CodeSOTA · Experiment · E-0005Portable protocol · public evidence graph
Why this run exists

OCRBench-EN sample-scope audit

Determine which stored OCR runs can be compared and which need a controlled reproduction.

Experiment
E-0005
Status
completed
Outcome
inconclusive
Forks
4 public or local starts recorded
Owner
Kacper Wikiel

Method before metrics.

Inspect total samples, valid samples, latency and error metrics before treating any two runs as direct evidence of model differences.

Hypothesis
H-0006 · OCR quality estimates from a 100-sample subset are not stable enough to stand in for a 1,000-sample evaluation.
Hypothesis
H-0007 · The valid/total sample ratio is necessary evidence for interpreting OCR run metrics.

The comparison contract.

A reproduction matches this protocol. A fork changes it and declares the deviation.

dataset
ocrbench-en
independent Variable
model and sample scope
benchmarks
accuracy · exact accuracy · CER · WER · latency

What this experiment produced or evaluated.

Produced
Tesseract 5 · OCRBench-EN · 100 samples
external-run:ER-0002
Produced
dots.ocr · OCRBench-EN · 100 samples
external-run:ER-0003
Produced
PaddleOCR-VL · OCRBench-EN · 1,000 samples
external-run:ER-0004

Change the evidence, not just the discussion.

Start locally, inspect the portable manifest, and sign in only if you choose to publish.

Attach a run to this protocol.

Record the measurement and its provenance. New evidence is unverified and pending review until CodeSOTA checks it.

Stable, portable, attributable.

Stable ID
E-0005
Visibility
public
Created
2026-08-20
Updated
2026-08-20