feat(dataset): add readable commit metadata and patch tables

CHANGELOG.md

@@ -1,5 +1,5 @@
-# Changelog
-
-## Local preparation
-
-Created bounded raw archives and four typed upstream indexes. No public release, remote revision, experimental rerun or canonical promotion.
+# Changelog
+
+## upstream-readable-v2
+
+Added directly readable messages, author/committer identities and dates, discovery observations, changed-file records and patch text. Added UTC date representations and explicit source-conflict handling. Simplified artifact fields and dataset documentation.

CITATION.cff

@@ -1,6 +0,0 @@
-# LOCAL DRAFT: authors and publication metadata are not yet confirmed.
-cff-version: 1.2.0
-message: "Unpublished local preparation; do not cite as a released dataset."
-title: "AI Coding-Agent Commit Upstream Corpus"
-type: dataset
-# Add confirmed authors and release version before publication.

DATA_DICTIONARY.md

@@ -1,69 +1,246 @@
-# Data dictionary
-
-Schema version: upstream-index-v1. Nullable fields retain absence; null does not mean zero or a negative finding.
-
-## artifacts
-
-| Field | Arrow type | Nullable |
-|---|---|---|
-| artifact_id | string | True |
-| collection_id | string | True |
-| kind | string | True |
-| relative_path | string | True |
-| bytes | int64 | True |
-| sha256 | string | True |
-| member_manifest_partition | string | True |
-| historical_license_flag | string | True |
-| publication_status | string | True |
-
-## commits
-
-| Field | Arrow type | Nullable |
-|---|---|---|
-| observation_id | string | True |
-| source_artifact_id | string | True |
-| repo_id | int64 | True |
-| repo_full_name | string | True |
-| commit_sha | string | True |
-| author_date | string | True |
-| language_bucket | string | True |
-| collection_id | string | True |
-| json_schema | string | True |
-| patch_artifact_id | string | True |
-| repository_observation_id | string | True |
-| metadata_mapping_status | string | True |
-| parse_status | string | True |
-
-## repositories
-
-| Field | Arrow type | Nullable |
-|---|---|---|
-| repository_observation_id | string | True |
-| source_artifact_id | string | True |
-| source_record_index | int64 | True |
-| repo_id | int64 | True |
-| full_name | string | True |
-| language | string | True |
-| stars | int64 | True |
-| forks | int64 | True |
-| license_key | string | True |
-| repository_updated_at | string | True |
-| metadata_observed_at | string | True |
-
-## sampling
-
-| Field | Arrow type | Nullable |
-|---|---|---|
-| sampling_record_id | string | True |
-| source_artifact_id | string | True |
-| source_record_index | int64 | True |
-| record_kind | string | True |
-| fields_json | string | True |
-| contains_nul | bool | True |
-
-artifact_id = SHA-256 of the UTF-8 bytes of "upstream", one NUL byte, and the source-relative path. sha256 separately identifies content bytes.
-A commit observation uses its source JSON artifact_id. A repository observation appends :row:<zero-based original JSONL line index>. Sampling uses the zero-based parsed CSV record index after its header.
-repo_id and repository_observation_id are populated only from existing raw repository metadata; multiple observations do not choose an arbitrary historical timestamp. metadata_mapping_status distinguishes an unambiguous observation from unresolved/multiple observations.
-author_date and repository_updated_at are source timestamp strings, not coerced dates. metadata_observed_at stays null because the raw repository response does not establish its collection time.
-sampling.fields_json preserves parsed CSV fields as JSON, including null or extra-field lists for irregular rows; no sampling logic is recomputed. contains_nul marks source NUL characters, preserved as JSON escapes. publication_status is local_candidate_not_approved for every artifact.
-parse_status reports commit JSON parsing. A non-ok value must be reviewed before claiming complete metadata extraction. language_bucket is the existing storage bucket, not a new language classification.
+# Data dictionary
+
+Schema: upstream-readable-v2. Null means unavailable or unresolved, not zero.
+
+## commits
+
+One saved commit record, with its message, author, committer, dates and collection labels.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `repo_full_name` | `string` | GitHub repository name in owner/repository form. |
+| `commit_sha` | `string` | Git commit identifier. Use it together with the repository name. |
+| `html_url` | `string` | Link to the commit on GitHub. |
+| `observed_agents` | `list<element: string>` | Distinct agent labels attached to matching discovery records; a commit can have several. |
+| `observed_channels` | `list<element: string>` | Distinct identification channels from matching discovery records. Agent/channel pairings remain in discovery. |
+| `repo_owner_login` | `string` | Owner component of repo_full_name: the person or organization GitHub login, not a display name. |
+| `repo_owner_type` | `string` | Owner account type, usually User or Organization, from a matching repository observation; never inferred from the login. |
+| `repo_owner_id` | `int64` | Numeric GitHub owner account ID from a matching repository observation. |
+| `message` | `string` | Full original commit message, including trailers and line breaks. |
+| `message_title` | `string` | First line of message; full message remains available. |
+| `author_name` | `string` | Name stored in the Git author entry. |
+| `author_email` | `string` | Email address stored in the Git author entry. |
+| `author_date` | `string` | Original timestamp, including its timezone stored in the Git author entry. |
+| `author_date_utc` | `string` | UTC ISO 8601 conversion of the corresponding timezone-aware source date. |
+| `committer_name` | `string` | Name stored in the Git committer entry. |
+| `committer_email` | `string` | Email address stored in the Git committer entry. |
+| `committer_date` | `string` | Original timestamp, including its timezone stored in the Git committer entry. |
+| `committer_date_utc` | `string` | UTC ISO 8601 conversion of the corresponding timezone-aware source date. |
+| `github_author_login` | `string` | GitHub username linked to the Git author; empty if the API provided no account. |
+| `github_committer_login` | `string` | GitHub username linked to the Git committer; empty if the API provided no account. |
+| `repo_id` | `int64` | Numeric GitHub repository ID, when available. |
+| `files_count` | `int64` | Number of changed-file entries in the saved delta. |
+| `additions` | `int64` | Added lines reported by the source. |
+| `deletions` | `int64` | Deleted lines reported by the source. |
+| `total_changes` | `int64` | Total changed lines reported for the commit. |
+| `parent_shas` | `list<element: string>` | Ordered parent commit SHAs from the API item. |
+| `tree_sha` | `string` | Git identifier of the commit tree. |
+| `observation_id` | `string` | Identity of the saved commit-delta observation. |
+| `source_artifact_id` | `string` | Original file reference in artifacts. |
+| `language_bucket` | `string` | Language directory used by the original collector. |
+| `collection_id` | `string` | Original source-directory group. |
+| `json_schema` | `string` | Schema label stored in the original delta JSON. |
+| `patch_artifact_id` | `string` | ID of the saved patch; joins to patches.artifact_id. |
+| `repository_observation_id` | `string` | ID of a repository record; joins to repositories. |
+| `metadata_mapping_status` | `string` | Whether the original repository-metadata join found one, no, or multiple records. |
+| `parse_status` | `string` | Whether the saved delta JSON could be parsed. |
+| `author_date_parse_status` | `string` | ok, missing, timezone_missing or invalid; a join conflict is separately recorded. |
+| `committer_date_parse_status` | `string` | ok, missing, timezone_missing or invalid; a join conflict is separately recorded. |
+| `github_author_type` | `string` | Account type, such as User or Bot linked to the Git author; empty if the API provided no account. |
+| `github_committer_type` | `string` | Account type, such as User or Bot linked to the Git committer; empty if the API provided no account. |
+| `api_url` | `string` | GitHub API URL of the source item. |
+| `verification_reason` | `string` | Source explanation for the signature-verification result. |
+| `verification_json` | `string` | Saved verification object; delta source in commits, API commit source in discovery. |
+| `github_author_id` | `int64` | Numeric account ID linked to the Git author; empty if the API provided no account. |
+| `github_committer_id` | `int64` | Numeric account ID linked to the Git committer; empty if the API provided no account. |
+| `comment_count` | `int64` | Number of commit comments reported by the source API response. |
+| `verification_verified` | `bool` | Signature-verification value in the saved source. |
+| `discovery_observation_count` | `int64` | Number of original matching discovery observations, including historical copies. |
+| `metadata_join_status` | `string` | matched, field_conflicts, or not_found for the discovery join. |
+| `metadata_conflict_fields` | `list<element: string>` | Fields with disagreeing non-null discovery values. |
+| `discovery_example_observation_id` | `string` | One source observation for navigation; consult all matching discovery rows for field provenance. |
+
+## discovery
+
+One collected API record. The same commit can appear in several records.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `discovery_observation_id` | `string` | Unique ID of one original discovery record. |
+| `source_artifact_id` | `string` | Original file reference in artifacts. |
+| `source_relative_path` | `string` | Original path below the source data root. |
+| `record_kind` | `string` | commit for a commit API item; other_api_item for other saved responses. |
+| `repo_full_name` | `string` | GitHub repository name in owner/repository form. |
+| `commit_sha` | `string` | Git commit identifier. Use it together with the repository name. |
+| `agent` | `string` | Agent label recorded by the collection channel. |
+| `channel` | `string` | Collection channel that produced this observation. |
+| `tier` | `string` | Sampling tier recorded by the collector. |
+| `mode` | `string` | Collection matching mode recorded by the collector. |
+| `query` | `string` | Original search query. |
+| `original_window_start_utc` | `string` | Start of the original collection query window. |
+| `original_window_end_utc` | `string` | End of the original collection query window. |
+| `segment_start_utc` | `string` | Start of the query segment used for this request. |
+| `segment_end_utc` | `string` | End of the query segment used for this request. |
+| `other_item_json` | `string` | Original non-commit API item as JSON; null for commit items. |
+| `source_record_index` | `int64` | Zero-based original JSONL line or API array-item index. |
+| `repo_id` | `int64` | Numeric GitHub repository ID, when available. |
+| `message` | `string` | Full original commit message, including trailers and line breaks. |
+| `message_title` | `string` | First line of message; full message remains available. |
+| `author_name` | `string` | Name stored in the Git author entry. |
+| `author_email` | `string` | Email address stored in the Git author entry. |
+| `author_date` | `string` | Original timestamp, including its timezone stored in the Git author entry. |
+| `author_date_utc` | `string` | UTC ISO 8601 conversion of the corresponding timezone-aware source date. |
+| `author_date_parse_status` | `string` | ok, missing, timezone_missing or invalid; a join conflict is separately recorded. |
+| `committer_name` | `string` | Name stored in the Git committer entry. |
+| `committer_email` | `string` | Email address stored in the Git committer entry. |
+| `committer_date` | `string` | Original timestamp, including its timezone stored in the Git committer entry. |
+| `committer_date_utc` | `string` | UTC ISO 8601 conversion of the corresponding timezone-aware source date. |
+| `committer_date_parse_status` | `string` | ok, missing, timezone_missing or invalid; a join conflict is separately recorded. |
+| `github_author_login` | `string` | GitHub username linked to the Git author; empty if the API provided no account. |
+| `github_author_type` | `string` | Account type, such as User or Bot linked to the Git author; empty if the API provided no account. |
+| `github_committer_login` | `string` | GitHub username linked to the Git committer; empty if the API provided no account. |
+| `github_committer_type` | `string` | Account type, such as User or Bot linked to the Git committer; empty if the API provided no account. |
+| `tree_sha` | `string` | Git identifier of the commit tree. |
+| `html_url` | `string` | Link to the commit on GitHub. |
+| `api_url` | `string` | GitHub API URL of the source item. |
+| `verification_reason` | `string` | Source explanation for the signature-verification result. |
+| `verification_json` | `string` | Saved verification object; delta source in commits, API commit source in discovery. |
+| `github_author_id` | `int64` | Numeric account ID linked to the Git author; empty if the API provided no account. |
+| `github_committer_id` | `int64` | Numeric account ID linked to the Git committer; empty if the API provided no account. |
+| `comment_count` | `int64` | Number of commit comments reported by the source API response. |
+| `parent_shas` | `list<element: string>` | Ordered parent commit SHAs from the API item. |
+| `verification_verified` | `bool` | Signature-verification value in the saved source. |
+
+## commit_files
+
+One changed-file entry from a saved commit delta.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `file_observation_id` | `string` | ID of one changed-file entry in a commit record. |
+| `observation_id` | `string` | Identity of the saved commit-delta observation. |
+| `source_artifact_id` | `string` | Original file reference in artifacts. |
+| `repo_full_name` | `string` | GitHub repository name in owner/repository form. |
+| `commit_sha` | `string` | Git commit identifier. Use it together with the repository name. |
+| `filename` | `string` | Changed file path within the repository. |
+| `previous_filename` | `string` | Previous path for a renamed file, when reported. |
+| `status` | `string` | Source change type, such as added, modified or removed. |
+| `blob_sha` | `string` | Git identifier of the file blob. |
+| `blob_url` | `string` | GitHub page for the file blob. |
+| `raw_url` | `string` | Source URL for the file contents. |
+| `contents_url` | `string` | GitHub contents API URL for the file. |
+| `file_language` | `string` | File language recorded by the original collector. |
+| `file_language_source` | `string` | How the original collector assigned the file language. |
+| `file_index` | `int64` | Zero-based position in the saved files array. |
+| `additions` | `int64` | Added lines reported by the source. |
+| `deletions` | `int64` | Deleted lines reported by the source. |
+| `changes` | `int64` | Changed lines reported for this file. |
+| `has_patch` | `bool` | Whether the collector recorded an available patch for this file. |
+
+## patches
+
+One saved patch file, available directly as text.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `artifact_id` | `string` | Path-based source-file identity; distinct from its content hash. |
+| `repo_full_name` | `string` | GitHub repository name in owner/repository form. |
+| `commit_sha` | `string` | Git commit identifier. Use it together with the repository name. |
+| `source_relative_path` | `string` | Original path below the source data root. |
+| `sha256` | `string` | SHA256 of the original complete file bytes. |
+| `patch` | `string` | UTF-8 decoded original patch, preserving exact round-trip bytes. |
+| `text_encoding_status` | `string` | utf8 or not_utf8; non-UTF8 originals remain available through artifacts. |
+| `bytes` | `int64` | Original file size in bytes. |
+
+## repositories
+
+One original repository-metadata observation.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `repo_id` | `int64` | Numeric GitHub repository ID, when available. |
+| `full_name` | `string` | GitHub repository name in owner/repository form. |
+| `owner_login` | `string` | Repository owner GitHub login recorded in the original API response. |
+| `owner_type` | `string` | Repository owner account type recorded by GitHub, usually User or Organization. |
+| `owner_id` | `int64` | Numeric GitHub account ID of the repository owner. |
+| `owner_html_url` | `string` | GitHub profile page of the repository owner. |
+| `language` | `string` | Primary repository language reported by GitHub. |
+| `stars` | `int64` | Stargazer count at the source observation. |
+| `forks` | `int64` | Fork count at the source observation. |
+| `license_key` | `string` | License key present in the original repository response. |
+| `repository_observation_id` | `string` | ID of a repository record; joins to repositories. |
+| `source_artifact_id` | `string` | Original file reference in artifacts. |
+| `source_record_index` | `int64` | Zero-based original JSONL line or API array-item index. |
+| `repository_updated_at` | `string` | Repository update timestamp from the API response. |
+| `metadata_observed_at` | `string` | Null when the repository response does not establish collection time. |
+
+## artifacts
+
+One original source file and its archive location.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `artifact_id` | `string` | Path-based source-file identity; distinct from its content hash. |
+| `collection_id` | `string` | Original source-directory group. |
+| `kind` | `string` | Original file category, such as commit-json or commit-patches. |
+| `relative_path` | `string` | Original source path relative to the data root. |
+| `bytes` | `int64` | Original file size in bytes. |
+| `sha256` | `string` | SHA256 of the original complete file bytes. |
+| `member_manifest_partition` | `string` | Manifest containing archive/member/offset mappings. |
+
+## sampling
+
+One original sampling CSV record.
+
+| Field | Type | Meaning |
+|---|---|---|
+| `sampling_record_id` | `string` | ID of a row in an original sampling CSV. |
+| `source_artifact_id` | `string` | Original file reference in artifacts. |
+| `source_record_index` | `int64` | Zero-based CSV data-record index, excluding the header; a record may span physical lines. |
+| `record_kind` | `string` | Original CSV filename without its extension, such as sampled_windows or query_segments. |
+| `fields_json` | `string` | Parsed original CSV fields, retaining irregular values and NUL escapes. |
+| `contains_nul` | `bool` | Whether the source CSV record contains a NUL character. |
+
+## Identity and joins
+
+`artifact_id = SHA256(UTF8("upstream" + NUL + source_relative_path))`. Content identity is separately stored as `sha256`.
+Commit observation IDs remain the delta JSON artifact IDs. Repository observations use the source JSONL artifact plus its original line index. Discovery IDs use the source artifact plus `:record:<zero-based index>`.
+Discovery joins use case-insensitive repository names plus exact commit SHA. Missing values do not conflict with observed values; differing non-null values do. The complete source observations remain in discovery.
+Verification fields, changed-file counts and additions/deletions in commits come from the delta JSON. Owner IDs and types come from matching repository observations. Messages, commit identities and dates come from discovery field consensus.
+Date fields retain their source strings. UTC conversion requires an explicit timezone. No filesystem timestamp is used as a commit date.
+`collection_id` identifies the source collection group; agent and identification-channel labels are available in their own fields.
+
+## Agent labels and identification channels
+
+`commits.observed_agents` and `commits.observed_channels` summarize the matching discovery observations. These lists are independent sets, not parallel arrays. Use discovery to retain each agent/channel pairing.
+
+A channel describes how a record was found. `author` channels search author identities; `coauthor` channels search message text, including co-author trailers; `marker` channels search named tool markers. A search match is collection evidence, not proof that an agent produced the entire commit. The table below retains the query recorded by the collector; collection windows are separate discovery fields.
+
+| Agent | Channel | Mode | Recorded query |
+|---|---|---|---|
+| `claude` | `app_bot_email` | `author` | `author-email:242468646+Claude@users.noreply.github.com` |
+| `claude` | `author_claude_code_email` | `author` | `author-email:claude-code@anthropic.com` |
+| `claude` | `author_claude_email` | `author` | `author-email:claude@anthropic.com` |
+| `claude` | `cli_email` | `author` | `author-email:noreply@anthropic.com` |
+| `claude` | `discover_claude_code` | `marker` | `"Claude Code"` |
+| `claude` | `msg_cli_email` | `coauthor` | `"noreply@anthropic.com"` |
+| `claude` | `msg_coauthored_by_claude` | `coauthor` | `"Co-authored-by: Claude"` |
+| `claude` | `msg_generated_with_claude_code` | `marker` | `"Generated with Claude Code"` |
+| `codex` | `cli_email` | `author` | `author-email:codex@openai.com` |
+| `codex` | `discover_openai_codex` | `marker` | `"OpenAI Codex"` |
+| `codex` | `legacy_connector` | `author` | `author:chatgpt-codex-connector[bot]` |
+| `codex` | `msg_coauthored` | `coauthor` | `"Co-authored-by: Codex"` |
+| `codex` | `msg_codex_openai_email` | `coauthor` | `"codex@openai.com"` |
+| `codex` | `msg_noreply` | `coauthor` | `"noreply@openai.com"` |
+| `codex` | `msg_openai_codex` | `coauthor` | `"Co-authored-by: OpenAI Codex"` |
+| `codex` | `new_bot_email` | `author` | `author-email:242516109+Codex@users.noreply.github.com` |
+| `copilot` | `msg_coauthored_by_copilot` | `coauthor` | `"Co-authored-by: Copilot"` |
+| `copilot` | `msg_github_copilot` | `coauthor` | `"Co-authored-by: GitHub Copilot"` |
+| `copilot` | `msg_noreply` | `coauthor` | `"copilot@users.noreply.github.com"` |
+| `copilot` | `swe_agent_bot` | `author` | `author:copilot-swe-agent[bot]` |
+| `cursor` | `bg_agent_email` | `author` | `author-email:agent@cursor.com` |
+| `cursor` | `bot_noreply` | `author` | `author-email:206951365+cursor[bot]@users.noreply.github.com` |
+| `cursor` | `cursoragent_email` | `author` | `author-email:cursoragent@cursor.com` |
+| `cursor` | `discover_cursor_agent` | `marker` | `"Cursor Agent"` |
+| `cursor` | `msg_coauthored_by_cursor` | `coauthor` | `"Co-authored-by: Cursor"` |
+| `cursor` | `msg_cursoragent` | `coauthor` | `"cursoragent@cursor.com"` |

LICENSES.md

@@ -1,3 +1,3 @@
-# Release rights status
-
-Local candidate only; public admission is not completed. Original source repositories retain applicable rights and license conditions. A historical repository license_key does not establish the rights of every file or historical commit. No blanket MIT or CC-BY grant is applied to third-party artifacts. Final notices, API conditions, personal-data handling and any approved exclusions must be reflected in a separate admitted release manifest before uploading or publishing this directory.
+# Source rights
+
+Content originating from GitHub repositories remains subject to its original rights and applicable licenses. This dataset does not grant a blanket license to third-party code, patches or messages. `repositories.license_key` records the API value and is not a file-level license determination.

PROVENANCE.md

@@ -1,9 +1,13 @@
-# Provenance
-
-The fixed v5 upstream scope was enumerated on 2026-09-15 23:45:37–23:57:30 UTC. Source observations are tied to existing collection files and historical language buckets. Per-file size/mtime checks and source SHA-256 were recorded during packing; no cross-file atomic snapshot is claimed. Files were not regenerated or deduplicated.
-
-Tar member ownership/time are normalized to uid=gid=mtime=0 and regular-file mode0644 for reproducible packaging. Content bytes remain unchanged. Files exceeding512MiB use explicit offset/length fragments. Gzip source content is stored as-is in tar; other groups use zstd level1. Parquet uses explicit nullable schemas, zstd compression, at most10000 rows per row group, and a page index. All tools are task-only local preparation tools; their code and private control records are not part of this candidate.
-
-Repository raw JSONL supplies the repository view, which is independently bound to its whole-artifact hash. The old flattened research CSV is not published. No author time or collection time is invented for delta JSON lacking those fields.
-
-The source pilot/nonclaude_optimized_2026-06-12T12-32-47Z/query_segments.csv contains 223 consecutive NUL bytes at offsets593981 through594203 inclusive. One parsed record is marked contains_nul=true and retains the NUL values in fields_json. The exact raw file is retained. The cause is unknown and no content repair was performed.
+# Provenance
+
+The corpus preserves existing GitHub collection artifacts. Source paths and SHA256 values are recorded in `artifacts`; archive containers and their members are described in `manifests`.
+
+The source enumeration covers 2026-09-15 23:45:37–23:57:30 UTC. This is a file-inventory interval, not the GitHub event interval or a globally atomic snapshot. Original collection windows are available in discovery and sampling records.
+
+Discovery records retain their original observation boundaries and source positions. Git author/committer identities, GitHub accounts and collection labels are distinct fields. The commit table enriches each existing delta observation by repository and SHA; it preserves conflicting values in discovery instead of choosing an arbitrary value.
+
+Saved delta JSON and assembled patches reflect the original collectors' representations. Raw bytes are preserved in tar/zstd containers; tar ownership and timestamps are normalized and do not represent source collection times. Oversized source files have offset-addressed fragments; reconstruct them in offset order and compare the complete-file SHA256 in artifacts.
+
+To recover an original file: locate its artifacts row, read member_manifest_partition, select its artifact_id, download the listed archive(s), extract the listed member(s), concatenate fragments if necessary, and verify bytes and sha256. SHA256SUMS covers dataset files other than itself and platform-managed .gitattributes.
+
+Known source anomaly: `pilot/nonclaude_optimized_2026-06-12T12-32-47Z/query_segments.csv` contains 223 consecutive NUL bytes; they are preserved and marked in the sampling table.

README.md

@@ -1,15 +1,27 @@
 ---
-pretty_name: AI Coding-Agent Commit Upstream Corpus
+pretty_name: AI Coding-Agent Commits on GitHub
 tags:
 - software-engineering
 - code-agents
 - tabular
 configs:
 - config_name: commits
+  default: true
   data_files:
   - split: all
     path: data/commits/*.parquet
-  default: true
+- config_name: discovery
+  data_files:
+  - split: all
+    path: data/discovery/*.parquet
+- config_name: commit_files
+  data_files:
+  - split: all
+    path: data/commit_files/*.parquet
+- config_name: patches
+  data_files:
+  - split: all
+    path: data/patches/*.parquet
 - config_name: repositories
   data_files:
   - split: all
@@ -23,42 +35,134 @@ configs:
   - split: all
     path: data/sampling/*.parquet
 ---
-# AI Coding-Agent Commit Upstream Corpus
+# AI Coding-Agent Commits on GitHub
+
+This dataset contains GitHub commits collected through searches for AI coding-agent activity. Start with `commits` to read messages, authors and dates. Use `patches` for the code changes and `repositories` for repository metadata. The repository is private; reading it requires an account with access.
+
+## Directory structure
+
+```text
+agent-commits-raw/
+├── README.md                 Overview, layout and examples
+├── DATA_DICTIONARY.md         Fields, keys and null semantics
+├── PROVENANCE.md              Sources and representation details
+├── LICENSES.md                Source rights
+├── CHANGELOG.md               Data-format changes
+├── data/
+│   ├── commits/               Messages, identities, dates and change summaries
+│   ├── discovery/             Original collection observations and context
+│   ├── commit_files/          Changed paths and file-level statistics
+│   ├── patches/               Directly readable patch text
+│   ├── repositories/          Repository metadata
+│   ├── artifacts/             Original-file identity and archive pointers
+│   └── sampling/              Sampling definitions and query segments
+├── raw/
+│   ├── commit-json/           Saved delta JSON archives
+│   ├── commit-patches/        Saved patch archives
+│   ├── discovery/             Original discovery JSON/JSONL archives
+│   ├── repository-metadata/   Original repository JSONL archive parts
+│   └── sampling/              Original sampling CSV archives
+└── manifests/
+    ├── release.json           Schema, counts and source identity
+    ├── archives.jsonl         Archive sizes and checksums
+    ├── members-*.jsonl.gz     Source files, archive members and offsets
+    └── SHA256SUMS             Checksums of dataset files
+```
+
+Most users only need `data/`. Its Parquet tables can be read with Hugging Face Datasets, pandas or another Parquet reader. The original files are also kept in `raw/`; use those when you need to check a source record or recover the original file. `manifests/` tells you where each original file is stored. The numbers in filenames identify storage parts.
 
-PRIVATE REVIEW COPY. Uploaded only for restricted organization review and transfer verification. Public release is not approved. Repository: ASSERT-KTH/agent-commits-raw. No public publication date exists; pin a verified commit SHA when citing this private snapshot.
+## Tables
 
-## Contents
+Choose a table using its configuration name. Every table uses the split `all`. A commit collected more than once may have several records, so row counts are not counts of unique commits.
 
-This local candidate preserves 3,300,414 upstream artifacts (93,377,856,881 bytes) in 128 bounded archives. It provides 1,650,139 commit-observation index rows, 441,371 repository-observation rows, and 27,968 sampling-definition rows. Counts refer to observations and physical artifacts, not unique commits or repositories. Historical copies are retained.
+| Table / files | Rows | All columns, on disk | Main fields | What one row means |
+|---|---:|---:|---|---|
+| [commits](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/commits) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/commits/commit-json-00000-0000.parquet) | 1,650,139 | 857.4 MiB / 42 files | repo_full_name, commit_sha, html_url, repo_owner_login/type, message, author/committer, dates, observed_agents, observed_channels | One saved commit record, with its message, author, committer, dates and collection labels. |
+| [discovery](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/discovery) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/discovery/source-000-0000.parquet) | 2,374,534 | 971.6 MiB / 102 files | repo_full_name, commit_sha, message, source position, agent, channel, query, windows | One collected API record. The same commit can appear in several records. |
+| [commit_files](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/commit_files) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/commit_files/commit-json-00000-0000.parquet) | 17,826,229 | 1,876.0 MiB / 102 files | repo_full_name, commit_sha, filename, status, additions, deletions, blob URLs | One changed-file entry from a saved commit delta. |
+| [patches](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/patches) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/patches/commit-patches-00000-0000.parquet) | 1,650,139 | 14,093.6 MiB / 294 files | artifact_id, repo_full_name, commit_sha, patch | One saved patch file, available directly as text. |
+| [repositories](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/repositories) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/repositories/repositories-0000.parquet) | 441,371 | 28.3 MiB / 1 file | repo_id, full_name, owner_login/type/id, language, stars, forks, license_key | One original repository-metadata observation. |
+| [artifacts](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/artifacts) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/artifacts/commit-json-00000.parquet) | 3,300,414 | 359.3 MiB / 123 files | artifact_id, kind, relative_path, bytes, sha256, member_manifest_partition | One original source file and its archive location. |
+| [sampling](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/sampling) · [first file](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/blob/main/data/sampling/sampling-00000.parquet) | 27,968 | 0.6 MiB / 1 file | source_artifact_id, record_kind, fields_json | One original sampling CSV record. |
 
-Raw artifacts include retained commit JSON and assembled patches, discovery/API records, raw repository metadata JSONL and sampling definitions. The original collectors used delta representations; not every JSON is a complete HTTP response. Patches may be partial because their original API evidence was partial. No new collection or experimental computation was performed.
+The size column is for the entire table. Each Parquet file is directly readable; `load_dataset` combines the files within a configuration automatically. No archive reconstruction is involved. Selecting a few columns can read less data. Commit metadata and repository metadata are separate from patch text: neither table requires `patches/` or `raw/`.
 
-Known source anomaly: one sampling CSV contains 223 consecutive NUL bytes. Its original bytes are preserved; 1 affected sampling record is marked contains_nul=true. The index retains those characters as escaped JSON values. No repair or inference about the cause was made.
+## Load a table
 
-Researcher-authored project code, complete repository trees, later parent/tree/blob/dependency/registry/OSV evidence and downstream experiment results are outside this candidate. The Parquet tables are lightweight derived indexes of upstream fields and artifact locations, not replacement raw evidence. Original content bytes are restored from raw archives and independently checked against whole-artifact SHA-256.
+The examples use your locally configured Hugging Face token (`token=True`).
 
-## Structure and relationships
+```python
+from datasets import load_dataset
+
+repo = "ASSERT-KTH/agent-commits-raw"
+revision = "main"  # replace with a commit SHA for a fixed version
+
+# Select the columns needed for your analysis.
+commits = load_dataset(
+    repo, "commits", split="all", revision=revision, token=True,
+    streaming=True,
+    columns=["repo_id", "repo_full_name", "repo_owner_login", "repo_owner_type",
+             "commit_sha", "html_url", "message", "author_name", "author_date_utc",
+             "committer_name", "committer_date_utc", "observed_agents", "observed_channels"],
+)
+print(next(iter(commits)))
+
+# Repository metadata is a separate, smaller table.
+repositories = load_dataset(
+    repo, "repositories", split="all", revision=revision, token=True,
+    streaming=True,
+    columns=["repo_id", "full_name", "owner_login", "owner_type",
+             "language", "stars", "forks", "license_key"],
+)
+print(next(iter(repositories)))
+```
+
+`streaming=True` reads as you iterate, and `columns` selects the fields you need. To keep a local copy, download only the relevant Parquet directory, such as `data/commits/` or `data/repositories/`. Avoid downloading the whole repository if you only need metadata.
 
-The four configurations are commits, repositories, artifacts and sampling, each with a single all split. No train/test split is implied. See DATA_DICTIONARY.md for explicit nullable schemas and row identities. Commit SHA alone is not a unique observation key. Repository observations may repeat; unresolved joins remain null. Large patches and nested payloads stay in the raw archives.
+## Commit identities and collection labels
 
-## Local loading
+The author is the identity recorded for the original change; the committer is the identity recorded when the commit was created or applied. GitHub may also link each identity to an account. For example, `author_name` is the name written into the commit, while `github_author_login` is the linked GitHub username. Either account link can be missing.
+
+`repo_owner_login` identifies the person or organization that owns the repository, independently of who authored or committed a change. It is the owner component of `repo_full_name`, not a person's display name. `repo_owner_type` and `repo_owner_id` come from the matching repository observation when its owner name agrees; otherwise they are empty. The repository table also provides these fields as `owner_login`, `owner_type` and `owner_id`, together with `owner_html_url`.
+
+Co-author trailers, such as `Co-authored-by:`, remain in the full `message`. The author fields describe the commit's single Git author entry, not every contributor named in its message.
+
+`observed_agents` lists the agent labels recorded during collection. A commit may have more than one. `observed_channels` lists the identification channels that found it. Both are available directly in `commits`, without reading patches or archives. To see which agent and channel belong to the same observation, read the matching rows in `discovery`, where `agent`, `channel` and the original query are kept together. An empty list means no label was available in the matching source records.
+
+## Find a commit and its patch
 
 ```python
-from datasets import load_dataset
-rows = load_dataset('/absolute/path/to/candidate-dataset', 'commits',
-                    split='all', streaming=True)
+target_repo = "OWNER/REPOSITORY"  # repository you want to inspect
+target_sha = "COMMIT_SHA"        # its full commit SHA
+commits = load_dataset(repo, "commits", split="all",
+                       revision=revision, streaming=True, token=True)
+matches = commits.filter(lambda r: r["repo_full_name"] == target_repo
+                         and r["commit_sha"] == target_sha)
+commit = next(iter(matches))
+print(commit["message"], commit["author_date_utc"])
+
+patches = load_dataset(repo, "patches", split="all",
+                       revision=revision, streaming=True, token=True)
+patch = next(r for r in patches
+             if r["artifact_id"] == commit["patch_artifact_id"])
+print(patch["patch"])
 ```
 
-Once a publication is approved and verified, replace the local path with the dataset ID and pin the verified release commit SHA. Do not treat main or an uncreated version tag as a fixed research snapshot. Loading an index does not download all raw archives.
+The examples read rows until they find a match, which can take time. If you will make many queries, download the tables you need once and query them locally. You do not need to download the raw archives to read messages or patches.
 
-## Recovery
+## Relationships and dates
 
-Use artifacts.member_manifest_partition to find each original artifact's archive members. For ordinary files, read the named tar member. For fragmented files, concatenate the fragment bytes in ascending offset order and verify the whole-artifact SHA-256. The manifest records each fragment hash and length. Decompress .tar.zst as a stream before reading tar members; .tar archives preserve existing gzip files without an extra compression layer. Check safe relative paths and do not execute recovered code.
+- `commits.source_artifact_id` and `commits.patch_artifact_id` join to `artifacts.artifact_id`.
+- `commit_files.observation_id` joins to `commits.observation_id`; `patches.artifact_id` joins to the patch artifact.
+- `commits.repository_observation_id` joins to `repositories.repository_observation_id` when the repository observation is unambiguous.
+- Use repository name plus commit SHA to retrieve all corresponding `discovery` observations. Original record positions and artifact IDs identify each source.
+- `author_date` and `committer_date` retain source timezone strings. Their `_utc` columns are ISO 8601 UTC representations. Invalid, absent or timezone-free values do not receive invented timestamps.
+- If repeated source records agree, their available values are used in `commits`. If they disagree on a field, that field is left empty and listed in `metadata_conflict_fields`. Read the matching `discovery` records to see the alternatives. An empty label list means no label was available.
 
-## Rights and privacy
+## Scope and limitations
 
-No dataset-wide source-code license is asserted. Historical license flags are screening information; missing/other/unmapped values are retained as uncertainty. Neither this local build nor a recognized license flag establishes publication permission. Source notices, API redistribution conditions, personal fields and embedded credentials still require final release admission. Internal review records remain outside this candidate directory. Index rows are also subject to admission.
+The source archive contains 3,300,414 files (93,377,856,881 bytes): existing delta JSON, saved patches, selected discovery responses, repository metadata and sampling definitions. It is a selected upstream corpus, not every historical local research artifact. Early CSV-only collections and some collection-side indexes/schedules are outside this snapshot.
 
-## Provenance, limits and citation
+Delta JSON is the collector's saved representation, not necessarily a full API response. Patches retain the collector's available file diffs and may be incomplete relative to the repository change. Repository observations and commit observations have different coverage. Account names and repository names reflect observed values, not guaranteed current identities. A source sampling CSV contains a preserved NUL anomaly marked by `contains_nul`.
 
-See PROVENANCE.md, LICENSES.md and manifests/release.json. Repository collection time is unknown where not present in the source and is kept null; disk mtime is not substituted. Observable agent-attribution evidence is not proof of causal model authorship. This build does not validate the full research pipeline or promote canonical/corrected results. Authorship/citation and any release DOI must be finalized before publication; CITATION.cff is explicitly a local draft.
+Use the repository URL and an immutable revision when referring to this dataset. See [DATA_DICTIONARY.md](DATA_DICTIONARY.md), [PROVENANCE.md](PROVENANCE.md) and [LICENSES.md](LICENSES.md) for details.

data/artifacts/commit-json-00000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c01fbc6c78149efeb3ea5f9364304cbda50cd295084aa34c00b7ec25d1e761b
-size 4439157
+oid sha256:31302007f2139f9c6e723f9d3bdea6767f219c1ede7210f30cb60a49b8bdc0d9
+size 4719392

data/artifacts/commit-json-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f342b1430b4c623d6a6d0c40f07fe65acbcc775c4c40eff5e341cbbf7d92ba33
-size 4427322
+oid sha256:257c27d73b00796aed2cdfae1abcb5a6c9df9f3d1da863725651d6edb47d2a54
+size 4697252

data/artifacts/commit-json-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:80bf2ca12b8aeed62f92212c5d4a4a6a311c683a5552c9265e16a3d3a2564b68
-size 4302221
+oid sha256:da79f8691c97e6b68470045cbaef62948c1988bd4058782bdfaa18e866b87c6e
+size 4559988

data/artifacts/commit-json-00003.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59f7ddc779cc1f56eb78d70300f5486d93db37f53f11f07e695964cb31388fd2
-size 4282652
+oid sha256:235a4ec2931de45ddb2d8d93f8d7d6cb386366256ef7756795def358f4fdb6e4
+size 4548219

data/artifacts/commit-json-00004.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39a9c544c2a55f28bf227d15d764d28b27637fff05fca3791bd3b40f1e6a93c6
-size 4288216
+oid sha256:0adb89e07a7cf7da52235ebc0b998b5d3d8254360a9fa1a541935d29bb16330f
+size 4564499

data/artifacts/commit-json-00005.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a75644ee20fd621a5193656f8eea423d17a5e326fbb50c71d27d28954e1ba9eb
-size 4277160
+oid sha256:b31d9fe2e971f25f604cfb81ec5a89183475596d2f2a3a8b746135228e568d09
+size 4529448

data/artifacts/commit-json-00006.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2ed53fee956fe69d2d562aa31e464903a136f4c839b0e2a2938001e166ffb03
-size 4306076
+oid sha256:5c268796cbe90b7062c9ad109463ee8f18fcc0c7ec207a93eb267de856fc6dcd
+size 4571046

data/artifacts/commit-json-00007.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7dc0ceee25a14be0ca669afb0a472d7a735935c9e013f8cd5ee44026987c5324
-size 4325753
+oid sha256:8f3a929d1d1e3a73c6dd8abc1f5d9572a48f471c3ac6af18530e0f19d55cf347
+size 4594376

data/artifacts/commit-json-00008.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82ade7ce814b9fcb1ad35e186fa25347dd4e3b421016f0e8af5409749a3c4cec
-size 4315987
+oid sha256:e836abb937a8279979070a00703da121e31796189f276669a5f3748859f2e6d2
+size 4589524

data/artifacts/commit-json-00009.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb526d5c465c72a8593f019678ded864f4f6ce97b8c63a85dd4d3d011b96472d
-size 4326467
+oid sha256:b144c303c4247b83410698fa57493e17cf70ab826bbfd45e9c681b5627259c93
+size 4593898

data/artifacts/commit-json-00010.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5185ad173c09adb9cce35d463a1498027d2045398d6a5237d4aa1952e22dcd62
-size 4335750
+oid sha256:4baf8ed26eee894d4577620c3036c92f2a500005e55818bdd7cf2049d4855d1f
+size 4595509

data/artifacts/commit-json-00011.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2924ce7a4a51fed111539574820e049ca5b5c4a04c24834726e376b25d9de066
-size 4316940
+oid sha256:50f2f4e3361344095a07c35bfa97aafcd297ba2ea167dbbbb66f323a5949b1e5
+size 4586822

data/artifacts/commit-json-00012.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:90de8c9b2c42271e52b865485ba9a27b400fd811c391c8ef826fa82480ddffcf
-size 4316725
+oid sha256:f1251b9c7e56f43f79eb5d0aebc492cec6c10683d2822e32edd5f15331ccb22f
+size 4588792

data/artifacts/commit-json-00013.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d332baa1fc5c75425d0202c6866c1ecb6b4b7b0b1b7e48f6e9710afd40812b13
-size 4331380
+oid sha256:4fa21b06494edf8d4a6ef2cad94cfc2debabf94ff68d1811a8a4a5d4b7a0cc93
+size 4605468

data/artifacts/commit-json-00014.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:336d51123c4d35dc2365950ad362b88bebb76f0dfe21d5eff59ab38c1ade42fc
-size 4327638
+oid sha256:bd16c4742353a70d0dc75cf33701daf502919560541cc6ca034e5c518261e7ba
+size 4588531

data/artifacts/commit-json-00015.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4232b8ec6f6266afcdae53ada50676659551598119e8af00325b533985fa21f
-size 4420442
+oid sha256:5518407081cfacec8e45b4d107e88d3030bc90b83b30b1382ea1ca70426fac94
+size 4669482

data/artifacts/commit-json-00016.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ffc8ea32dd271910e6f3ff5aaf2eca76403803740198258c7abce487fc65e7ab
-size 4317984
+oid sha256:c4a817c24d5ab3b575bc64ede580ed3a56406156ec3c49aadcc19035c28c5e5c
+size 4579406

data/artifacts/commit-json-00017.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f444100947b4d19431e3fe91fcfa49630e94b65d0a7902f340ad06f8369f3d70
-size 4298362
+oid sha256:bc5c7ee98b1e37c21270b116a9fde2722e5943539cdd18d6b96290ff560a9165
+size 4553185

data/artifacts/commit-json-00018.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fae9cff301620cd9a7717d1f0a085b55c387a355ead2fdacbd60e051e9abb1ed
-size 4318429
+oid sha256:ae4cf6b2060ba3e69b6a4205226dc7ff25a9b42015620795a6a0645845b24764
+size 4577039

data/artifacts/commit-json-00019.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2da8a7c36923428977e0f8d6b8da03aaf30108387ce2a3e9c1385f030bd94843
-size 4318015
+oid sha256:ea3cc3037e3ab1b6c52a5ef60c93d2d0f4225cb886e0e5f65c61dbf36ec5bda1
+size 4571999

data/artifacts/commit-json-00020.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:64d5eeb8bb27074929504bb4b2ec6cceceea5a40a2c5f179d57ee1d34f671410
-size 4325865
+oid sha256:f2019cc4dbc4a65888d5c9c789afcf175f83da21443ca985b5be7c0659188f23
+size 4582612

data/artifacts/commit-json-00021.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e1af6ddd52b201ed08bb5e7f4476342f6cda681608a31e08e2af5b436629f0fa
-size 4305220
+oid sha256:862ff134785f35b22f015d18c1c0f5e8879bcdbd454c23e31cd8b716ef320a80
+size 4569052

data/artifacts/commit-json-00022.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2e3d43aa304d73f644e95f1c6f3d92bdeffcdc42dcf121f70443888d9e1f1b15
-size 4304420
+oid sha256:85ab048fc59b0d503fe49d035e4e28dbda5ccba4e6519fcaa130bd9a1f84d2b6
+size 4562114

data/artifacts/commit-json-00023.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4406aefb73bb666a3b8132178f835ae7af0d0a6bd611f471f8cb83958842fa46
-size 4384878
+oid sha256:66bf5e634bab972454ff2414378e2f8de53fb5035aad0a7edc9257040fc93804
+size 4655953

data/artifacts/commit-json-00024.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a11f8ba36f226e436bdbd2deb967981ee46601b8011dd78ab2830eb1d87c4e0
-size 4383238
+oid sha256:b307d1c2c1e4732b6bfad917e578f6ec23c10b4dd4beae8357c9b60e3cdb80d3
+size 4642514

data/artifacts/commit-json-00025.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35dee22d1bd34ced0af1fe35f8e0fa22de95e36b2ed47bbc8aa9084b31b2eea1
-size 4354690
+oid sha256:857ccf11022d157ca5fd71ba638f6186f1f61674857107698e8eeeeb890dc363
+size 4616105

data/artifacts/commit-json-00026.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e7b5abd4287e599eaecd3bbd29b2ac49a6291b7729234fd8d450e5157f1ef50
-size 4360502
+oid sha256:1d41635afcd6e99e74f48c67f133f855ce969d4dbc11ae66eed92e23a14aa48f
+size 4641734

data/artifacts/commit-json-00027.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07eaebc4e5f11ed5ea5da9947fd37bd1fbe5378769027cc20ab05a161c8bc0db
-size 4366442
+oid sha256:1cbcfff697da2a45fcf2b4135d96b8e238b2ef521a824a9bb4ac1198559f4b91
+size 4632268

data/artifacts/commit-json-00028.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c022bddf2ad6eae4c103172b073341960003b15d1897ee7433fae4c88be0643d
-size 4377428
+oid sha256:df98b1c360eed5ae6d8bc021d524d50bac0838fb4d5f1f5fddc85d880008acf9
+size 4647210

data/artifacts/commit-json-00029.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0078a188a6500d8be46f1e501b01c749e2fff0d45d97a48c0f508e0bf278bdde
-size 4440044
+oid sha256:d6ceccafc580e127596fd4ad4f4a3eced4bc2c8413c4a6a8e70efe83033f8a2e
+size 4699310

data/artifacts/commit-json-00030.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84e941074183d94d7acac1192b6c533e2ead1c734a944e4e02de8882c4e38e20
-size 4088431
+oid sha256:41bed5f09c21c53cf60ab390be4c608f2f4d497a6ea1f75b03b961db60e493c5
+size 4302252

data/artifacts/commit-json-00031.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee4fc94f22bbc0a8d0310dc757d37f3a292057ecc02076546a1b96fdbf9db2db
-size 3691454
+oid sha256:27bc1af894ddf4176cd49722a3d1e942a03b0abbc508fd154fedbbfedd6ca61b
+size 3912950

data/artifacts/commit-json-00032.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b000a8c33dec19c1840d56d55f3235cec0117facd3667ac9cae2af02d6c81f5
-size 3691991
+oid sha256:78c435e7517127d31768c265b0445272572566fe134a20ad4159ff94a615257d
+size 3916441

data/artifacts/commit-json-00033.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd9c28736913a6c718629f75425296f1d14d77c81d17a200087d67c3e155da29
-size 3603740
+oid sha256:ebb4d5ae7df1b9c6a0d858e11bf451ad0d7580ed743ce4343953d1cf3df4cfd0
+size 3828016

data/artifacts/commit-json-00034.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a9720d265ece4f84bce3da4503c24f3e2a0907a29fd0a37ee77a5f610d60ad7
-size 4037747
+oid sha256:1f870d1ff79e1dc8f62ad4c68c43c902bfb19d7ff72450d23906cfff60042692
+size 4277708

data/artifacts/commit-json-00035.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3406cfcbb582271ceb0ded1e70fe7fea70ffdcc561ce35e61298d78fd3191d73
-size 4113767
+oid sha256:69606feb67e2ed8c5cbaae14d5711e5decad9a985e11fe0ff65943d1e01a1087
+size 4349527

data/artifacts/commit-json-00036.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ead05db97558b8fef6643b268284f34e67129fda21236ffcdeb57f9b134cfc41
-size 4373474
+oid sha256:5f266e7ebe9f7cbfd74a6cf71f0ac186f45f87d3a99df9e48e0bb57ba3483908
+size 4651569

data/artifacts/commit-json-00037.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:25f88132f4fbc89e38613a99e9827e0908ebdb7291afc62bf1d404994af6dbe8
-size 4393523
+oid sha256:af1004244292347c00e71a2685da1c0913edc7c54a8c524d2e3faea03eaf8b83
+size 4655253

data/artifacts/commit-json-00038.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da23f6ddc5c9a60ebc74bd8a04400d44d6766727c8ea01df399b8f0ac21571df
-size 4310348
+oid sha256:f339746f63f647b1848e1b69413cd3f069d68429c814337ace07c5508c60050d
+size 4566394

data/artifacts/commit-json-00039.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b8b061dda69496423f0a41765c9f4ba4c736dbbed42db272cfbb8fc7fe1b7af0
-size 4423816
+oid sha256:5d791927dceb253de54ba0fe2d75fbac082ce0c6b1d4ce09ab72ef8dce85fba5
+size 4684248

data/artifacts/commit-json-00040.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f5f78d7f0e862616aa9a79a93b0d17b6c2ae07ff2b3db8980c6857c45daa9cc6
-size 4357802
+oid sha256:73761cb756fa9486e2b07b7458933448d1294f052b8f73d322f25a44a718e05c
+size 4616726

data/artifacts/commit-json-00041.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f032d7854f4bd7315002bf7601cfcf07977cf8840a19e1a123fe3b39f2455728
-size 4234759
+oid sha256:3797ac164ed6d41b9355ad234edd28b1b0dc3e7de2b74871c03899e1cc3a40fb
+size 4477577

data/artifacts/commit-patches-00000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:82388b090f9f60e0955e988317e2236e71b099dbbf3b1fa2b163a756f3743888
-size 4464604
+oid sha256:1c5233492a8931a6b22a63361352f647d0225bb445813d409033724a7d8d2b15
+size 4729053

data/artifacts/commit-patches-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b8617047e87f0c5381fa5392610ddc22060585862c2aeb9c51a02cb701994aa
-size 3504500
+oid sha256:b4fd7d5d074c841095bdcf3ce04f58ea638ce5f7da406c2cd75370a43e51ac2f
+size 3701594