A Cognitive Benchmark for Code-RAG Retrieval: Part 2 — Why Model Rankings Depend on the Pipeline

When developers enter an unfamiliar project, they rarely search for a specific
file by name. They usually ask about system behavior: where incoming
connections are accepted, which component cleans logs, or how a request travels
between architectural layers.

Code-RAG tries to answer such questions through semantic search. It splits and
indexes the source code, then retrieves the context most closely related to a
developer's query.

The quality of this search is often reduced to the choice of embedding model:
compare several candidates and select the one with the highest metric. In
practice, the result also depends on how the code was split and which retrieval
mode was used.

To study these dependencies, I built a Code-RAG benchmark on the Apache Kafka
4.0.0 broker core, a real polyglot project written in Java and Scala. For
thirty questions about system behavior, I identified the correct files in
advance, allowing me to measure how accurately the retrieval pipeline finds
the relevant code.

The results show that a model ranking exists only within a specific
configuration. Changing the chunking strategy, retrieval mode, or query
phrasing can change both the metric value and the order of models in the
ranking.

1. Experimental Setup

In this part of the study, I compare sixteen embedding models, five chunking
configurations, and three retrieval modes: BM25, vector search, and hybrid
search. Each of the thirty questions was expressed in five forms, ranging from
a natural developer question to a query with inaccurate terminology or a
reference to a neighboring module. The structure of these variants and the
evaluation methodology are described in
Part 1 of the study.

I compared four groups of variables:

Variable	What changed	What it tested
Embedding model	7 local models through Ollama and 9 commercial APIs	How strongly quality depends on the vector representation of the query and code
Chunking	Whole-file indexing and four fixed-size chunks with overlap	How the indexed fragment size affects a particular model
Retrieval mode	BM25_ONLY, VECTOR_ONLY, HYBRID_RRF	Whether lexical search, vector search, or their combination works best
Query phrasing	Natural question, technical query, keywords, inaccurate terminology, and selected cross-module queries	How strongly the result depends on the language of the query

The three retrieval modes work as follows:

Mode	How the ranking is produced
BM25_ONLY	Lucene lexical search. Files rank highly when query terms match terms in the code. No embedding model is used.
VECTOR_ONLY	The query and code fragments are converted into embeddings. Ranking is based on vector similarity, so exact word overlap is unnecessary.
HYBRID_RRF	BM25 and vector search run independently, then their positions are combined using Reciprocal Rank Fusion. RRF uses rank positions rather than directly adding incomparable scores.

The primary metric in this article is recall@10. For a single question, it
equals 1 when the primary correct file appears in the first ten results and 0
otherwise. The final value is the average across thirty questions. For example,
recall@10 = 0.900 means that the correct file appeared in the top ten for 27
of 30 questions.

The model ranking also reports a 95% CI, a 95% bootstrap confidence interval.
To calculate it, I repeatedly resampled the set of questions with replacement
and recalculated recall for each sample. A wide interval means that thirty
questions are insufficient for precisely estimating small differences.
Overlapping intervals are not themselves a formal pairwise test, but they warn
against treating the order of neighboring rows as stable.

The chunking label c500-o100 means fragments of 500 characters with a
100-character overlap. whole-file means that an entire file is indexed as one
fragment.

I did not test the complete Cartesian product of all parameters. Models were
compared under a fixed baseline configuration; local and commercial models were
tested across five chunking configurations; and VECTOR_ONLY was compared
with HYBRID_RRF at the fixed c1500-o200 chunking. The full interaction
between retrieval mode and chunking remained outside the study. BM25 was run as
a single baseline because it does not depend on an embedding model.

2. Conditions for Comparing Models

To compare embedding models, the remaining retrieval-pipeline parameters must
be fixed. Otherwise, it is impossible to tell whether a difference was caused
by the model, fragment size, or retrieval method.

The baseline comparison used natural human questions, HYBRID_RRF, and
c1500-o200 chunking. For each model, I measured the share of thirty questions
for which the correct file appeared in the first ten results.

This ranking compares models under identical conditions, but it does not
describe their quality outside the selected configuration. For example, OpenAI
text-embedding-3-large achieved recall@10 = 0.833 with c1500-o200,
0.900 with the smaller c500-o100 fragments, and 0.433 when files were
indexed whole.

The value 0.833 therefore cannot be treated as an independent property of the
model. It describes one combination of model, chunking, retrieval mode, corpus,
and question set. The baseline ranking is a useful starting point, but it
cannot identify the best configuration without testing the other parameters.

3. The Effect of Chunking

Ideally, code should be split along logical boundaries such as methods, classes,
or other structural units. Structural chunking, however, requires a dedicated
parser for every language.

This study deliberately uses a polyglot Java and Scala project. I therefore
split the code into fixed-size fragments. This is not presented as the optimal
way to index code; it provides a common denominator across languages and makes
it possible to isolate the effect of fragment size.

Every value in the table is recall@10 for natural human questions using
hybrid retrieval. The best observed result for each model is shown in bold.

Model	c500-o100	c1500-o200	c3000-o300	c5000-o500	whole-file	Type
all-minilm	0.733	0.700	0.667	0.700	0.567	local
bge-m3	0.833	0.767	0.767	0.633	0.733	local
granite-embedding	0.800	0.767	0.767	0.633	0.433	local
mxbai-embed-large	0.900	0.833	0.733	0.633	0.433	local
nomic-embed-text	0.733	0.667	0.700	0.633	0.133	local
qwen3-embedding:0.6b	0.800	0.800	0.933	0.833	0.900	local
snowflake-arctic-embed2	0.800	0.800	0.800	0.733	0.733	local
EmbeddingsGigaR	0.767	0.767	0.800	0.789	0.167	commercial
GigaEmbeddings-3B	0.767	0.867	0.833	0.767	0.300	commercial
codestral-embed	0.800	0.900	0.867	0.800	0.467	commercial
mistral-embed-2312	0.800	0.800	0.900	0.800	0.400	commercial
text-embedding-3-large	0.900	0.833	0.867	0.800	0.433	commercial
text-embedding-3-small	0.867	0.867	0.867	0.867	0.433	commercial
voyage-4-large	0.900	0.833	0.867	0.833	0.867	commercial
voyage-code-2	0.933	0.867	0.800	0.800	0.533	commercial
voyage-code-3	0.900	0.867	0.900	0.833	0.833	commercial

Smaller Fragments Helped Most Local Models

For five of the seven local models, c500-o100 produced the highest observed
result. One possible explanation is that a small fragment contains less
unrelated code. Its embedding can describe a local implementation more
precisely, while BM25 benefits from matching specific terms.

The experiment does not establish this mechanism directly. Doing so would
require inspecting retrieved fragments and comparing hybrid and vector-only
search at every chunk size.

Some Models Benefit from Larger Fragments

qwen3-embedding:0.6b achieved its highest result with c3000-o300 and still
reached 0.900 when indexing whole files. Unlike most local models, it retained
quality on larger fragments.

A possible explanation is the model's ability to process longer context. A
larger fragment preserves relationships between methods and their surrounding
class that smaller fragments may lose. A similar pattern appeared for
mistral-embed-2312, EmbeddingsGigaR, and partly for voyage-code-3.

This remains a hypothesis: the experiment measured retrieval outcomes, not the
internal cause of each model's behavior.

Whole-File Indexing Is the Riskiest Choice

With whole-file, results ranged from 0.133 to 0.900. The approach
remained viable for qwen3-embedding, voyage-4-large, and voyage-code-3,
but quality dropped sharply for nomic-embed-text and EmbeddingsGigaR.

The likely explanation is context-window limits and truncation of long files.
Because I did not directly measure truncation by provider tokenizers, this must
also remain a hypothesis.

The matrix does not reveal a universally best fragment size. Instead, it shows
three kinds of behavior:

• models that benefit substantially from small fragments;
• models that require more surrounding context;
• models that remain stable across chunking configurations.

Chunking should therefore be selected together with the embedding model. When
tuning time is limited, c500-o100 is a reasonable starting point, but at
least one larger alternative should also be tested, and whole-file should
not be used without separate validation.

4. The Effect of Retrieval Mode

After choosing how to split the code, the next question is how to retrieve the
relevant fragments. The experiment compared three modes:

• BM25_ONLY matches words in the query against words in the code;
• VECTOR_ONLY compares semantic similarity between embeddings;
• HYBRID_RRF combines the rank positions from BM25 and vector search.

The retrieval-mode comparison used c1500-o200. In an earlier experiment, the
combination c1500-o200 + HYBRID_RRF produced the strongest result available
at the time and became the control configuration for later runs.

The subsequent chunking matrix showed that there is no universally optimal
fragment size. Keeping c1500-o200, however, allowed retrieval modes to be
compared under identical conditions without mixing their effect with a
chunking change.

The full matrix of retrieval modes and chunking configurations was not tested.
The results below therefore describe retrieval-mode behavior only at
c1500-o200.

Every value is recall@10 for natural human questions. The best mode for
each model is shown in bold.

Model	BM25_ONLY	VECTOR_ONLY	HYBRID_RRF	Type
No embedding model	0.600	—	—	lexical baseline
all-minilm	—	0.667	0.700	local
bge-m3	—	0.867	0.767	local
granite-embedding	—	0.733	0.767	local
mxbai-embed-large	—	0.833	0.833	local
nomic-embed-text	—	0.667	0.667	local
qwen3-embedding:0.6b	—	0.800	0.800	local
snowflake-arctic-embed2	—	0.900	0.800	local
EmbeddingsGigaR	—	0.711	0.767	commercial
GigaEmbeddings-3B	—	0.833	0.867	commercial
codestral-embed	—	0.967	0.900	commercial
mistral-embed-2312	—	0.900	0.800	commercial
text-embedding-3-large	—	0.867	0.833	commercial
text-embedding-3-small	—	0.833	0.867	commercial
voyage-4-large	—	0.878	0.833	commercial
voyage-code-2	—	0.933	0.867	commercial
voyage-code-3	—	0.933	0.867	commercial

Commercial-model values are averaged across three repeated runs, so some
values are not multiples of one question out of thirty.

Hybrid Search Was Not a Universal Improvement

Adding BM25 to vector search helped two local and three commercial models. It
made no difference for three local models. In the remaining cases, hybrid
retrieval reduced recall@10.

Among local models, the clearest differences appeared for bge-m3 and
snowflake-arctic-embed2: vector-only search improved their results by
0.100. Among commercial models, mistral-embed-2312 showed the same
improvement.

One possible explanation is that BM25 helps when the correct file contains
query terms missed by vector search. It can also promote lexically similar but
semantically incorrect files and weaken an already strong vector ranking. The
experiment did not test this mechanism directly.

BM25 Remains a Useful Baseline

For natural questions, BM25 achieved recall@10 = 0.600, below every tested
embedding-based combination. Its result, however, depended strongly on query
language.

For queries composed of technical terms and keywords, BM25 reached
0.833–0.867. With inaccurate terminology, it fell to 0.400. Lexical search
works well when the developer already knows the names of relevant entities,
but it is less effective when system behavior is described in the developer's
own words.

The choice of retrieval mode, like the choice of chunking, depends on the
embedding model. Hybrid retrieval cannot be assumed to improve vector search:
it helped some models, left some unchanged, and reduced the results of others.

A practical evaluation should compare at least VECTOR_ONLY and HYBRID_RRF
on the selected model and representative queries. BM25 remains both a useful
control point and a standalone option for precise technical searches.

5. The Effect of Query Phrasing

The same question about code can be expressed in different ways. A developer
may describe system behavior in natural language, list known technical terms,
or use a plausible but incorrect name for a component.

To test retrieval robustness under these changes, each question was represented
in several forms:

• human — a natural developer question;
• ai_optimized — a detailed query using precise technical terminology;
• keyword — a short list of keywords;
• wrong_terminology — the original intent with one controlled terminology error;
• cross_module — a question connecting multiple system components.

The construction rules for these variants are described in
Part 1 of the study.

This comparison fixed chunking at c1500-o200 and used HYBRID_RRF. Every
value is recall@10. The cross_module variant existed for only ten
applicable questions, while the other results were calculated across all
thirty.

Model	human	ai_optimized	keyword	wrong_terminology	cross_module	Type
BM25 without an embedding model	0.600	0.833	0.867	0.400	0.600	baseline
all-minilm	0.700	0.933	0.933	0.433	0.600	local
bge-m3	0.767	1.000	0.833	0.633	0.700	local
granite-embedding	0.767	1.000	0.967	0.567	0.700	local
mxbai-embed-large	0.833	0.967	0.867	0.633	0.700	local
nomic-embed-text	0.667	0.700	0.733	0.467	0.700	local
qwen3-embedding:0.6b	0.800	1.000	0.900	0.600	0.700	local
snowflake-arctic-embed2	0.800	1.000	1.000	0.633	0.700	local
EmbeddingsGigaR	0.767	1.000	0.900	0.600	0.700	commercial
GigaEmbeddings-3B	0.867	1.000	0.800	0.633	0.700	commercial
codestral-embed	0.900	1.000	1.000	0.667	0.700	commercial
mistral-embed-2312	0.800	1.000	1.000	0.633	0.700	commercial
text-embedding-3-large	0.833	1.000	0.933	0.600	0.700	commercial
text-embedding-3-small	0.867	1.000	0.967	0.567	0.700	commercial
voyage-4-large	0.833	1.000	1.000	0.733	0.700	commercial
voyage-code-2	0.867	1.000	1.000	0.700	0.700	commercial
voyage-code-3	0.867	1.000	1.000	0.700	0.700	commercial

Precise Technical Queries Make Retrieval Easier

The ai_optimized variant contains class names, component names, and
operations already present in the code. On these queries, all nine commercial
and four of the seven local models reached recall@10 = 1.000.

This does not mean that code retrieval is solved. It shows that Code-RAG works
far better when the user already knows the terminology and approximate
location of the answer. In practice, retrieval is often needed precisely
because that knowledge is missing.

Short keyword queries also performed well. Even BM25 reached 0.867, because
the keywords often matched names and terms in the source code directly.

Inaccurate Terminology Hurts Every Model

Replacing one term with a plausible but incorrect alternative reduced the
result of every tested model. Among commercial models, the drop relative to
human ranged from 0.100 for voyage-4-large to 0.300 for
text-embedding-3-small.

The model ranked first on natural questions was not the most robust to
terminology distortion. codestral-embed fell from 0.900 to 0.667, while
voyage-4-large fell from 0.833 to 0.733.

Code specialization also failed to predict robustness. The smallest and
largest observed drops among commercial models both belonged to
general-purpose models.

Cross-Module Queries Need a Separate Study

The cross_module values barely distinguish the embedding models: every model
except all-minilm received 0.700, while all-minilm received 0.600.
This variant existed for only ten questions, so the result cannot be
interpreted as evidence of equal robustness.

A meaningful comparison would require a separate question set focused on
relationships between modules and containing more examples of this type.

Query phrasing is another parameter of the retrieval pipeline. Precise
terminology can bring almost every model close to the maximum result, while a
small terminology error can reduce quality substantially.

Model selection should therefore account for where queries come from. A system
for developers familiar with the codebase and a system for new team members or
non-technical users may require different configurations.

6. Model Comparison Results

For the baseline comparison, every model ran under the same conditions:
natural human questions, HYBRID_RRF, and c1500-o200 chunking.

Rank	Model	Model types in row	Recall@10	95% CI
1	mistral/codestral-embed	commercial	0.900	[0.800–1.000]
2–5	GigaEmbeddings-3B, text-embedding-3-small, voyage-code-2, voyage-code-3	commercial	0.867	[0.733–0.967]
6–8	mxbai-embed-large, text-embedding-3-large, voyage-4-large	local and commercial	0.833	[0.700–0.967]
9–11	qwen3-embedding:0.6b, snowflake-arctic-embed2	local	0.800	[0.633–0.933]
9–11	mistral-embed-2312	commercial	0.800	[0.666–0.933]
12–14	bge-m3, granite-embedding, EmbeddingsGigaR	local and commercial	0.767	[0.600–0.900]
15	all-minilm	local	0.700	[0.533–0.867]
16	nomic-embed-text	local	0.667	[0.500–0.833]

At first glance, this looks like a conventional ranking: a specialized
commercial model takes first place, and the remaining results gradually fall
from 0.867 to 0.667. With thirty questions, however, a difference of
0.033 represents only one retrieved file.

One question separates codestral-embed from the next four models. Those four
retrieved the correct files for the same 26 questions out of 30, while the
leader retrieved one additional file. A paired bootstrap analysis showed that
the confidence interval for every pairwise difference among the top five
models included zero. The available data is therefore insufficient to treat
their order as stable.

The separation between local and commercial models was also less pronounced
than expected. Local mxbai-embed-large achieved 0.833, two correct answers
behind the leader out of thirty, and its confidence interval overlaps those of
commercial models.

Larger and more expensive models did not always produce better results.
text-embedding-3-large achieved 0.833, while the cheaper
text-embedding-3-small reached 0.867. The compact 62 MB
granite-embedding tied the 1.2 GB bge-m3 at 0.767. Two generations of
specialized Voyage models, voyage-code-2 and voyage-code-3, also completed
the baseline comparison with the same result of 0.867.

These observations do not prove that the models are equal: thirty questions
are insufficient for confidently comparing small differences. They do show
why an embedding-model ranking is meaningful only together with its
measurement conditions. Changing chunking, retrieval mode, or query phrasing
can alter both the metric and the order of models in the table.

7. Practical Recommendations

The experiment does not identify one configuration suitable for every
Code-RAG project. The decision depends on the queries the system will receive,
where it will run, and how much time is available for tuning.

Requirement	Candidate	Evidence from this study
Highest observed recall@10	codestral-embed + c1500-o200 + VECTOR_ONLY	Highest point estimate among completed runs: 0.967
No commercial APIs	qwen3-embedding:0.6b + c3000-o300	Highest tuned local-model result: 0.933
Minimal initial tuning	voyage-4-large or voyage-code-3	Smallest observed range across chunking configurations: 0.067
Restricted memory	granite-embedding	A 62 MB model with the same baseline point estimate, 0.767, as the 1.2 GB bge-m3
Precise technical queries	BM25 as standalone search or a baseline	BM25 reached recall@10 = 0.867 on keyword queries without embedding infrastructure
Inaccurate user queries	Vector or hybrid retrieval after testing the chosen model	The advantage of semantic retrieval over BM25 was most visible with inaccurate terminology

These candidates reflect the best observed results inside the experiment, not
universal production configurations. For example, codestral-embed produced
the highest recall@10, but its advantage was measured at one chunking
configuration, and statistical superiority over nearby models was not
established. Local models avoid API charges but move cost into hardware and
operations.

Practical Code-RAG tuning should begin with a description of future queries,
not a large model leaderboard. If the system is used by developers familiar
with project terminology, BM25 may be a strong starting point. If questions
come from new team members or users describing behavior in their own words,
semantic retrieval becomes more important.

The next step is to choose a short list of models that meet cost, memory, and
deployment constraints. For each candidate, test several fragment sizes, then
compare VECTOR_ONLY and HYBRID_RRF. A chunking or retrieval-mode choice
should not be transferred from one model to another without retesting.

The final comparison should include not only convenient technical queries, but
also natural phrasing, inaccurate terminology, and plausible but incorrect
files. Results should be retained per question so that an apparent advantage
can be traced to a stable pattern rather than a few favorable examples.

In practice, selecting a Code-RAG configuration becomes a process of narrowing
the search space:

1. identify realistic query types and system constraints;
2. measure BM25 and one accessible embedding model as starting points;
3. select a short list of candidate models;
4. test several chunking configurations for each candidate;
5. compare vector-only and hybrid retrieval;
6. repeat the evaluation across different query phrasings;
7. test the statistical stability of the resulting ranking.

8. Conclusion

The central result of this study is that an embedding model cannot be evaluated
independently from the retrieval pipeline around it. Each model has its own
effective combination of chunking, retrieval mode, and query phrasing.

These parameters are connected. Fragment size determines how much code enters
an embedding. Retrieval mode sets the balance between semantic similarity and
exact terminology. Query phrasing determines how easily the system can connect
a developer's intent to the vocabulary of the source code.

There is therefore no universal ranking of Code-RAG models. Models can only be
compared under explicit conditions: on a particular codebase, with a selected
chunking strategy and retrieval mode, and for a known distribution of user
queries.

The practical question is not "Which model is best?" but "Which configuration
best solves the tasks of this project's users?" Answering it requires joint
tuning of the pipeline and evaluation on a project-specific question set.

This study used one polyglot project and a small gold set, so its selected
configurations should not be transferred to other codebases without retesting.
Part 3 will describe the reproducible benchmark harness used to run these
comparisons.

Reproducibility

The raw results are published with the project. The main tables in this article
can be verified using these artifacts:

results/E003-full/all.parquet
results/E007-commercial-chunking-merged/all.parquet
results/E004-vector/all.parquet
results/E005-production/all.parquet
results/E006-production-merged/all.parquet
results/E007-commercial-vector-merged/all.parquet
results/E004-bm25/all.parquet
results/forest_plot_data.csv

The confidence intervals for the baseline ranking can be recalculated from the
source CSV files:

git clone https://github.com/Daeryss/karta-rag-map
cd karta-rag-map
python3 -m venv .venv
.venv/bin/pip install -r scripts/requirements.txt
.venv/bin/python scripts/bootstrap_cis.py \
  results/E005-production/all.csv \
  results/E006-production-merged/all.csv

The script fixes the baseline conditions at k=10, the human query variant,
2,000 bootstrap samples, and seed 42.

---

A Cognitive Benchmark for Code-RAG Retrieval · Part 2 of 3 · Previous:
Part 1 — Methodology · Next: Part 3 —
Engineering a Reproducible Benchmark