Pete Nicholas voice repair experiments

Post-hoc transforms

Prefix rewrite, seed-intro, intro-brief, section-brief, hybrid full-brief opening plus section body, plain brief, full brief, slop cleanup, and candidate-pool selection.

Candidate selection

Portfolio selection scans raw rewrites, runs cheap deterministic gates, then confirms same-author and quality only for plausible candidates.

External worktrees

Diagnostic scalpel helped modestly. Multi-model bakeoff is topic-dependent. Drift forcing, moves rewrite, and aggressive rule polish hurt or did nothing.

Next tool

Generate from topic and factual brief with the seed corpus as the voice target, then gate at batch level for same-author and repeated fingerprints.

BC/AD accepted repair

Hybrid full-brief opening plus section-brief body. This is the cleanest post-hoc win: 0/3 source votes to 3/3 candidate votes.

• source-articles / bc-and-ad-how-the-christian-calendar-became-a-quiet-daily-confession-of-a-contro.md
• same-author-after / bc-and-ad-how-the-christian-calendar-became-a-quiet-daily-confession-of-a-contro.md
• same-author-raw_rewrites / bc-and-ad-how-the-christian-calendar-became-a-quiet-daily-confession-of-a-contro.md
• same-author-raw_rewrites / hybrid-brief-section-bc-and-ad-how-the-christian-calendar-became-a-quiet-daily-confession-of-a-contro.md

Tithing accepted repair

Brief rewrite passed repeated same-author voting, but the word ratio is close to the lower bound, so it is evidence rather than a production-ready default.

• source-articles / christian-tithing-should-the-ten-percent-law-apply-to-christians-and-what-the-ne.md
• same-author-after / christian-tithing-should-the-ten-percent-law-apply-to-christians-and-what-the-ne.md
• same-author-raw_rewrites / christian-tithing-should-the-ten-percent-law-apply-to-christians-and-what-the-ne.md
• same-author-raw_rewrites / brief__anthropic-claude-sonnet-4.6__christian-tithing-should-the-ten-percent-law-apply-to-christians-and-what-the-ne.md

Portfolio report

The portfolio selector prevents bad repairs but still tops out at 3/10 on the first-ten sample.

• reports-and-code / PORTFOLIO_REPORT.md

Iteration notes excerpt

# Same-author lift iteration notes

Date: 2026-05-18

Goal: raise Pete outputs to at least 52/61 `same_author_llm` passes while
preserving voice and avoiding new repetitive tells.

## Tooling added

- `experiments/same_author_lift/run.py`
  - cached same-author evaluation
  - optional `single` judge mode matching `simple_writer.pipeline.character_check`
  - optional `multi` judge mode using all three seed excerpts
  - opener-prefix repair that writes candidates under `experiments/same_author_lift/after/`
  - rejection gates for same-author failure, word-count loss, drift rise, slop rise,
    scaffold increase, and quality-guard failures
  - rejected candidates preserved under `experiments/same_author_lift/rejected/`

## Metric finding

The production `same_author_llm` check is weak as an optimization target. It
compares every article to only the second seed representative excerpt:

> When we feel hard-pressed, God's grace fills us up...

That excerpt is a 68-word devotional paragraph. The generated articles are
longer discursive essays. Optimizing directly against this single excerpt would
push the articles toward devotional paragraph cadence and away from the broader

Toolbox excerpt

# Same-author fix toolbox

Date: 2026-05-18

Goal: reach at least 52/61 Pete articles passing `same_author_llm` while
preserving author voice and avoiding repeated batch tells.

## Current result

The best post-hoc path is still only 3/10 on the first-ten sample:

- baseline repeated same-author vote: 2/10
- hybrid full-brief opening + section-brief body: 3/10
- portfolio selection across existing raw candidates: 3/10

That makes the answer fairly sharp: post-hoc repair is not the complete fix.
It is a safety net for individual articles, not the production strategy.

## Tools that help

1. `experiments/same_author_lift/run.py`
   - runs single/multi/any same-author judges
   - generates candidate transforms
   - gates candidates by same-author, word count, slop, drift, scaffold hits,
     and hostile quality guard
   - useful transforms: `hybrid-brief-section`, `section-brief`,
     `plain-brief`, `brief`, `slop-cleanup`, `candidate-pool`

2. `experiments/same_author_lift/portfolio.py`
   - selects the best existing raw candidate per article