Datascreen
Public dataset case study / August 2026

Identical prompts received opposite answers in 68 groups inside OpenHermes-2.5.

Datascreen checked all 1,001,551 rows in the recorded source snapshot, connected identical prompts with different answers, and traced the strongest verified pattern back to one contributing source.

SOURCE ROWS CHECKED 1,001,551 The recorded source snapshot
TRAINING EXAMPLES 1,008,268 Some rows contained multiple examples
PROMPTS WITH OPPOSITE ANSWERS 68 One row answered Yes or True while another answered No or False
AFFECTED TRAINING EXAMPLES 191 Examples connected to those 68 prompts
01 / SOURCE SCOPE

The scan checked the recorded source snapshot before narrowing the evidence.

OpenHermes-2.5 contained 1,001,551 source rows. Some rows produced more than one user and assistant exchange, resulting in 1,008,268 training examples. No source rows were skipped.

Source coverage Every row processed

The result is based on all 1,001,551 rows in the recorded source snapshot, not a sample.

Comparison Identical prompts connected

Answers to the same prompt were grouped so disagreements could be inspected together.

Verification Opposite answers isolated

The complete responses had to be Yes, No, True, or False after capitalization and punctuation were ignored. Each counted group contained at least one Yes or True answer and at least one No or False answer.

02 / FROM BROAD SCAN TO VERIFIED RESULT

The broad comparison led to a smaller result that could be stated plainly.

Across the recorded dataset snapshot, 4,641 groups contained identical prompts with different answers. A contributing source named UnnaturalInstructions accounted for 145 of those groups. Of those, 68 contained a Yes or True answer in one row and a No or False answer in another for the exact same prompt.

VERIFICATION PATH
Groups found across the full dataset4,641
Groups traced to UnnaturalInstructions145
Groups with Yes or True and No or False answers68
Examples in the verified groups191
Only these 68 groups are presented as verified conflicts.
03 / OPPOSITE ANSWERS

One prompt appeared three times with incompatible answers.

These were three separate training examples from the same contributing source. Each used the identical prompt below.

IDENTICAL PROMPT

You are given a description of an event, and you need to determine if it is credible or not by outputting True or False.

A rock was thrown into the air and disappeared.
ROW 947,905False.
ROW 947,906Yes.
ROW 947,907True

False, Yes, and True cannot all be correct answers to the same True-or-False prompt. Datascreen surfaced the disagreement without pretending to determine the correct answer automatically.

04 / ANSWERS FROM OTHER TASKS

Five copies of one programming prompt included four answers to unrelated questions.

The shared prompt asked for code that sums a list of integers. One response answered that request. The other four responses belonged to different tasks, making the affected examples unsuitable without repair or removal.

SHARED PROMPT

Write a function that takes in a list of integers and outputs the sum of all the integers. List: [1, 2, 3].

RELEVANT ANSWER

Python code that adds the list and outputs 6.

ANSWER FROM A DIFFERENT TASK

['What color was the lamb's fleece?'] - 5

05 / WHY THIS MATTERS

Conflicting examples can teach incompatible outputs for the same input.

Training data is supposed to provide a useful target for each prompt. When identical prompts carry opposite labels or answers from unrelated tasks, the training signal no longer expresses one coherent outcome. Datascreen makes those rows reviewable together instead of leaving the disagreement buried across a million-row file.

WHAT THE REVIEWER RECEIVES
Identical promptShown once
Different answersGrouped together
Source rowsPreserved
Final actionReviewer controlled
The product reduces a scattered dataset problem to a concrete decision about specific source rows.
06 / WHAT THIS ESTABLISHES

A verified dataset conflict with clear limits on the claim.

OBSERVED

All 1,001,551 rows in the recorded source snapshot were checked. One named contributing source contained 68 groups where the exact same prompt received a Yes or True answer in one row and a No or False answer in another.

NOT ESTABLISHED

This result does not show that OpenHermes-2.5 is broadly low quality, that the conflicts were intentional, or that a downstream model was harmed.

Exact source

OpenHermes-2.5 at commit b82037821055c377bed0d495e72e46de3bc72e84

SHA-256 · abe573d17eade4161aac321028027dd5ba614a6d9516d51bde9299d5353e1609

View source dataset →
Inspect your own dataset

Trace inconsistent answers back to the exact prompts, rows, and contributing sources.

Datascreen turns disagreements spread across large training datasets into source-linked findings that can be reviewed and resolved.