Update README.md

README.md

@@ -1,101 +1,100 @@
 ---
 license: cc-by-4.0
 language:
-  - en
+- en
 size_categories:
-  - n>1T
+- 100B<n<1T
 task_categories:
-  - text-generation
+- text-generation
 source_datasets:
-  - nebius/SWE-rebench-V2
-pretty_name: SWE-rebench V2 CodeWorldModeling Traces
+- nebius/SWE-rebench-V2
+pretty_name: SWE-rebench V2 Code World Modeling Traces
 tags:
-  - code
-  - execution-traces
-  - swe-rebench
-  - marin
+- code
+- execution-traces
+- swe-rebench
+- marin
 ---
 
 # SWE-rebench V2 — CodeWorldModeling Traces
 
-> **This is a derived dataset.** Every record is produced from an instance of
-> [`nebius/SWE-rebench-V2`](https://huggingface.co/datasets/nebius/SWE-rebench-V2).
-> It is **governed by the SWE-rebench V2 license** — see [License](#license)
-> below — including the requirement to respect each source repository's own
-> license.
+This dataset turns SWE-rebench V2 software-engineering tasks into execution-trace
+training examples. For each traced test, it records the test source together with
+a line-by-line account of how the relevant Python code executed: which functions
+were entered, what arguments were passed, which branches were taken, what values
+were assigned, what returned, and where exceptions occurred.
 
-Line-by-line Python execution traces for the test suites of SWE-rebench V2
-instances, captured by running each instance's tests under a tracer inside
-Nebius ConTree sandboxes.
+The goal is to make the runtime behavior behind SWE-style bug fixes explicit.
+Instead of seeing only a repository, a failing test, and a patch, a model or
+researcher can inspect how the program behaved before and after the fix.
 
-Each instance comes with a fix patch. The pipeline traces two kinds of tests:
+> **Derived dataset.** Every record is produced from an instance of
+> [`nebius/SWE-rebench-V2`](https://huggingface.co/datasets/nebius/SWE-rebench-V2).
+> See [License](#license) for licensing details.
 
-- **Fix-verifying tests** (`affected=True`) — tests that fail before the fix
-  patch and pass after it. These are the tests that demonstrate the bug and its
-  fix. Such a row is traced *twice*: once before the patch (the failing run) and
-  once after (the passing run).
-- **Regression-suite tests** (`affected=False`) — the rest of the repository's
-  existing test suite ("broad phase"), each traced once at the fixed revision.
+## What is included
 
-Each row is one ``(instance_id, test_id)`` trace.
+Each SWE-rebench V2 instance contains a bug-fixing patch and a set of tests. This
+dataset traces those tests in two ways.
 
-## Schema
+For tests that directly verify the fix, the dataset captures the story of the bug:
+the test fails before the patch, the patch is applied, and the same test passes
+afterward. These rows have `affected=True`. Their `text` field contains the test
+source, the pre-patch trace, the patch itself, and the post-patch trace.
 
-Trace shards under `data/`:
+For the rest of the repository's test suite, the dataset captures normal
+regression behavior at the fixed revision. These rows have `affected=False`.
+Their `text` field contains the test source followed by a single execution trace.
 
-| column | type | notes |
-| --- | --- | --- |
-| `instance_id` | string | SWE-rebench-V2 row id |
-| `test_id` | string | pytest node id |
-| `affected` | bool | `True` = fix-verifying test (fails before the patch, passes after); `False` = regression-suite test traced once at the fixed revision |
-| `text` | string | plain-text test source + line-by-line execution trace (see below) |
+Each row corresponds to one `(instance_id, test_id)` pair. Rows for instances
+that failed before any trace could be captured are excluded.
 
-Sentinel rows from instances that failed before any trace was captured are
-excluded.
+## How to read a row
 
-## Reading a trace
+The central field is `text`. It is designed to be readable as plain text: first
+the test, then the trace material associated with that test.
 
-The `text` of a regression-suite row (`affected=False`) is the test source
-followed by a single trace:
+For a regression-suite row (`affected=False`), the format is:
 
-```
+```text
 <test source>
 
 # --- trace ---
 <line-by-line execution trace>
 ```
 
-A fix-verifying row (`affected=True`) carries the test source, the trace from
-*before* the fix, the fix patch itself, and the trace from *after* the fix:
+For a fix-verifying row (`affected=True`), the format is:
 
-```
+```text
 <test source>
 
 # --- pre-patch trace ---
 <line-by-line execution trace, before the fix>
 
 # --- patch ---
-<the fix patch, in `git diff` form>
+<the fix patch, in git diff form>
 
 # --- post-patch trace ---
 <line-by-line execution trace, after the fix>
 ```
 
-A trace replays execution one source line at a time. Inline `#` comments record
-runtime state:
+Within a trace, comments on the right-hand side or on standalone lines describe
+runtime state. For example, the trace records function entry, argument values,
+branch outcomes, assignments, return values, and exceptions.
 
-| comment | meaning |
-| --- | --- |
-| `# === file::function (line N) ===` | execution entered this function frame |
-| `# ENTER: arg=value, ...` | argument values at the call |
-| `# branch=if:True` / `:False` | which way a conditional went |
-| `name = expr  # name=value` | value bound by an assignment (right-margin) |
-| `# RETURN from function: value` | the frame's return value |
-| `# EXCEPTION in function: ...` | an exception raised in the frame |
+| Trace annotation                         | Meaning                                                           |
+| ---------------------------------------- | ----------------------------------------------------------------- |
+| `# === file::function (line N) ===`      | Execution entered this function frame.                            |
+| `# ENTER: arg=value, ...`                | Function argument values at call time.                            |
+| `# branch=if:True` / `# branch=if:False` | Which branch a conditional took.                                  |
+| `name = expr  # name=value`              | Value bound by an assignment, recorded as a right-margin comment. |
+| `# RETURN from function: value`          | Function return value.                                            |
+| `# EXCEPTION in function: ...`           | Exception raised in the frame.                                    |
 
-### Example — a regression-suite trace (`affected=False`)
+## Example: regression behavior
 
-Source file `mathx.py`:
+A regression-suite trace shows how a passing test executes at the fixed revision.
+Consider this simple source file:
 
 ```python
 def clamp(value, low, high):
@@ -106,9 +105,10 @@ def clamp(value, low, high):
     return value
 ```
 
-Test `test_clamp_within_range` and its trace:
+A test that exercises the in-range case appears together with its execution
+trace:
 
-```
+```text
 def test_clamp_within_range():
     assert clamp(5, 0, 10) == 5
 
@@ -126,13 +126,18 @@ def clamp(value, low, high):
     # RETURN from clamp: 5
 ```
 
-### Example — a fix-verifying trace (`affected=True`)
+The trace shows that `clamp` was called with `value=5`, that neither conditional
+branch was taken, and that the function returned `5`.
 
-The buggy `clamp` returns `low` instead of `high` when `value` is above the
-range, so `test_clamp_above_range` fails. The fix patch corrects it. The
-pre-patch trace shows the failure, the post-patch trace shows it passing:
+## Example: bug and fix behavior
 
-```
+A fix-verifying trace shows the same test before and after the patch. In the
+example below, the buggy implementation returns `low` instead of `high` when the
+input is above the allowed range. The pre-patch trace exposes the failing
+behavior, the patch corrects the implementation, and the post-patch trace shows
+the corrected return value.
+
+```text
 def test_clamp_above_range():
     assert clamp(99, 0, 10) == 10
 
@@ -174,14 +179,27 @@ def clamp(value, low, high):
     # RETURN from clamp: 10
 ```
 
-## Joining the per-instance license
+## Dataset fields
 
-`metadata/licenses.parquet` is a companion file mapping `instance_id` ->
-`license` (the source repository's license at the instance commit). It is kept
-separate so the multi-GB trace shards do not carry a redundant per-row string.
-Join it on `instance_id` to attach the license to any trace row.
+Trace shards are stored under `data/`.
 
-With `datasets`:
+| Column        | Type   | Description                                                                                                                                                                                                   |
+| ------------- | ------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| `instance_id` | string | SWE-rebench V2 row identifier.                                                                                                                                                                                |
+| `test_id`     | string | Pytest node identifier.                                                                                                                                                                                       |
+| `affected`    | bool   | Whether the test is fix-verifying. `True` means the test fails before the patch and passes after it. `False` means the test is part of the broader regression suite and is traced once at the fixed revision. |
+| `text`        | string | Test source and trace material in plain text.                                                                                                                                                                 |
+
+## Loading per-instance licenses
+
+The companion file `metadata/licenses.parquet` maps each `instance_id` to the
+license of the corresponding source repository. It is stored separately so the
+multi-GB trace shards do not repeat a per-instance license string on every row.
+
+Join this file on `instance_id` when you need license metadata attached to trace
+rows.
+
+### Using `datasets`
 
 ```python
 from datasets import load_dataset
@@ -195,12 +213,13 @@ license_by_instance = dict(zip(licenses["instance_id"], licenses["license"]))
 traces = traces.map(lambda row: {"license": license_by_instance[row["instance_id"]]})
 ```
 
-With pandas:
+### Using pandas
 
 ```python
 import pandas as pd
 
 base = "hf://datasets/marin-community/swe-rebench-v2-CodeWorldModeling"
+
 traces = pd.read_parquet(f"{base}/data")
 licenses = pd.read_parquet(f"{base}/metadata/licenses.parquet")
 traces = traces.merge(licenses, on="instance_id", how="left")
@@ -208,9 +227,11 @@ traces = traces.merge(licenses, on="instance_id", how="left")
 
 ## Provenance
 
-- **Source dataset**: [`nebius/SWE-rebench-V2`](https://huggingface.co/datasets/nebius/SWE-rebench-V2)
-- **Generator**: [`experiments/swe_rebench_trace/contree_pipeline.py`](https://github.com/marin-community/marin/blob/main/experiments/swe_rebench_trace/contree_pipeline.py)
-  in [marin-community/marin](https://github.com/marin-community/marin).
+The traces are derived from [`nebius/SWE-rebench-V2`](https://huggingface.co/datasets/nebius/SWE-rebench-V2).
+They were generated with
+[`experiments/swe_rebench_trace/contree_pipeline.py`](https://github.com/marin-community/marin/blob/main/experiments/swe_rebench_trace/contree_pipeline.py)
+in [`marin-community/marin`](https://github.com/marin-community/marin), by
+running tests under a tracer inside Nebius ConTree sandboxes.
 
 ## License
 
@@ -221,36 +242,34 @@ This dataset is derived from SWE-rebench V2 and inherits its license terms:
 > particular instance is based. To facilitate this, the license of each
 > repository at the time of the commit is provided for every instance.
 
-The per-instance source-repository license is provided in the companion file
-`metadata/licenses.parquet` (join on `instance_id`). When using or
-redistributing these traces, honor the license of the originating repository
-for each instance.
+Per-instance source-repository licenses are provided in
+`metadata/licenses.parquet` and can be joined on `instance_id`.
 
 ## Citation
 
-Please cite both this dataset and the source dataset it derives from.
+Please cite both this dataset and the source dataset from which it is derived.
 
-This dataset:
+### This dataset
 
 ```bibtex
 @misc{marincommunity2026swerebenchcodeworldmodeling,
-      title={SWE-rebench V2 CodeWorldModeling Traces},
-      author={Marin Community},
-      year={2026},
-      howpublished={\url{https://huggingface.co/datasets/marin-community/swe-rebench-v2-CodeWorldModeling}},
+  title        = {SWE-rebench V2 CodeWorldModeling Traces},
+  author       = {Marin Community},
+  year         = {2026},
+  howpublished = {\url{https://huggingface.co/datasets/marin-community/swe-rebench-v2-CodeWorldModeling}},
 }
 ```
 
-Source dataset:
+### Source dataset
 
 ```bibtex
 @misc{badertdinov2026swerebenchv2languageagnosticswe,
-      title={SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale},
-      author={Ibragim Badertdinov and Maksim Nekrashevich and Anton Shevtsov and Alexander Golubev},
-      year={2026},
-      eprint={2602.23866},
-      archivePrefix={arXiv},
-      primaryClass={cs.SE},
-      url={https://arxiv.org/abs/2602.23866},
+  title         = {SWE-rebench V2: Language-Agnostic SWE Task Collection at Scale},
+  author        = {Ibragim Badertdinov and Maksim Nekrashevich and Anton Shevtsov and Alexander Golubev},
+  year          = {2026},
+  eprint        = {2602.23866},
+  archivePrefix = {arXiv},
+  primaryClass  = {cs.SE},
+  url           = {https://arxiv.org/abs/2602.23866},
 }
-```
+```