From ded70c3853f5922f6ad0b6e82383a345e7620d07 Mon Sep 17 00:00:00 2001 From: mARTin-B78 Date: Sat, 19 Sep 2026 20:54:22 +0200 Subject: [PATCH] docs(audiobook): record the rejected scene-cast experiment Co-Authored-By: Claude Sonnet 5 --- TODO.md | 5 +++-- architecture/audiobook-attribution.md | 3 ++- 2 files changed, 5 insertions(+), 3 deletions(-) diff --git a/TODO.md b/TODO.md index 855aa8b..b0853a5 100644 --- a/TODO.md +++ b/TODO.md @@ -345,8 +345,9 @@ Still open (each needs a real decision or new evidence, not more prompt tuning): until the book itself ties it to a name. - [ ] Look-alike main characters in one scene (Marcian/Zerwas/Lancorian) remain the biggest real error class (~1/5 of spoken words). Tried and rejected: reasoning mode (25× slower, no gain), larger/smaller - windows, tag-based overrides, reasoning adjudication of disputes (made disputed lines worse). Untried: - an explicit scene-cast constraint (characters present in the scene) built from the text. + windows, tag-based overrides, reasoning adjudication of disputes (made disputed lines worse), an explicit + scene-cast hint (names mentioned in the passage: +0.3 points, noise). What remains is mostly model + capacity plus reference noise; the useful lever left is human review guided by the agreed/disputed flags. --- diff --git a/architecture/audiobook-attribution.md b/architecture/audiobook-attribution.md index 441bb2f..ca96f6b 100644 --- a/architecture/audiobook-attribution.md +++ b/architecture/audiobook-attribution.md @@ -132,7 +132,8 @@ lines) but compared against a biased baseline; on the real pipeline it made the so the adjudication step (and the `candidates` request field) was removed. The only reliable signal is agreement itself: agreed lines are right 82 % of the time, contested first answers 67 %. Sampling matters: windows of ~1 200 characters were clearly worse (−4 points) and ~5 000 about equal to -the default, so the default stays. +the default, so the default stays. A scene-cast hint (which known characters are named in the passage or +just before it) gave +0.3 points — noise — and was not adopted. Request additions: `attribute_only` `[ids]` — the other spans are context; unrequested lines are never reported missing.