Skip to content

docs: ara-diac-tiny clean-label verdict — from-scratch collapses (74.68) #72

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ronaldtse merged 1 commit into from
Aug 29, 2026
Merged

docs: ara-diac-tiny clean-label verdict — from-scratch collapses (74.68) #72

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
14 changes: 14 additions & 0 deletions docs/PUBLICATION-NOTES.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,3 +116,17 @@ data, not same capacity law. Two lessons for the experiments section:
inferred from filenames; (2) a reproduced number is only evidence of
reproducibility when the input pipeline is versioned. run-005
(regenerating labels live from the r6 teacher) is the honest test.

## ara-diac-tiny run-005: the law holds (2026-08-29)

The clean-label rerun scores 74.68 vs the poisoned run's 83.08 (gate
3.07): corruption explained 8pp, capacity explains the rest. The
"pretrained backbone or nothing" claim now has Arabic evidence
matching the Thai ablation, closing the retraction arc
(verdict -> retraction -> poisoned rerun -> clean rerun). For the
paper: report the pair (83.08, 74.68) as the label-quality and
capacity bounds of the same from-scratch rung, with the teacher
reproducing 1.3205 across all three measurements as the harness
control. Next rung on the frontier: stitch-down from ByT5-small
pretraining (ridge-fit width bridge), which changes init, not data or
capacity alone.
22 changes: 22 additions & 0 deletions docs/RESULTS.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,6 +195,28 @@ Verdict: run-004 says nothing about from-scratch capacity; the
retraction stands. **run-005** (fresh teacher labels, no snapshot,
2026-08-29) is the actual clean-label falsification test — in flight.

## ara-diac-tiny run-005 — clean-label verdict: from-scratch collapses (2026-08-29)

The falsification test the Aug-24 retraction called for: same 33M
from-scratch student (d384, 8+8), labels regenerated live from the r6
teacher (11,793 units, no snapshot), 4,422 steps / 3 epochs, final CE
~0.9. Windowed gate, 300 SadeedDiac-25 paragraphs, teacher reproduces
1.3205:

| Model | DER-CE (300) |
|---|---|
| Teacher (r6) | 1.3205% |
| Tiny, mojibake labels (run-004) | 83.08% |
| **Tiny, clean labels (run-005)** | **74.68% — REJECTED** |

Clean labels recover ~8pp of the collapse — the student learns real
signal — but remains two orders off the <= 3.07 gate. **The
pretrained-backbone law now rests on Arabic evidence as well as
Thai**: from-scratch byte-level students at this width do not work.
The viable path to a sub-100MB browser tier is width reduction FROM a
pretrained ByT5-small (closed-form stitch across widths, microkimi
protocol), not from-scratch training.

## ara-diac-small-1.0 — Arabic client tier (2026-08-24)

Sequence-level KD from the r6 teacher (rababa_arabic_byt5/run-006-morph,
Expand Down
Loading