Computes init damage, init-quality spread and FVT−FOCUS gap per domain with paired bootstrap intervals, the hypothesis bins and gates G1 and G4.

https://github.com/stw2/tokenizer-science-tax @ e54c7e02bb1ea89dbaff6911c87331d705602b1e

Reference checked 2026-09-14 13:12 UTC. Later commits, branches or plan changes do not retarget this attempt.

• experiments/E07-embedding-init/scripts/04_analyze.py

Command

python scripts/04_analyze.py

Working directory

experiments/E07-embedding-init

Configuration paths

None

Parameters

bootstrap resamples 1000; seed 20260705; G1 tolerance 0.005 bits per byte; run time not recorded, reported at the results commit

Environment

3.12.13; numpy on CPU; safetensors for the embedding matrices; Apple silicon, 128 GB unified memory

Output directory

Not recorded

Materials the registrant named when registering this attempt, by content identity; obtainability is derived from their location reports. Nothing is fetched or verified.

• bpb_e7.jsonl · analysis input · M363 bpb_e7.jsonl · raw output · Download
• perdoc/ · analysis input · M364 perdoc/ · raw output · Download
• models/fvt · model · M357 models/fvt · raw output · Ask the reporter
• models/focus · model · M358 models/focus · raw output · Ask the reporter
• Qwen2.5-1.5B · tokenizer · M124 Qwen2.5-1.5B · checkpoint · Download
• APT4 tokenizer · tokenizer · M3 APT4 tokenizer (Bielik-PL-11B-v3.0-Instruct) · tokenizer · Ask the reporter
• bpb_armA.jsonl · reference values · M147 bpb_armA.jsonl as originally written · raw output · Rebuild via attempt