> Canonical page: https://posttrainllm.com/experiments/sql-hygiene-dpo-refanchored

Experiment · sql

# Hygiene DPO reference-anchored retry (2026-07-11)

Reference-anchored DPO is the correct fix for the SimPO collapse Retained outcome: failed.

**Outcome**
 failed

**Family**
 sql

**Evidence confidence**
 exact

**Record**
 training run

## Evidence readout

composed execution `0.860 -> 0.900` (+0.040, exact `0.840 -> 0.860`)

## What the attempt taught

Reference-anchored DPO is the correct fix for the SimPO collapse

## Why it stopped or stayed bounded

DPO pressure too low to change output FORMAT — 41/50 raw

## Method and scope

- Experiment family: sql.
- Record kind: training run.
- Objective: output format.
- Methods: dpo, lora.
- Base models: qwen3-0.6b.
- Recorded data rows: 108.

## Disposition

~~Higher-pressure reference-anchored DPO...~~ Done 2026-07-11 (next entry).

This retained disposition is historical evidence, not authorization to restart the experiment. A new run needs a fresh question, frozen evaluator, explicit resource budget, and scoped tracking issue.

## How to read this dossier

This canonical page separates the retained repository decision from the material that informed it. A source may describe an external claim, historical measurement, local observation, or planned procedure; those evidence classes are not interchangeable. Follow the provenance links before reusing a number or method.

Status and disposition describe what PostTrainLLM retained when this record was indexed. They are not a live product promise, a newly run benchmark, or permission to restart historical work. Unknown or unmeasured fields remain unknown rather than being treated as zero.

To reuse the record, first name the exact claim you need and trace it to the linked source. Then check whether the original environment, model revision, data split, evaluator, hardware, and budget match the proposed use. If they do not, treat the record as a hypothesis or design reference and run the smallest fresh comparison that can falsify it. Preserve negative outcomes and regressions beside any improvement; a local win on one slice does not silently become a general capability claim.

## Source provenance

The normalized record comes from docs/attempts.json . The links below are the tracked evidence and explanatory sources preserved with the record.

- [docs/NEXT.md ↗](https://posttrainllm.com/docs/NEXT/)
- [docs/attempt-ledger.md ↗](https://posttrainllm.com/docs/attempt-ledger/)

[Read the Markdown equivalent →](https://posttrainllm.com/experiments/sql-hygiene-dpo-refanchored.md)

Related evidence

- evidence [Primary proof](https://posttrainllm.com/docs/NEXT/)
- recipe [Applicable recipe](https://posttrainllm.com/recipes/sql-lineage)
- learning path [Learning path](https://posttrainllm.com/learn/paths/post-training)

 Canonical record experiment:sql-hygiene-dpo-refanchored