Introducing HorizonSWE — Research 01
FlamkiLabs
FLAMKI LABS / HOW WE WORK

Good data has
a paper trail.

From a repository-level problem to an inspectable task package. Every requirement connects to an environment, a reference solution, and evidence.

Follow the work
TASK DESIGN

Something
worth solving.

A good task asks for connected engineering decisions: understand the repository, trace the failure, plan a change, and check what it affects.

We define the required behavior and acceptance criteria. The solver chooses the implementation path.

For long-horizon work, stages, complexity, and execution budget are agreed before delivery. Duration and difficulty need measurement on the task set.

REPOSITORY CONTEXT
01 / TRACE02 / CHANGE03 / CHECK

Follow dependencies.
Understand the consequences.

A SCHEMATIC OF CONNECTED ENGINEERING WORK
REPEATABLE STARTING CONDITIONS
REPOSITORYPinned state
RUNTIMEDefined setup
EXECUTIONDocumented reset
RESTORE. RUN. REVIEW.
REPRODUCIBILITY

Same start.
Every time.

A prompt alone isn’t a task environment. The repository, dependencies, and commands need to describe the same starting conditions for each attempt.

Snapshot & setup

A pinned repository revision, runtime, dependency versions, and installation instructions. Container requirements are agreed for the pilot.

Reset & execution

Documented steps restore the starting state. Validation commands and the environment used accompany the recorded output.

Delivery checks

Setup and runtime requirements are checked in the delivered environment. Supported platforms and execution limits are documented with the task.

VALIDATION

A patch is a proposal.
Tests are the evidence.

Check the reference against the specification. Check that the baseline exposes the missing behavior. Keep the commands and results available for review.

Reports distinguish task outcomes from execution errors and resource limits. Passing tests establish the behavior covered by those checks.

Explore the research
PUBLIC SAMPLE / AUTH-001JAVASCRIPT

The change.
The check.

BASELINE6/9

tests passed

REFERENCE9/9

tests passed

Recorded locally · Node.js 22.14.0 · macOS arm64
Public tests with an injected transport.

Inspect the source, patch & results
THE ENGINEER BEHIND THE WORK
Flamki Labs mint clover and asterisk emblemINDEPENDENT AI DATA & EVALUATION

Founded on engineering.
Built for frontier AI.

Flamki Labs is an independent AI data and evaluation company focused on high-quality software engineering datasets, reproducible task environments, and benchmarks for advanced AI systems.

Founded by Ayush Singh, Flamki Labs brings hands-on experience in software engineering and AI data production to the development of coding datasets and evaluations.

LET’S MAKE THE WORK CONCRETE

Your capability.
Our next task.

Start with five tasks. Define the scope, acceptance criteria, pricing, and schedule together.

Scope a pilot