Files
the-nexus/eval-results.json
Alexander Whitestone 25cca98636
Some checks failed
CI / validate (pull_request) Failing after 7s
CI / auto-merge (pull_request) Has been skipped
feat: eval harness results as 3D bar chart — baseline vs fine-tuned (#283)
Adds a holographic 3D bar chart floating beside the platform showing
eval harness metrics comparing baseline vs fine-tuned model performance.

- eval-results.json: data file with 5 metrics (accuracy, F1, BLEU,
  recall, precision) for baseline and fine-tuned model scores
- buildEvalBarChart(): constructs BoxGeometry bars (blue=baseline,
  green=fine-tuned) with edge glow, y-axis gridlines, metric labels,
  title banner, and colour legend using canvas textures
- loadEvalResults(): fetches eval-results.json, falls back to stub
- Gentle vertical float animation in the main animate loop
- Glow PointLight inside chart group for ambient atmosphere

Fixes #283

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-03-24 00:57:51 -04:00

11 lines
379 B
JSON

{
"title": "Eval Harness",
"metrics": [
{ "name": "Accuracy", "baseline": 0.72, "finetuned": 0.89 },
{ "name": "F1", "baseline": 0.68, "finetuned": 0.85 },
{ "name": "BLEU", "baseline": 0.41, "finetuned": 0.63 },
{ "name": "Recall", "baseline": 0.65, "finetuned": 0.83 },
{ "name": "Precision", "baseline": 0.71, "finetuned": 0.87 }
]
}