FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteTie8 / 162 · bc30ba197adadd03

Metagenomic Analysis using Phylogenetic Placement -- A Review of the First Decade

Quantitative biology · 2202.03534v2

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields, 51 of 81 left-column positions preferred the Evidence version. Ties, pairs where both versions were unusable and incomplete reviews remain in the denominator.

51Evidence version preferred
÷
81All positions in this group
=
63%Evidence preference in this group

Source contribution is a separate review: 16 of 22 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

Metagenomic Analysis using Phylogenetic Placement -- A Review of the First Decade

4 Conclusion and Outlook

…created by, and shared in, research communities that investigate the same group(s) of organisms. This would not only yield obtaining high-quality reference trees trivial, but would also immensely increase the comparability across studies, as well as their reproducibility. Consequently, we would highly encourage such collaborations, and the public sharing of (perhaps even versioned instances of) gold-standard reference trees. Notably, for some environments, first efforts into this direction have already been undertaken Berney2017 ; DelCampo2018 ; Rubinat-Ripoll2019 ; Rajter2021 ; Rajter2021a .

Furthermore, as mentioned, there is a lack of established methods that evaluate placement quality in a standardized and meaningful way. In particular, robust metrics are missing to distinguish the case where reference sequences of known species are missing from the tree from the case where the placed data actually contains yet undescribed species. A classification based on the LWR and pendant length of the placement locations might offer a solution here.

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: Tie

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Beyond such classification, the field would benefit from a shared benchmark that reports placement quality under a common set of reference trees and query datasets, so that methods can be compared across studies rather than within individual evaluations.

Passages supplied to the Evidence version

pplacer: linear time maximum-likelihood and Bayesian phylogenetic placement of sequences onto a fixed reference tree

The posterior probability has a clear statistical interpretation as the probability that the fragment is correctly placed on that edge, assuming the reference tree, the alignment, and the priors on pendant branch length. Because the reference tree is fixed, direct numerical quadrature over the likel…

Read full passage excerpt

The posterior probability has a clear statistical interpretation as the probability that the fragment is correctly placed on that edge, assuming the reference tree, the alignment, and the priors on pendant branch length. Because the reference tree is fixed, direct numerical quadrature over the likelihood function can be performed to obtain the posterior probability rather than relying on Markov chain Monte-Carlo procedures as is typically done in phylogenetics [ 44 , 45 ] . In maximum likelihood (ML) mode, pplacer evaluates the “likelihood weight ratio,” [ 39 ] i.e. the ML likelihood values across all placement locations normalized to sum to one.

No Evidence

Same model and position · no retrieved passages · first output

Beyond such classification, the field would benefit from a shared benchmark that reports placement quality under a common set of reference trees and query datasets, so that methods can be compared across studies rather than within individual evaluations.

Blind model judgment

The blind model judge did not prefer either output.

Read the original model judgment

Both outputs are identical. They provide a coherent, well-grounded continuation that logically extends the draft's argument: the draft discusses missing standardized methods for evaluating placement quality and proposes LWR/pendant length classification as a potential solution; both outputs add a complementary and natural next step (shared benchmarks for cross-study comparison) without inventing methods, results, or unsupported claims. No citation markers are generated. The continuation is consistent with the draft's scope and the source's context about pplacer's LWR metric, without overreaching into unsupported territory.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.