SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2608.26762 · 2026

Equal Ranking Quality, Different Decisions: Training Order-Consistent LLM Scorers

Navid Rekabsaz, Markus Frohmann, Mahdiyar Alavi, Elizabeth Lingg

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 16 functions out of this paper's own repositories and ran 14 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
thomsonreuters/presentation-dependence canonical 14 of 16
FunctionStatusWhere it lives
canonical_method_key Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/terminology.py
code served (permissive licence) · get_code("0b18e643c678e842")
detect_surface Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/amortization.py
code served (permissive licence) · get_code("52ffd26e79d3940a")
hierarchical_bootstrap_ci Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/metrics.py
code served (permissive licence) · get_code("9d94581fe52f5f71")
load_b1_scores Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/copresence_scores.py
code served (permissive licence) · get_code("a4d7d5f474a39ce8")
load_b20_order_averaged_scores Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/copresence_scores.py
code served (permissive licence) · get_code("516a0a73bb992598")
model_signature Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/bundle.py
code served (permissive licence) · get_code("b7956f205862a59d")
normalize_recipe_token Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/amortization.py
code served (permissive licence) · get_code("d052ee9cfc9e1da1")
paired_bootstrap_ci Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/metrics.py
code served (permissive licence) · get_code("d0479097f346085d")
paper_method_label Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/terminology.py
code served (permissive licence) · get_code("68ba74adfebf2269")
published_presentations Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/context_decomposition.py
code served (permissive licence) · get_code("cb864446516f5c03")
rank_ids_from_scores Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/context_decomposition.py
code served (permissive licence) · get_code("12b8497af8a4338e")
shard_qids Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/context_decomposition.py
code served (permissive licence) · get_code("64a0c08a1d67fc27")
stem_of Ran thomsonreuters/presentation-dependence/src/presentation_dependence/analysis/amortization.py
code served (permissive licence) · get_code("5b507438e60abedc")
summarize_score_pairs Ran thomsonreuters/presentation-dependence/src/presentation_dependence/eval/copresence_scores.py
code served (permissive licence) · get_code("75bc657ab9a4ae00")
assert_shared_model Not yet run thomsonreuters/presentation-dependence/src/presentation_dependence/eval/bundle.py
code served (permissive licence) · get_code("6eeaae096d235160")
run_bundle Not yet run thomsonreuters/presentation-dependence/src/presentation_dependence/eval/bundle.py
code served (permissive licence) · get_code("121652f9d430c54a")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Rerankers, reward models and multi-document QA scorers score candidate documents or responses in one LLM prompt, so each score depends on their order. Such scorers are selected on ranking quality, but their scores determine a decision: what a score threshold retains, a reader answers, or a preference model selects. However, equal ranking quality does not imply equal decisions: on passage reranking, five trained scorers within 0.010 nDCG@10 retain sets that overlap by only 0.66-0.84 when reordered. A published reranker takes the highest retained-set F1 in our comparison and still overlaps by only 0.667. No prompt-time change we test removes that order dependence: the only one that gains ranking quality leaves all three decisions unchanged. Order-consistency SFT (OC-SFT) attenuates it in the weights, training a candidate's score not to depend on the order. It holds ranking quality and leads every decision-stability measure among trained scorers on all three tasks: it flips the reader's answer on 0.125 of permutation pairs against 0.149-0.164 for three other objectives that target order. It is more stable than order-averaged distillation on 12 base models, and one OC-SFT permutation retains sets that overlap more than ten averaged off-the-shelf permutations. A comparison should therefore report what a threshold retains and a reader answers, not ranking quality alone. Code is available at github.com/thomsonreuters/presentation-dependence.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2608.26762")
get_code_for_paper("2608.26762")
have("2608.26762")

Connect an agent — have() is free.