fix(dataset): unify commit metadata and simplify entity tables

CHANGELOG.md

@@ -1,3 +0,0 @@
-# Data format
-
-The tables use schema `upstream-readable-v3`: commit lookup locations, direct patch storage, compact changed-file metadata and typed sampling fields. Field definitions are in DATA_DICTIONARY.md.

DATA_DICTIONARY.md

@@ -1,284 +1,117 @@
 # Data dictionary
 
-Schema: upstream-readable-v3. Null means unavailable or unresolved, not zero.
+Null means unavailable or unresolved. Zero is an explicit numeric value.
 
 ## commits
 
-One saved commit record, with its message, author, committer, dates and collection labels.
+One commit within a repository.
 
 | 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. |
+| `commit_id` | `string` | Dataset key for a repository-scoped commit; use it to join the entity tables. |
 | `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 matching discovery records, including repeated observations. |
-| `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. |
+| `repo_full_name` | `string` | Recorded repository name in owner/repository form. |
+| `commit_sha` | `string` | Full Git commit SHA. The same SHA can belong to different repositories. |
+| `commit_url` | `string` | GitHub commit URL from the source response. |
+| `repo_owner_login` | `string` | Owner component of the recorded repository name. |
+| `message` | `string` | Complete commit message, including line breaks and co-author trailers. |
+| `identifications` | `list<item: struct<agent: string, channel: string>>` | Distinct agent/channel pairs attached to the commit. |
+| `author_name` | `string` | Name in the Git author entry. |
+| `author_email` | `string` | Email in the Git author entry. |
+| `author_date_utc` | `string` | Git author timestamp converted from an explicitly zoned source value to UTC. |
+| `github_author_login` | `string` | GitHub login associated with the Git author. |
+| `github_author_id` | `int64` | Numeric GitHub account ID associated with the author. |
+| `github_author_type` | `string` | Author account type, such as User or Bot. |
+| `committer_name` | `string` | Name in the Git committer entry. |
+| `committer_email` | `string` | Email in the Git committer entry. |
+| `committer_date_utc` | `string` | Git committer timestamp converted from an explicitly zoned source value to UTC. |
+| `github_committer_login` | `string` | GitHub login associated with the Git committer. |
+| `github_committer_id` | `int64` | Numeric GitHub account ID associated with the committer. |
+| `github_committer_type` | `string` | Committer account type, such as User or Bot. |
+| `parent_shas` | `list<item: string>` | Ordered parent commit SHAs recorded by the source. |
+| `files_count` | `int64` | File-entry count in the response selected by primary_change_set_id. |
+| `additions` | `int64` | Added lines reported by the source response. |
+| `deletions` | `int64` | Deleted lines reported by the source response. |
+| `repository_observation_id` | `string` | Key of the corresponding repository observation. |
+| `primary_change_set_id` | `string` | change response selected for this commit's summary statistics. |
+| `metadata_conflict_fields` | `list<item: string>` | Fields left null because source observations disagree. |
 
-## discovery
+## repositories
 
-One collected API record. The same commit can appear in several records.
+One repository metadata observation.
 
 | 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. |
+| `repository_observation_id` | `string` | Key of the corresponding repository observation. |
 | `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. |
+| `repo_full_name` | `string` | Recorded repository name in owner/repository form. |
+| `owner_login` | `string` | Owner login recorded in this repository observation. |
+| `owner_id` | `int64` | Numeric GitHub account ID of the owner. |
+| `owner_type` | `string` | Owner account type, usually User or Organization. |
+| `language` | `string` | Primary repository language reported by GitHub. |
+| `stars` | `int64` | Stargazer count in the repository observation. |
+| `forks` | `int64` | Fork count in the repository observation. |
+| `license_key` | `string` | License key recorded in the repository response. |
+| `repository_updated_at` | `string` | Repository update timestamp reported by the GitHub API. |
 
 ## commit_files
 
-One changed-file entry from a saved commit delta.
+One changed file in a commit response.
 
 | 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. |
+| `commit_id` | `string` | Dataset key for a repository-scoped commit; use it to join the entity tables. |
+| `change_set_id` | `string` | Identity of the change response containing this file entry. |
+| `file_index` | `int64` | Zero-based position in that response's files array. |
 | `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. |
-| `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. |
+| `previous_filename` | `string` | Previous path for a renamed file, when recorded. |
+| `change_type` | `string` | File change type, such as added, modified, removed or renamed. |
+| `blob_sha` | `string` | Git blob SHA recorded for the file. |
+| `file_language` | `string` | File language recorded by the collector. |
+| `additions` | `int64` | Added lines reported by the source response. |
+| `deletions` | `int64` | Deleted lines reported by the source response. |
+| `has_patch` | `bool` | Whether a patch snippet is available for this file entry. |
 
 ## patches
 
-One saved patch file as text. This version has 139,721 empty strings, 1,510,418 nonempty strings and no null patch values. Empty files are preserved source bytes; they do not imply no changes or no dependencies. Nonempty text may still be partial.
+One patch file with its commit ID and content hash.
 
 | Field | Type | Meaning |
 |---|---|---|
-| `artifact_id` | `string` | Path-based source-file identity; distinct from its content hash. |
-| `repo_full_name` | `string` | Repository label decoded from the saved directory path. Path normalization can differ from the GitHub name in commits; join through artifact_id, not this label. |
-| `commit_sha` | `string` | Exact Git commit SHA for identity checks. Join the patch through artifact_id; its repository label comes from the source path. |
-| `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; an empty string is a known zero-byte file. Null is unavailable/undecodable text, not an empty file. Nonempty does not guarantee completeness. |
-| `text_encoding_status` | `string` | UTF-8 decoding status. All saved patches in this version are utf8 and preserve the original bytes on encoding. |
-| `bytes` | `int64` | Original file size; zero for an empty source file. Filter bytes > 0 only for tasks requiring text, not to redefine a commit population. |
+| `patch_id` | `string` | Identity of the original patch file; distinct from its content hash. |
+| `commit_id` | `string` | Dataset key for a repository-scoped commit; use it to join the entity tables. |
+| `patch` | `string` | Original patch decoded as UTF-8, preserving its bytes when re-encoded. |
+| `bytes` | `int64` | Original patch file size in bytes. |
+| `sha256` | `string` | SHA256 of the complete original patch file bytes. |
 
-## repositories
+## Keys and missing values
 
-One original repository-metadata observation.
+- Commit identity uses a recorded numeric repository ID where unambiguous; otherwise it remains scoped to the recorded repository name. Commit identity includes the repository and SHA.
+- `repository_observation_id` selects a particular repository observation. Joining on `repo_id` alone can multiply rows.
+- `primary_change_set_id` selects the response used for the commit's change summary. Filter `commit_files.change_set_id` to that value when computing those totals.
+- When several change responses exist, the response with the lexicographically smallest source identity supplies the summary. This gives a stable choice for the summary. All file entries remain available.
+- Unavailable change statistics and primary_change_set_id are null.
+- Patch rows represent available patch files. Empty files have bytes=0 and an empty string; patch completeness is unspecified.
+- `identifications` is a list of structures, each containing `agent` and `channel`. A channel records the account, email or message signal used to identify the commit.
+- Git names and emails are commit attributes; GitHub logins and numeric IDs describe linked accounts.
+- The owner component of a recorded commit repository name can differ from a later repository observation after a rename or transfer. Owner ID and type belong to the linked repository observation.
+- Conflicting nonempty metadata values remain null in the summary and are named in metadata_conflict_fields. Source alternatives are located through indexes/sources and the original files.
+- Dates use UTC ISO 8601 strings. Dates require an explicit source timezone; other values remain null.
 
-| 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 storage location.
+## Supporting indexes
 
-| 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` | Archive/member/offset manifest for tar_archive; null for parquet_utf8. |
-| `storage_format` | `string` | tar_archive for archived sources (.tar or .tar.zst) or parquet_utf8 for patch text. |
-| `content_file` | `string` | Dataset-relative patch Parquet path for parquet_utf8; otherwise null. |
-| `content_row_group` | `int32` | Zero-based Parquet row group; select artifact_id within that group. |
+`indexes/commits/<SHA prefix>.parquet` contains commit_id, repo_full_name, commit_sha, commit_file and row_group. Recorded repository aliases can point to the same commit.
 
-## 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. |
-| `agent` | `string` | Agent label recorded by the collector. |
-| `channel` | `string` | Identification channel used for the query. |
-| `tier` | `string` | Original tier label where recorded; not inferred from other rows. |
-| `mode` | `string` | Collector search mode. |
-| `query` | `string` | Original GitHub search query. |
-| `sample_date_utc` | `string` | UTC calendar date of the sampled window. |
-| `window_start_utc` | `string` | Start of the recorded sampling window in UTC. |
-| `window_end_utc` | `string` | End of the recorded sampling window in UTC. |
-| `original_window_start_utc` | `string` | Start of the sampling window before query subdivision. |
-| `original_window_end_utc` | `string` | End of the sampling window before query subdivision. |
-| `segment_start_utc` | `string` | Start of this query segment in UTC. |
-| `segment_end_utc` | `string` | End of this query segment in UTC. |
-| `bucket` | `string` | Sampling bucket label from the source record. |
-| `status` | `string` | Collector outcome for the recorded query. |
-| `error` | `string` | Recorded error text, when present. |
-| `seed` | `int64` | Recorded random seed. |
-| `depth` | `int64` | Query subdivision depth. |
-| `total_count` | `int64` | Result count reported for the query. |
-| `fetched_unique_sha` | `int64` | Unique SHAs fetched for the recorded query. |
-| `attempts` | `int64` | Recorded number of query attempts. |
-| `incomplete_results` | `bool` | Whether the API reported incomplete results. |
-| `cap_bound` | `bool` | Whether the collector marked this query as cap-bound. |
-
-Direct columns copy existing fields from `fields_json`; blank or absent fields are null. Raw strings, including irregular values, remain in `fields_json`. Window timestamps are sampling boundaries, separate from commit author/committer dates.
-
-## Commit lookup index
-
-`indexes/commits/<prefix>.parquet` uses the first two characters of a full lowercase commit SHA. Each row locates one entry in `commits` and its saved patch. It covers the commit table, not every discovery observation.
-
-| Field | Type | Meaning |
-|---|---|---|
-| `repo_full_name` | `string` | Casefolded owner/repository lookup key. |
-| `commit_sha` | `string` | Exact full commit SHA. |
-| `commit_file` | `string` | Dataset-relative commit Parquet path. |
-| `commit_row_group` | `int32` | Zero-based row group; match repository and SHA within it. |
-| `patch_artifact_id` | `string` | Original patch identity. |
-| `patch_file` | `string` | Dataset-relative patch Parquet path. |
-| `patch_row_group` | `int32` | Zero-based row group; match patch_artifact_id within it. |
+`indexes/patches/<SHA prefix>.parquet` contains commit_id, patch_id, patch_file and row_group. It returns all patch files for a commit.
 
-## Identity and joins
+`indexes/sources/<SHA prefix>.parquet` contains commit_id, source_artifact_id, source_record_index and source_kind. A null source_record_index denotes the delta JSON file itself; API indexes identify a zero-based physical JSONL line or a JSON array item.
 
-`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.
+`indexes/repository_sources.parquet` locates each repository observation by source_artifact_id and source_record_index.
 
-## Agent labels and identification channels
+`indexes/artifacts/` locates original files using artifact_id, collection_id, kind, relative_path, bytes, sha256, member_manifest_partition, storage_format, content_file and content_row_group. The collection_id identifies a source-file group.
 
-`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.
+An artifact ID is SHA256 of UTF-8 `upstream + NUL + original relative path`. Patch IDs retain that file identity. Content hashes are stored separately as sha256.
 
-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.
+## Identification channels
 
-| 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"` |
+Author channels match agent accounts or email addresses. Co-author channels match message text such as co-author trailers. Marker channels match tool names or generated-with text. Channel names retain the labels recorded with each source observation.

PROVENANCE.md

@@ -1,32 +1,11 @@
-# Data sources
+# Sources and file access
 
-The data comes from GitHub API records and saved commit diffs. `discovery` contains the collected API records, their search queries, agent labels and identification channels. Query windows are recorded in `discovery` and `sampling`; commit author and committer dates are separate fields.
+The data comes from GitHub search and repository API responses, commit change responses and patch files. The four tables expose commit messages, identities, repository observations and code changes directly.
 
-Commit JSON uses the collector's delta format. Patches contain the file diffs available at collection time and may omit part of a commit's changes. Repository metadata is present where a matching observation was collected. Account and repository names reflect the values returned at collection time.
+Repository-scoped commit identities use numeric repository IDs when unambiguous. Recorded names remain visible. Source observations with the same identity contribute paired agent/channel labels and available metadata. Disagreeing nonempty metadata values remain null and are named in metadata_conflict_fields.
 
-Each row in `commits` combines a saved delta record with discovery metadata for the same repository and commit SHA. Git author and committer entries, linked GitHub accounts and collection labels have separate columns. When observations disagree on a field, `commits` leaves that field empty and lists it in `metadata_conflict_fields`. The values remain available in `discovery`.
+Commit and repository source references are under indexes/. To inspect an original response, locate its source_artifact_id in indexes/artifacts/. The member_manifest_partition identifies the compressed JSONL member inventory, which gives archive paths, member names and fragment offsets. Concatenate fragments in offset order and verify the original byte count and SHA256. Original record indexes locate the JSONL line or API array item within the recovered file.
 
-## Sampling and coverage
+Patch text is stored in data/patches/. Encoding patch as UTF-8 reproduces the bytes; bytes and sha256 verify that content. Patch IDs refer to original files, so distinct files remain separate even when their content hashes match.
 
-Matched sampling used a reproducible 10-minute window per day shared by Claude, Codex, Copilot and Cursor (seed 202506). Expanded sampling used a 4-hour window per day (240 minutes) for Codex, Copilot and Cursor. These sampled windows span June 2025 through May 2026. Their exact times and queries are recorded in `discovery` and `sampling`.
-
-Search exposure and detection rules differ across agents and channels. For comparisons, account for the recorded sampling windows and channels, and use a consistent counting unit. The calendar span describes the sampling period rather than continuous collection of GitHub activity.
-
-## Original files
-
-`artifacts` records each source file's path, byte count, SHA256 and storage location. It uses two storage formats:
-
-- `parquet_utf8`: read the row group named by `content_file` and `content_row_group`, select the row with the same `artifact_id`, and encode `patch` as UTF-8. Check the resulting byte count and SHA256. This also restores zero-byte files. The commit lookup example in README performs these checks.
-- `tar_archive`: open `member_manifest_partition`, select entries with the same `artifact_id`, and download the listed archives. Files ending in `.tar.zst` need Zstandard decompression; `.tar` files are uncompressed archives. Extract the named members and concatenate fragments in byte-offset order. Check the complete file against `artifacts.bytes` and `artifacts.sha256`.
-
-These recover the source file contents; archive timestamps and ownership are packaging metadata. Use the table dates and query windows for time-based analysis. When writing a recovered file, treat `relative_path` as relative to your chosen output folder and reject paths that escape it.
-
-The JSON, JSONL and CSV originals retain fields beyond the table projections. For example, the three file-content URLs are available in the original delta JSON: find it through `commit_files.source_artifact_id`, then select `file_index` in its files array. `manifests/archives.jsonl` lists archive objects and `manifests/SHA256SUMS` lists dataset file checksums, excluding itself and the platform-managed `.gitattributes`.
-
-One sampling CSV record contains 223 consecutive NUL bytes. The original file and `fields_json` retain them; `contains_nul` identifies that record. Direct sampling columns use the original field names. Empty fields become null; tier values are copied only where recorded, without assigning a sampling tier to other rows.
-
-## Empty patch text
-
-The patch table preserves 139,721 empty source files (8.4672%) and 1,510,418 nonempty files; no patch text is null in this version. Empty text is a saved zero-byte file, not evidence of transfer loss, no code changes or no dependencies. Keep those commits in the relevant denominator. A nonempty patch can still be incomplete.
-
-For a task that needs patch text, select `bytes > 0`. This filters text availability only; it is not a rule for excluding commits from an analysis. `artifact_id`, `bytes` and `sha256` retain the identity and checksum of empty files.
+SHA256SUMS lists delivery-file hashes. manifests/archives.jsonl describes the retained original archives; manifests/release.json records the table sizes and source-file inventory. These files support download verification and source inspection.

README.md

@@ -10,10 +10,10 @@ configs:
   data_files:
   - split: all
     path: data/commits/*.parquet
-- config_name: discovery
+- config_name: repositories
   data_files:
   - split: all
-    path: data/discovery/*.parquet
+    path: data/repositories/*.parquet
 - config_name: commit_files
   data_files:
   - split: all
@@ -22,187 +22,137 @@ configs:
   data_files:
   - split: all
     path: data/patches/*.parquet
-- config_name: repositories
-  data_files:
-  - split: all
-    path: data/repositories/*.parquet
-- config_name: artifacts
-  data_files:
-  - split: all
-    path: data/artifacts/*.parquet
-- config_name: sampling
-  data_files:
-  - split: all
-    path: data/sampling/*.parquet
 ---
 # AI Coding-Agent Commits on GitHub
+This dataset contains 1,853,915 commit records across 444,055 repositories, associated with Claude, OpenAI Codex, GitHub Copilot and Cursor. Messages and author identities are in commits; repository metadata, file changes and patch text are in separate tables.
 
-This dataset brings together GitHub commits associated with Claude, OpenAI Codex, GitHub Copilot and Cursor. It supports research on how coding agents appear in commit histories, with messages, identities, repository metadata and code changes available in separate tables.
-
-## Collection and size
+GitHub searches targeted daily windows between June 2025 and May 2026. Identification used agent accounts, email addresses, co-author trailers and tool names in commit messages. The matched collection used ten-minute daily windows; the expanded collection used four-hour daily windows for Codex, Copilot and Cursor. Counts describe this corpus and reflect each agent's search exposure.
 
-Records were collected through GitHub API searches for agent author accounts and email addresses, co-author trailers and tool names in messages. The queries cover sampled windows between June 2025 and May 2026, rather than continuous collection of every commit.
+For example, this query illustrates a four-hour search window:
 
-The collection combines **matched sampling**, with a shared 10-minute window per day for Claude, Codex, Copilot and Cursor, and **expanded sampling**, with a 4-hour window per day for Codex, Copilot and Cursor. Detection rules and search exposure differ by agent and channel, so raw agent counts should not be read as adoption rates. The recorded queries and windows are available in `discovery` and `sampling`.
+```text
+author-email:codex@openai.com author-date:2026-01-15T00:00:00Z..2026-01-15T04:00:00Z
+```
 
-| Measure | Count |
-|---|---:|
-| Commits (unique SHA) | 1,650,139 |
-| Repositories represented in `commits` | 414,783 |
-| Discovery records | 2,374,534 |
-| Repository metadata records | 441,371 |
+It finds commits whose Git author email is `codex@openai.com` and whose author timestamp falls between 00:00 and 04:00 UTC on January 15, 2026. The four hours describe the date range covered by the search. Other channels search agent accounts or message text, such as co-author trailers.
 
-Repository counts use case-insensitive `owner/repository` names. Discovery and repository metadata retain their own observation records.
+Search results provide the repository, commit SHA, message and URL. The repository and SHA are then used to retrieve commit details and patch text from the GitHub API. Results from different searches are merged by repository identity and SHA, retaining each agent/channel pair in `identifications`. See GitHub's [commit search syntax](https://docs.github.com/en/search-github/searching-on-github/searching-commits).
 
-Start with `commits` for messages, dates, authors and agent labels. Read `repositories` for repository details or `patches` for code changes.
+Start with **commits** for messages, identities and agent labels. Load repository details, file changes and patch text as needed.
 
-## Directory structure
+## Agents
 
-```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 storage pointers
-│   └── sampling/              Sampling definitions and query segments
-├── indexes/commits/          Commit and patch locations, grouped by SHA prefix
-├── raw/
-│   ├── commit-json/           Saved delta JSON 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
-```
+| Agent | Count |
+|---|---:|
+| Claude | 401,025 |
+| OpenAI Codex | 45,020 |
+| GitHub Copilot | 980,957 |
+| Cursor | 425,955 |
 
-Start with `data/` for analysis. Each Parquet file can be opened independently. `indexes/commits/` locates a commit and its patch without scanning every file. JSON, JSONL and CSV originals are in `raw/`; patch originals are stored as UTF-8 text in `data/patches/`. Their paths, byte counts and SHA256 checksums are recorded in `artifacts`.
+Each commit counts once per agent, regardless of identification channel. The 385 commits with multiple agent labels count in each relevant row.
 
 ## Tables
 
-Choose a table using its configuration name. Every table uses the split `all`. In `commits`, each row has a distinct SHA in this version. `discovery` preserves repeated observations. The 441,371 repository-metadata rows are observations, while 414,783 counts repository names represented in `commits`; these are different units.
-
-| Table / files | Rows | Size / files | Contents |
+| Table | Count | Size | Contents |
 |---|---:|---:|---|
-| [commits](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/commits) | 1,650,139 | 857.4 MiB / 42 files | 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) | 2,374,534 | 971.6 MiB / 102 files | 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) | 17,826,229 | 1,094.7 MiB / 102 files | One changed-file entry from a saved commit delta. |
-| [patches](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/patches) | 1,650,139 | 14,093.6 MiB / 294 files | One saved patch file as text, including 139,721 empty files. |
-| [repositories](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/repositories) | 441,371 | 28.3 MiB / 1 file | One original repository-metadata observation. |
-| [artifacts](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/artifacts) | 3,300,414 | 339.0 MiB / 123 files | One original source file and its storage location. |
-| [sampling](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/sampling) | 27,968 | 0.8 MiB / 1 file | One sampling record, with query, window and result fields. |
+| [commits](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/commits) | 1,853,915 | 782.3 MiB | One commit within a repository. |
+| [repositories](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/repositories) | 441,371 | 21.7 MiB | One repository metadata observation. |
+| [commit_files](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/commit_files) | 17,826,229 | 855.2 MiB | One changed file in a commit response. |
+| [patches](https://huggingface.co/datasets/ASSERT-KTH/agent-commits-raw/tree/main/data/patches) | 1,650,139 | 14,065.9 MiB | One patch file with its commit ID and content hash. |
+
+Repository totals count distinct IDs, or names where IDs are unavailable. The repositories table contains metadata observations; a repository may appear more than once.
 
-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/`.
+Each table consists of independently readable Parquet files. Select the tables and columns needed for your analysis. The configuration name is the table name, and the split is `all`.
 
-## Quick start
+## Files
+
+```text
+data/
+  commits/          Messages, identities and paired agent/channel labels
+  repositories/     Recorded repository metadata
+  commit_files/     Changed paths, change types and line counts
+  patches/          Patch text and content hashes
+raw/                Original API content
+indexes/            Record locations and source-file references
+manifests/          File inventories and checksums
+README.md
+DATA_DICTIONARY.md
+PROVENANCE.md
+LICENSES.md
+SHA256SUMS
+```
 
-Use Python with `datasets`, `huggingface_hub` and `pyarrow`. The examples use your locally configured Hugging Face token (`token=True`); do not put credentials in notebooks. The preview uses a fixed revision. The lookup example below resolves the current revision once for all its downloads.
+## Read commit metadata
 
 ```python
 from itertools import islice
 from datasets import load_dataset
+from huggingface_hub import HfApi
 
 REPO = "ASSERT-KTH/agent-commits-raw"
-REVISION = "9ad49e07344dc4138a5042163ecfa3292bdb8728"
-
+REVISION = HfApi().dataset_info(REPO).sha
 commits = load_dataset(
     REPO, "commits", split="all", revision=REVISION,
-    streaming=True, token=True,
-    columns=["repo_full_name", "commit_sha", "html_url", "observed_agents"],
+    streaming=True,
+    columns=["commit_id", "repo_full_name", "commit_sha",
+             "commit_url", "message", "identifications"],
 )
-for row in islice(commits, 3):
-    print(row)
+for commit in islice(commits, 3):
+    print(commit)
 ```
 
-Static preview from the same revision:
-
-| Repository | Commit SHA (short) | Agent labels |
-|---|---|---|
-| zzstoatzz/leaflet-search | `71cecc5fe66c` | claude |
-| zzstoatzz/leaflet-search | `9af6ef294580` | claude |
-| zml/zml | `bfb792342b42` | copilot |
-
-Only the selected columns are returned. For repository metadata, use the `repositories` configuration; neither metadata table requires patch archives.
+Save REVISION with your analysis. `identifications` keeps each agent and channel together. `message` includes the full text and co-author trailers.
 
-## Identities and agent labels
+## Find a commit and its patches
 
-`author_*` and `committer_*` describe the identities stored in Git; `github_author_*` and `github_committer_*` identify linked GitHub accounts when available. `repo_owner_*` identifies the repository's owning user or organization. Full co-author trailers remain in `message`.
-
-`observed_agents` and `observed_channels` list the labels attached to a commit. To see the agent, channel and search query for each observation, use `discovery`. Field definitions and channel queries are listed in [DATA_DICTIONARY.md](DATA_DICTIONARY.md).
-
-## Find a commit and read its patch
-
-Set the repository name and full SHA below. The first two SHA characters select one small index file, which points to the commit and patch Parquet files and their row groups. The complete index is 123.0 MiB across 256 files. This example downloads one index file and the two data files it identifies; it reads only the indexed row groups. For metadata only, stop after reading `commit`.
-
-All downloads use the same revision, resolved once at the start. Save `REVISION` with your analysis to reuse that snapshot.
+Set a recorded repository name and the full commit SHA. The small SHA-prefix index locates the metadata file and row group. Read the commit metadata first, then load its patches as needed.
 
 ```python
 import hashlib
-import re
 import pyarrow.parquet as pq
 from huggingface_hub import HfApi, hf_hub_download
 
 REPO = "ASSERT-KTH/agent-commits-raw"
-REVISION = HfApi().dataset_info(REPO, token=True).sha
+REVISION = HfApi().dataset_info(REPO).sha
 target_repo = "Tkute0401/Urban-Realty"
 target_sha = "8b0c1a43d2e80ad40929d4e949b7d49f6648cfa2"
-if not re.fullmatch(r"[0-9a-f]{40}", target_sha):
-    raise ValueError("Use the full lowercase 40-character commit SHA")
-
-def download(path):
-    return hf_hub_download(REPO, path, repo_type="dataset",
-                           revision=REVISION, token=True)
-
-index_path = download(f"indexes/commits/{target_sha[:2]}.parquet")
-locations = pq.read_table(index_path, filters=[
-    ("commit_sha", "=", target_sha),
-    ("repo_full_name", "=", target_repo.casefold()),
-]).to_pylist()
+
+def local(path):
+    return hf_hub_download(REPO, path, repo_type="dataset", revision=REVISION)
+
+prefix = target_sha[:2]
+locations = pq.read_table(local(f"indexes/commits/{prefix}.parquet"),
+    filters=[("commit_sha", "=", target_sha),
+             ("repo_full_name", "=", target_repo.casefold())]).to_pylist()
 if len(locations) != 1:
-    raise ValueError("No unique commit found for this repository and SHA")
+    raise ValueError("Select a unique repository identity from the matching records")
 location = locations[0]
-
-def read_one(file_key, group_key, predicate):
-    table = pq.ParquetFile(download(location[file_key]))
-    rows = table.read_row_group(location[group_key]).to_pylist()
-    matches = [row for row in rows if predicate(row)]
-    if len(matches) != 1:
-        raise ValueError("The indexed row could not be found uniquely")
-    return matches[0]
-
-commit = read_one("commit_file", "commit_row_group", lambda r:
-    r["commit_sha"] == target_sha and
-    r["repo_full_name"].casefold() == target_repo.casefold())
-patch = read_one("patch_file", "patch_row_group", lambda r:
-    r["artifact_id"] == location["patch_artifact_id"])
-payload = patch["patch"].encode("utf-8")
-assert len(payload) == patch["bytes"]
-assert hashlib.sha256(payload).hexdigest() == patch["sha256"]
+rows = pq.ParquetFile(local(location["commit_file"])).read_row_group(
+    location["row_group"]).to_pylist()
+commit = next(r for r in rows if r["commit_id"] == location["commit_id"])
 print(commit["message"])
-print(patch["patch"][:500])
+
+patches = pq.read_table(local(f"indexes/patches/{prefix}.parquet"),
+    filters=[("commit_id", "=", commit["commit_id"])]).to_pylist()
+for item in patches:
+    rows = pq.ParquetFile(local(item["patch_file"])).read_row_group(
+        item["row_group"]).to_pylist()
+    patch = next(r for r in rows if r["patch_id"] == item["patch_id"])
+    payload = patch["patch"].encode("utf-8")
+    assert len(payload) == patch["bytes"]
+    assert hashlib.sha256(payload).hexdigest() == patch["sha256"]
+    print(patch["patch"][:500])
 ```
 
-The example patch has 39,961 bytes. The same lookup works for other commits in `commits`, including saved empty patches. There are 139,721 empty patch records; empty text does not establish that a commit has no changes. Nonempty text may still be partial. See [DATA_DICTIONARY.md](DATA_DICTIONARY.md) for details.
+## Join the tables
 
-## Relationships and dates
+- `commits.repository_observation_id` joins to the same field in `repositories`, where owner type, owner ID, language, stars, forks and license are available.
+- `commit_files.commit_id` and `patches.commit_id` join to `commits.commit_id`.
+- For the file entries behind a commit's summary, filter `commit_files.change_set_id` to `commits.primary_change_set_id`.
+- Git author/committer names and emails belong to the commit. GitHub account fields describe linked accounts.
 
-- `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.
+Field definitions are in [DATA_DICTIONARY.md](DATA_DICTIONARY.md), source-file access in [PROVENANCE.md](PROVENANCE.md), and source rights in [LICENSES.md](LICENSES.md).
 
 ## Citation
 
-Use the dataset repository URL and a fixed commit revision when citing the data. Field definitions are in [DATA_DICTIONARY.md](DATA_DICTIONARY.md), collection and file details in [PROVENANCE.md](PROVENANCE.md), and source rights in [LICENSES.md](LICENSES.md).
+Cite the dataset URL together with the fixed revision used in your analysis.

data/commit_files/commit-json-00000-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6572936b5c59fcc729295891a2acfaeaaba87e2200247f28a69089012fc068f7
-size 17082572
+oid sha256:54042ab72459b4a3953db1f2314d0cf2ebfd3fa5faf8e42183b55c0da688121f
+size 12583751

data/commit_files/commit-json-00001-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f5c46d3c452ed73a03944ca02f84a74f3c7e576304386b94f6cb147e6a750a8
-size 15600522
+oid sha256:f8ec137062344324a7ebf1e742f389504f9a13948ddbc4a8cd70b4eaf59c1e04
+size 11906058

data/commit_files/commit-json-00001-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe28e7894ee78af0ac22cc2249bef585c8942ecf55557845317956019edda217
-size 10892713
+oid sha256:8aa0117cf7a74b18c3e14f9fab498dfd47138d93bf6215315e7eb269149f82a8
+size 8656845

data/commit_files/commit-json-00002-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:23361c570ee60f96bad6f92b55db825ce300d484fcc2c7d733bd43496f2661af
-size 5194340
+oid sha256:f1a6add201b33754e1ed6bad1505a51cea7ef893773f75a6e817e92d648612d9
+size 3936345

data/commit_files/commit-json-00003-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:75901d2aeec9a09c6a0a8bc52d05afd9c2f7e5640afea233c498717a95fcd244
-size 4662585
+oid sha256:32ec2b2f4a110cf8eaca140a9ed8620b64768cde936d325dd80f82581951ec6c
+size 3519125

data/commit_files/commit-json-00004-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb9d5448e66b56c23936df2bc821e230ead4561d7c879e9b3e3a04bec741c703
-size 3779811
+oid sha256:94c798d136250493d1b34803a79f0459d8a4fff0c6b784e1880092cf2b9ccc35
+size 2720305

data/commit_files/commit-json-00005-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6eb0c4e625117aedc242e06c2796afcc384bb3e0b0880c55c9d56252a5eb625
-size 5249528
+oid sha256:b515406d593115e53b11c3e68b0609f98a71e24b9ca8195b048be69cab67aa6a
+size 4068937

data/commit_files/commit-json-00006-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:688fb83895903708cbe508cdd3b4c3746486a0f635bcd6624ce4cd2bd46710e6
-size 11229604
+oid sha256:dcc097e4a27798bfcfb1163f260642f4e76921104c9afc77d7be99aa6bb8647c
+size 8589827

data/commit_files/commit-json-00007-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:183ebb6d967e8b7ae28866cbb60609d5bc2ccd7864893f1c504fe0bf1613d8f1
-size 14795828
+oid sha256:db795d3eb5d51ae32d24a50fff5db7b76d13a378672dcdd57eaf6fa9036eede6
+size 11479142

data/commit_files/commit-json-00007-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:895bd2dc5ba919a0e944e87de357d52f922522490f84d9bd68b5b28239559765
-size 13236742
+oid sha256:44b982f05ef861347f7f8731a5479aecc35905ea667d5127b99a17d473d546e0
+size 10279565

data/commit_files/commit-json-00008-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b9fd55d8d9f88a3b92bfaf063f33f78eec0d3c73684911466757dd97e73cd1e
-size 15460821
+oid sha256:92938573428ab1719796c359a14c20a90d756e084808a505cfd97ff6d58a69d7
+size 11929691

data/commit_files/commit-json-00008-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0c9a6edd60a11113c16731816277131f9f062ba785ae4b80c63d13cbd46890c5
-size 10690350
+oid sha256:6b52b18d27342378f36b6799ef87069efe65423f1ba34ca857f56bf73a5cde02
+size 8240905

data/commit_files/commit-json-00009-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c18a4641cc31cdff6679cb60f418b58d0705ae641eb79feb544ad006b618bba0
-size 14948735
+oid sha256:384620d0cadaab321b7f6cfcb1bdf199633279dea660f95adb4d70efc66b5e38
+size 11601402

data/commit_files/commit-json-00009-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3263d0aae2362f4b1eb4b0d2a0fa51d63beb1f7da854e81a4b30b8b2b7d37a73
-size 13240194
+oid sha256:2b146aa5904e659fa0ee4810d486fb4e17c857521f9ece37405a64bd3344fdc5
+size 10284119

data/commit_files/commit-json-00010-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a281841c3839d8735fb9d8ba943724ecda5c01a787bf26dc3624156aea818d66
-size 15568711
+oid sha256:8f1b9c99fec633e785a6c387bd1112c15a5a42951d133f0c84a106aced3fe954
+size 11986904

data/commit_files/commit-json-00010-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:48be1635f3596d0a6a32c8894fa5fa4471efd44b0c729a865a3b20ce52d5cc93
-size 9178911
+oid sha256:a0329120d8c32fc7a8f23bc4625006356c3dac4cc1dd0da1105dbb69a6ed25a7
+size 6997217

data/commit_files/commit-json-00011-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0052d3702e3f45e81d93a186879d29d677a1257213bda554b976826012a483b
-size 17228186
+oid sha256:e504ce409ec0ad395356f09080505868ddc2afd8b86183bb378245561931e5f6
+size 12978470

data/commit_files/commit-json-00011-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e72444bb89fd716e698a22a23ed9361aa821cbd9f6c8d5a97b1015c69092ee7e
-size 3794516
+oid sha256:2eb14289652e9f78c7c8213e4d26508737ecc04b4350be8642fa6777b5579243
+size 2874651

data/commit_files/commit-json-00012-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aba916fd244540c4cf5366fe5cb9bfcf7c80f64c0c4b6820a14b6a240be2658d
-size 16908065
+oid sha256:50011595d131fa52b190d37a62e99129768426319306bb9a0af142cea528e583
+size 12817335

data/commit_files/commit-json-00012-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fd05c01f9da9f3d9fd7d1ab9d65dbdec52d0ee706db26d68d241ca35bbd776eb
-size 5117461
+oid sha256:26fd46eda528324127ebd1272a4de1ac6a82f31bc4b77e038d71a821798a54cd
+size 3867403

data/commit_files/commit-json-00013-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f8f8e83cd4875de9ca0e0c3c02ab41c95446dc50e1dba49d216793fffdfa3529
-size 17014384
+oid sha256:b0c56bd205f38060cb393f2ef78d6985d4b3bf551d5e00e5da7aec9c07032603
+size 12895919

data/commit_files/commit-json-00013-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:165dca5d91635742fdf05a41bd96ef75615a41c6d5c8a454984578969c053db9
-size 4767146
+oid sha256:a458f3d69b0fb0a55e4fde8eeeee70b8e44db5b5db2d23668b9d195ad737b74a
+size 3601760

data/commit_files/commit-json-00014-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:948bf1b8c5d23e93b34a84f33b564514f6d1255ecad70a081758a6e968c39fd1
-size 16444550
+oid sha256:4244bea9bdab2eb3963821ff9500bfb9af0980403872fc474e1285982cbf28a9
+size 12530187

data/commit_files/commit-json-00014-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3c352fcfe6590d84c2f99d48e402784cefd2b53d6fbae26d32de979ff1f2cc7
-size 6455028
+oid sha256:334a374868ca82cc6ee3684e2590a02e2e9b29c6fa83646181e4564dcf54fdd8
+size 4896792

data/commit_files/commit-json-00015-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:29ea589288c22b2bbd1291cf0d1d966a5f2b1a7a75114f87f67f9691b1dc174e
-size 17174739
+oid sha256:93d7297fb7b24f7e67108b0c57399b91d8afdcdfef6f8929e04b4cb6e966b90a
+size 12922336

data/commit_files/commit-json-00015-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:99d47cb068957f7e50eb9693cc9cafb7f2f5e7521932330d5c1bf74b1c8a43c6
-size 3115021
+oid sha256:ce36b8272a19429185b193de2ca86a72b54081ba72074538dc0a4ef7ab96f723
+size 2322881

data/commit_files/commit-json-00016-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3dfd1ad54f2936f02237af969c681b3bb7723fbaae95291d5fbcd22336fda501
-size 15786451
+oid sha256:f9204c37547ffb72929472b280ec8ac1e40161a821f7ccdcb7c8a2d260ea2f5f
+size 11996432

data/commit_files/commit-json-00016-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c80555e648ab90ec1e893567506c38c6d6ca9da4cfa8986ed5fb58690368dd67
-size 8948779
+oid sha256:53ea2d01c1298f578c1cfe09807dc1ba5c64411efb1f98cc14d02ab91c2c4a6e
+size 7024070

data/commit_files/commit-json-00017-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef6df1d5f5006b21fde154c0eb2cfd1f42e05066d3d2eedf094ad9d4151a1d02
-size 13109936
+oid sha256:e81d3660f5f41c1370ea1d5f668ae2f35c11c368c7d085f2b80c1870b080134d
+size 10427533

data/commit_files/commit-json-00017-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:42fe96023fdf83d43833d876ec3c7827187023db24116e6855c7a2d7939d83a0
-size 12674581
+oid sha256:f826efb14b1baf996ab30a38077a57dd5365ba1d69180c1611d6fa3cf466e916
+size 10164780

data/commit_files/commit-json-00017-0002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c817a3e9c95a6cc02b1704253939f6a0e6e52ca81d12f3ef88f71b94f280567
-size 11192987
+oid sha256:e68a923b82d99700737e4d9cb8f12f023b854e9d9df79fc63cb67e3286a1285e
+size 8665064

data/commit_files/commit-json-00018-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b468371b9c5214abb9a740c95acfe7d2eb83bfb612d5731a6a34ec70aa9b909
-size 15541560
+oid sha256:109d264d84a0261674605a1c0cdad70f849b1b323b19ea695a222f3fc144a678
+size 11879719

data/commit_files/commit-json-00018-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a06996d7ec0e186b5e0f878b883057b3cc95ddcdbf8485e2d168f68c013de1f4
-size 10035544
+oid sha256:2176699ded7b7bf282a766e823e5fbfbb7e23e4a93f320a510847be556e6a0d6
+size 7748796

data/commit_files/commit-json-00019-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3e93a650cd9b977d40b0f0d20a94187ccc6d68ced6d811ecf70039cf2a21d5c
-size 14943523
+oid sha256:402f985b879548ddc17e33b4a7d79182ee03a6a36809a901123470b92c90fd7f
+size 11493600

data/commit_files/commit-json-00019-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a38187974f9198bfe670c5ea4570ea8ad17189d4e92b01405d6045a5855ddd74
-size 12198984
+oid sha256:b4abd6155952a8c647bcecc66bdcaecd4b2a3b10e00b44ac2c241da05c1fa413
+size 9432871

data/commit_files/commit-json-00020-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:259dbf11504748afc6466f3e2e31c19bb674d7e18cd7b3a2dcd68621a71b23ea
-size 15271351
+oid sha256:5c7b5190b95b79fd96aab8807d2f96389f3cfe68e282e3afc4ba472567440575
+size 11703017

data/commit_files/commit-json-00020-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d62c67e4b2ce03aad3144e81097d23d43869639110adc938df80ebd34ea1d51f
-size 8952878
+oid sha256:c917871e421bcbfbf453d013093cc61bbd16397c98b5f3a3ecc7c0fd9da6a447
+size 6767459

data/commit_files/commit-json-00021-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6535337ff230db077876e713d3ae356a364bdc4801a5bc7fd89b05211c77adc8
-size 15076906
+oid sha256:c160718c69523f3875d1ce1bf656d493f5cd98b56e1ce75c9147e2e4deda83c6
+size 11557515

data/commit_files/commit-json-00021-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:653782134abddab92046d00602dc14d40b450387848d78fa59723ee82100b2a2
-size 11660148
+oid sha256:1638bf2ab58436ef09694fb4b31fa5fa5001449ce151e19b5fbda27ca4282e6c
+size 9076671

data/commit_files/commit-json-00022-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5cb569df8e85e8faa34111533c2304d8e6414c4bf59b22f56ff8209669bff82b
-size 14819792
+oid sha256:939ad6cecd26cca92f51251537a4987c340700350c1e8fafae40295de5111250
+size 11525926

data/commit_files/commit-json-00022-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1916d3f1aedcbc366c1532daf9a508e1949656ee07c9ba1aecd3023b2724f4c
-size 13876330
+oid sha256:3e6686de3f5ca56000b560ca2b9d6f4bcec2c430093c019616d646e520ddfd2a
+size 10965612

data/commit_files/commit-json-00022-0002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:84657e1a07b3e49c1859742bb98c91dca54062aca5fd5c5958f8baa5006a3ef7
-size 2495686
+oid sha256:d380bcef0160ea9ce2e1736d4f8bb223ba913126c4539a307c986c05bb2ca76a
+size 1955118

data/commit_files/commit-json-00023-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef601ce96d705a22391a891e87d7e5ec700888c38acaedb6af0bd1ed71297008
-size 16461630
+oid sha256:a97d21a0ed62eaf3ddcfab31b03acc036dd6c71fab2d4251a1c967799fce5a83
+size 12557605

data/commit_files/commit-json-00023-0001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13d46d612c3c15cd328c72898eb5147644af4e24f4ed5f7e64e6012e2296ee76
-size 6266124
+oid sha256:39373948f9ac1c473580d08018e7fab2c367239057c1cb51d0e95c9b9688f6f2
+size 4806142

data/commit_files/commit-json-00024-0000.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47262410c16ebdcafc0ba6865aba21109659588c952662f266934cfd5aa3c95c
-size 16580836
+oid sha256:6b87f5e157475a6138e2cfe93f7c94c03f6227552817b820522ea2554402e178
+size 12606665