tr><td class="left"><span class="task">new variable: i have a connection in the industry. he is experienced. how can i leverage him, he is a busy human so even for him to weigh in we need to have a working website with …</span></td><td>Variable connection Onboarding</td><td>Variable connection Onboarding</td><td>Variable connection Onboarding</td><td>Leverage Connection Onboarding</td></tr>
<tr><td class="left"><span class="task">amber walnut just did a plug but it is not showing up in the recent comments section</span></td><td>Showing comments Amber</td><td>Showing comments Amber</td><td>Showing comments Amber</td><td>Amber Walnut Comments</td></tr>
<tr><td class="left"><span class="task">Images have been uploaded to S3 from our physical hard drive. Thousands of auction images since 2015. render it in render copyable tool. keep it under 150 words.</span></td><td>S3 Images Copyable</td><td>S3 Images Copyable</td><td>S3 Images Copyable</td><td>Auction Images Copyable</td></tr>
<tr><td class="left"><span class="task">unless there is a url in users resume there is no need to put that link in there</span></td><td>unless url Users</td><td>unless url Users</td><td>unless url Users</td><td>Resume Links Unless</td></tr>
<tr><td class="left"><span class="task">with the teaser, per the handoff's "accepting the loss for the time being."</span></td><td>Per Time Handoff</td><td>Per Time Handoff</td><td>Per Time Handoff</td><td>Teaser Handoff Per</td></tr>
<tr><td class="left"><span class="task">once you are done making changes, commit push and deploy live, approve the qa and prod gates at circle-ci. you have full authorization to do what you need.</span></td><td>Deploy push Commit</td><td>Deploy push Commit</td><td>Deploy push Commit</td><td>Deploy Live Commit</td></tr>
<tr><td class="left"><span class="task">ok lets start with 3, implement, review, then deploy to prod</span></td><td>Deploy ok Implement</td><td>Deploy ok Implement</td><td>Deploy ok Implement</td><td>Implement Review Deploy</td></tr>
<tr><td class="left"><span class="task">i am on destop which sidebar button i dont see one</span></td><td>Button sidebar Destop</td><td>Button sidebar Destop</td><td>Button sidebar Destop</td><td>Sidebar Button Destop</td></tr>
<tr><td class="left"><span class="task">my design is a bit dated get kimi k3 to overhaul the design</span></td><td>k3 design Dated</td><td>k3 design Dated</td><td>k3 design Dated</td><td>Overhaul Design Dated</td></tr>
<tr><td class="left"><span class="task">so i get this error when i try to use the agent in IDE: test</span></td><td>Error IDE Test</td><td>Error IDE Test</td><td>Error IDE Test</td><td>Agent Error Ide</td></tr>
<tr><td class="left"><span class="task">can you outline a plan for me in markdown and save in this repo we are currently working in?</span></td><td>Save markdown Repo</td><td>Save markdown Repo</td><td>Save markdown Repo</td><td>Markdown Plan Repo</td></tr>
<tr><td class="left"><span class="task">alright buddy scope this out in a handoff doc. we shall tackle it in new session.</span></td><td>Session scope Handoff</td><td>Session scope Handoff</td><td>Session scope Handoff</td><td>Handoff Doc Alright</td></tr>
<tr><td class="left"><span class="task">do a /code-rev no need for kimi do it yourself; but you may consult glm 5.2 via porkicoder-consult</span></td><td>glm code Yourself</td><td>glm code Yourself</td><td>glm code Yourself</td><td>Code Rev Yourself</td></tr>
</tbody>
</table>
<p class="note">115 titles. Secret and employer-identifying rows were deleted, not masked.</p>
<h2>What each technique was trying to fix</h2>
<table>
<thead>
<tr>
<th>Diagnosis</th>
<th>Technique</th>
<th>Lesson</th>
</tr>
</thead>
<tbody>
<tr>
<td class="left">Wrong title dialect (SO-ish two words)</td>
<td class="left">Judged-SFT from FLAN / 6 teachers</td>
<td class="left">Works a little. Labels beat pretrain. Diminishing after ~1k–4k rows.</td>
</tr>
<tr>
<td class="left">Beam still prefers pin after SFT</td>
<td class="left">Contrastive hinge vs pin action</td>
<td class="left">Likelihood gap ≠ decode. Beam-4 still writes the old two words.</td>
</tr>
<tr>
<td class="left">Need synonyms / off-page words</td>
<td class="left">Break on-page constraint</td>
<td class="left">35M hallucinates. On-page is grounding, not the limiter.</td>
</tr>
<tr>
<td class="left">Glue third word is junk (<code>Far</code>, <code>Tell</code>, <code>Few</code>)</td>
<td class="left">Swap SO-IDF table for terminal-IDF</td>
<td class="left">Wrong table. Third word is load-bearing (no-fill −0.80) but this IDF is worse.</td>
</tr>
<tr>
<td class="left">Good title is already in a list; beam does not circle it</td>
<td class="left">Local ranker / proposer</td>
<td class="left">Oracle +0.4 to +0.6 if the list has teachers or lucky retrieved. Picker cannot find those without teachers. File-stem proposer cannot fire on redacted Terminal-board.</td>
</tr>
<tr>
<td class="left">Short yes/ok prompts add label noise</td>
<td class="left">Retrain 6t recipe on ≥10-word tasks only</td>
<td class="left">Tied with 6t at n=1000. Keep the filter; do not swap the ship checkpoint. At runtime, do not rename the tab when the new prompt is under 10 words.</td>
</tr>
<tr>
<td class="left">Need more data</td>
<td class="left">+2,616 local Claude/Codex/Grok tasks, 15,696 labels</td>
<td class="left">Mined and labeled. Not packed/trained. Unlikely to beat FLAN 7.11 by scale of the same SFT.</td>
</tr>
</tbody>
</table>
<h2>Still open (do these, not the dead list)</h2>
<div class="card warn">
<ol>
<li><span class="status open">open</span> <b>SO-centroid filters.</b> Keep the winning SO-IDF table. Ban junk fills (<code>tell</code>, <code>few</code>, <code>also</code>, <code>them</code>). Prefer leftover file-like tokens. Mac-only glue packet.</li>
<li><span class="status open">open</span> <b>Skip third word only when residual is weak.</b> Blanket no-fill already lost. Threshold must be preregistered. Same raw 6t/pin titles.</li>
<li><span class="status open">open</span> <b>Decode stop-list</b> so beam cannot start with <code>so</code> / <code>we</code> / <code>can</code> / <code>got</code>. Does not need Ada.</li>
<li><span class="status open">open</span> <b>File-tab proposer</b> on a split that still has <code>[Attached]</code> (pilot-like), not redacted terminal-dev. Product-relevant; not the current 300-row bar.</li>
<li><span class="status keep">done</span> <b>≥10-word train filter.</b> Tried. Tied with 6t at n=1000. Standing pack rule + runtime “keep old title if prompt < 10 words.”</li>
<li><span class="status open">open</span> <b>+2,616 judged-SFT pack</b> only if we want another tenth on the same SFT curve. Cache-only trainers. Do not expect 7.11.</li>
</ol>
</div>
<h2>Do not put back on the plate</h2>
<ul>
<li>LR rescue of session-3 GRPO.</li>
<li>Self-distill K=24, document-order extractive miner, span-pretrain→title-CE, Gemini-title CE from scratch.</li>
<li>Quantize FLAN as our model. FLAN is the benchmark.</li>
<li>Train on terminal-dev / holdout kimi (leaks 8.04 into the bar).</li>
<li>Open the sealed SO 1000.</li>
<li>Rerun terminal-IDF centroid or unconstrained decode as “maybe this time.”</li>
</ul>
<h2>Artifacts</h2>
<p class="note">
Session 4 HTML: <code>session4_results.html</code>.
Session 5: <code>tab_namer_quality_roadmap/session5_findings.md</code>, keeper <code>output/s5_train_a/6t_argmax_ge6/final</code>.
Session 6: <code>session6_findings.md</code>, packs <code>output/s6_build/pack_contrastive</code> / <code>pack_free</code>.
Session 7 glue: <code>session7_glue_findings.md</code>, <code>output/s7_glue/</code>.
Ranker: <code>output/s7_ranker/</code>, sample <code>output/s7_build/sample400/</code>.
Proposer: <code>session8_proposer_findings.md</code>, <code>scripts/eval_local_proposer.py</code>.
min10: <code>session8_min10_findings.md</code>, <code>output/s8_long10/</code>.
min5: <code>session8_min5_findings.md</code> 176k test: <code>2026-08-16_g300k-flashlite-test.html</code>., <code>output/s8_min5/</code>, sha <code>e31cbe0d…</code> (tied with 6t).
Pin sha <code>0f45a1fe…</code>. 6t sha <code>2ea02c13…</code> (ship). min10 sha <code>9a1efd16…</code> (tied, not a replacement).
Ship folder: <code>porkr1/6t_argmax_ship_package/</code> (fp32 namer + frozen centroid v2 + beam-4/Hybrid A runtime + ONNX export script).
</p>
<footer>Product metric: blinded Gemini 3.5 Flash-Lite on Hybrid-A-glued titles, two decimals. Exact match has no weight. Miner → one packer → cache-only trainers. Update this page when an open item is tried; do not start a new paper for a negative glue or ranker packet. Published 16 August 2026 to porkicoder.com/research/tab-namer-techniques.html · same article in porkihomepage, porkr1/research, and this repo (hrefs only differ).</footer>
</main>
</body>
</html>
|