assets/captioner-blind-read.svg

<svg xmlns="http://www.w3.org/2000/svg" width="1200" height="420" viewBox="0 0 1200 420" role="img">
<style>
text{font-family:'Cascadia Mono','JetBrains Mono','SFMono-Regular',Menlo,Consolas,monospace;fill:#d6dbe4}
.eyebrow{font-size:11px;letter-spacing:.18em;fill:#7c8596}
.title{font-size:20px;letter-spacing:.06em}
.subtitle{font-size:12px;fill:#7c8596}
.label{font-size:13px}
.value{font-size:13px;fill:#d6dbe4}
.big{font-size:34px;letter-spacing:.04em}
.muted{font-size:11px;fill:#7c8596}
.tiny{font-size:9px;fill:#7c8596}
.cell{font-size:10px}
.cellbold{font-size:10px;font-weight:600}
.dim{fill:#4a5060}
.chip{font-size:11px;letter-spacing:.08em;fill:#5ccfe6}
.amber{fill:#e0a458}
</style>
<rect width="1200" height="420" rx="14" fill="#16161e"/>
<text x="40" y="44" class="eyebrow">CAPTIONER // BLIND READ</text>
<text x="40" y="76" class="title">Read blind against the audio, the GGUF was never the wrong one</text>
<text x="40" y="100" class="subtitle">Six clips · original int8 (transformers) against Q4_K_M + Q8_0 proj (llama.cpp) · both captions shown in random order, judged by ear before the key was opened</text>
<rect x="40" y="116" width="1120" height="1" fill="#262836"/>
<text x="40" y="194" class="big" style="fill:#d6dbe4">4</text>
<text x="74" y="194" class="label" style="fill:#d6dbe4">rated the same</text>
<text x="420" y="194" class="big" style="fill:#5ccfe6">2</text>
<text x="454" y="194" class="label" style="fill:#5ccfe6">GGUF preferred</text>
<text x="800" y="194" class="big" style="fill:#7c8596">0</text>
<text x="834" y="194" class="label" style="fill:#7c8596">original preferred</text>
<rect x="40" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#262836" stroke-width="1"/>
<text x="54" y="258" class="eyebrow">CLIP 01</text>
<text x="54" y="281" class="label" style="fill:#d6dbe4">tie</text>
<rect x="230" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#262836" stroke-width="1"/>
<text x="244" y="258" class="eyebrow">CLIP 02</text>
<text x="244" y="281" class="label" style="fill:#d6dbe4">tie</text>
<rect x="420" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#5ccfe6" stroke-width="2"/>
<text x="434" y="258" class="eyebrow">CLIP 03</text>
<text x="434" y="281" class="label" style="fill:#5ccfe6">GGUF preferred</text>
<rect x="610" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#262836" stroke-width="1"/>
<text x="624" y="258" class="eyebrow">CLIP 04</text>
<text x="624" y="281" class="label" style="fill:#d6dbe4">tie</text>
<rect x="800" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#262836" stroke-width="1"/>
<text x="814" y="258" class="eyebrow">CLIP 05</text>
<text x="814" y="281" class="label" style="fill:#d6dbe4">tie</text>
<rect x="990" y="238" width="172" height="76" rx="6" fill="#1d1e27" stroke="#e0a458" stroke-width="2"/>
<text x="1004" y="258" class="eyebrow">CLIP 06</text>
<text x="1004" y="281" class="label" style="fill:#5ccfe6">GGUF preferred</text>
<text x="1004" y="302" class="muted amber">[contradiction]</text>
<text x="40" y="348" class="muted">One contradiction in six, and it came from the original: on clip 06 it described a synth pad and a chorus the track does not have.</text>
<text x="40" y="386" class="muted">Captions shown in random A/B order without engine names; the key was opened after the verdicts. Six clips is a read, not a statistic. Full transcript: see the</text>
<text x="40" y="402" class="muted">evidence report linked in the README.</text>
</svg>