Skip to content
RETRIDGE
Case studies

Results, published only when a client permits it.

Retridge does not publish invented numbers. Below is the exact structure every engagement report follows, and the anonymized template published once a client signs off. Everything on this page is labeled.

Anatomy

Seven sections, in the same order every time.

The sequence matters: measurement comes before diagnosis, and diagnosis comes before change. If a case study skips “what we measured”, it is marketing.

01
Client context
Industry, system type, users, scale. Anonymized where required.
02
Problem
The symptom as the client described it, in their words.
03
What we measured
Query set size, metrics, and the layers evaluated.
04
What we found
Root causes, ranked. Including the ones that were not the model.
05
What we changed
The fixes actually shipped, and who shipped them.
06
Results
Before/after on the same evaluation set. No cherry-picked queries.
07
Client quote
Attributed where permitted, role-only otherwise. Never written by us.
Template preview

How a published case study will read

Demonstration content · not a real client
Financial services Internal policy assistant RAG System Audit

A policy assistant that retrieved the right document and still gave the wrong answer

Sample structure showing how findings, changes, and before/after metrics are presented in a published Retridge case study.

01 · Client context

A mid-size lender with an internal assistant over policy documents, procedures, and product terms. Around 1,200 employees; roughly 400 daily queries at peak.

02 · Problem

“It cites the right policy and then gets the number wrong.” Operations staff had stopped trusting it for approval thresholds — the exact task it was built for.

03 · What we measured

78 real queries from support logs, scored on recall@5, ranking position of the gold chunk, groundedness, completeness, plus 40 security probes and full token accounting.

04 · What we found
Root cause 1
Archived duplicates outranked current policy

Superseded versions had no recency metadata and scored higher on lexical overlap.

Root cause 2
Tables split across chunk boundaries

Threshold tables were cut mid-row, so the model saw a figure without its condition.

Root cause 3
No abstention behavior

With weak context the system guessed instead of saying it could not answer.

05 · What we changed
Recency + status metadata filter Table-aware chunking Cross-encoder rerank, top_n 3 → 5 Abstention on low context score Archive tier removed from index Regression eval in CI
06 · Results
Same 78-query evaluation set · demonstration figures
Recall@5
0.61 0.89
Groundedness
0.44 0.86
p95 latency
6.2s 3.4s
Cost / answer
$0.041 $0.028
07 · Client quote

Quote slot — filled with the client's own words once a case study is approved for publication. Retridge does not draft client quotes.

Role · Company (or “Head of Engineering, anonymized”)
All figures on this page are demonstration data illustrating report structure. They are not client results.
Publication queue

Engagements in progress. Published on client sign-off.

Slot 01 · reserved

Legal · contract assistant

Multihop retrieval across clause libraries. Awaiting permission to publish.

Slot 02 · reserved

Healthcare admin · red team

Access-boundary and leakage testing before launch. Likely anonymized.

Slot 03 · open

Your system

Audit engagements are scheduled two to three weeks out.

Want your before/after in this format?

It starts with measuring what the system does today.