Add 2011-02-15 — 40.9K events, 11 files

README.md

@@ -51,60 +51,90 @@ configs:
   data_files: "data/public_events/**/*.parquet"
 - config_name: discussions
   data_files: "data/discussions/**/*.parquet"
+- config_name: live
+  data_files: "today/raw/**/*.parquet"
 ---
 
 # OpenGitHub
 
-Every public GitHub event since 2015, parsed into 16 structured Parquet tables, ready for analysis, training, and research.
+## What is it?
 
-OpenGitHub covers **2011-02-12** to **2011-02-14** (3 days), totaling **108,663 events**. The original 19.7 MB of raw GH Archive NDJSON is compressed into 10.9 MB of Zstd-compressed Parquet, partitioned as `data/TABLE/YYYY/MM/DD.parquet` (one file per event type per day). All 16 event types are fully flattened into typed columns, so there is no JSON parsing needed downstream. The dataset is updated daily and released under the [Open Data Commons Attribution License (ODC-By) v1.0](https://opendatacommons.org/licenses/by/1-0/).
+This dataset contains every public event on GitHub: every push, pull request, issue, star, fork, code review, release, and discussion across all public repositories. GitHub is the world's largest software development platform, home to over 200 million repositories and the daily work of tens of millions of developers, from individual open-source contributors to the engineering teams behind the most widely used software on earth.
 
-The underlying data comes from [GH Archive](https://www.gharchive.org/), created by [Ilya Grigorik](https://www.igvita.com/), which has been recording every public GitHub event via the [Events API](https://docs.github.com/en/rest/activity/events) since 2011.
+The archive currently spans from **2011-02-12** to **2011-02-15** (4 days), totaling **149,533 events** across 16 fully structured Parquet tables. New events are fetched directly from the GitHub Events API every few seconds and committed as 5-minute Parquet blocks through an automated live pipeline, so the dataset stays current with GitHub itself.
 
-## Why this dataset?
+We believe this is the most complete and regularly updated structured mirror of public GitHub activity available on Hugging Face. The original 27.5 MB of raw GH Archive NDJSON has been parsed, flattened, and compressed into 15.0 MB of Zstd-compressed Parquet. Every nested JSON field is expanded into typed columns — no JSON parsing needed downstream. The data is partitioned as `data/TABLE/YYYY/MM/DD.parquet`, making it straightforward to query with DuckDB, load with the `datasets` library, or process with any tool that reads Parquet.
 
-GitHub is the world's largest open-source platform, with over 200 million repositories, millions of developers, and billions of events. [GH Archive](https://www.gharchive.org/) captures every public event as gzipped NDJSON, one file per hour. Querying raw JSON at this scale is painful.
+The underlying data comes from [GH Archive](https://www.gharchive.org/), created by [Ilya Grigorik](https://www.igvita.com/), which has been recording every public GitHub event via the [Events API](https://docs.github.com/en/rest/activity/events) since 2011. Released under the [Open Data Commons Attribution License (ODC-By) v1.0](https://opendatacommons.org/licenses/by/1-0/).
 
-OpenGitHub fixes that. Every event is parsed, every nested field is flattened into typed Parquet columns, and the output is partitioned by date. You can `SELECT` stars, issues, PRs, and commits directly with DuckDB, Spark, pandas, or HuggingFace Data Studio.
+## Live data (today)
 
-Every field from the original GitHub API response is preserved. You can reconstruct the original event from these tables.
+Events from today are captured in near-real-time from the GitHub Events API and stored as 5-minute blocks in `today/raw/YYYY/MM/DD/HHMM.parquet`. Each block contains a generic event record with the full JSON payload preserved for later processing. Live blocks are committed to this dataset within minutes of the events occurring.
+
+
+### Live event schema
+
+| Column | Type | Description |
+|---|---|---|
+| `event_id` | string | Unique GitHub event ID |
+| `event_type` | string | Event type (PushEvent, IssuesEvent, etc.) |
+| `created_at` | timestamp | When the event occurred |
+| `actor_id` | int64 | User ID |
+| `actor_login` | string | Username |
+| `repo_id` | int64 | Repository ID |
+| `repo_name` | string | Full repository name (owner/repo) |
+| `org_id` | int64 | Organization ID (0 if personal) |
+| `org_login` | string | Organization login |
+| `action` | string | Event action (opened, closed, started, etc.) |
+| `number` | int32 | Issue/PR number |
+| `payload_json` | string | Full event payload as JSON |
+
+```python
+# Query today's live events with DuckDB
+import duckdb
+duckdb.sql("""
+  SELECT event_type, COUNT(*) as n
+  FROM read_parquet('hf://datasets/open-index/open-github/today/raw/**/*.parquet')
+  GROUP BY event_type ORDER BY n DESC
+""").show()
+```
 
 ## Events per year
 
 ```
-  2011  ██████████████████████████████  108.7K
+  2011  ██████████████████████████████  149.5K
 ```
 
 | Year | Days | Events | Avg/Day | Raw Input | Parquet Output | Download | Process | Upload |
 |------|-----:|-------:|--------:|----------:|---------------:|---------:|--------:|-------:|
-| 2011 | 3 | 108,663 | 36,221 | 19.7 MB | 10.9 MB | 27s | 36s | 1m07s |
+| 2011 | 4 | 149,533 | 37,383 | 27.5 MB | 15.0 MB | 46s | 51s | 1m27s |
 
 
 ### Pushes per year
 
 ```
-  2011  ██████████████████████████████  57.8K
+  2011  ██████████████████████████████  80.4K
 ```
 
 
 ### Issues per year
 
 ```
-  2011  ██████████████████████████████  4.1K
+  2011  ██████████████████████████████  6.0K
 ```
 
 
 ### Pull requests per year
 
 ```
-  2011  ██████████████████████████████  2.5K
+  2011  ██████████████████████████████  3.5K
 ```
 
 
 ### Stars per year
 
 ```
-  2011  ██████████████████████████████  11.9K
+  2011  ██████████████████████████████  16.9K
 ```
 
 
@@ -126,6 +156,11 @@ ds = load_dataset("open-index/open-github", "issues",
 
 # Load all pull requests into memory
 ds = load_dataset("open-index/open-github", "pull_requests")
+
+# Query today's live events
+ds = load_dataset("open-index/open-github", "live", streaming=True)
+for row in ds["train"]:
+    print(row["event_type"], row["repo_name"], row["created_at"])
 ```
 
 ### DuckDB
@@ -483,28 +518,25 @@ GitHub Discussions lifecycle: created, answered, category_changed, labeled, and
 
 | Table | GitHub Event | Events | % | Description |
 |-------|-------------|-------:|---:|-------------|
-| `pushes` | PushEvent | 57,768 | 53.2% | Git pushes with commits |
-| `issues` | IssuesEvent | 4,113 | 3.8% | Issue lifecycle events |
-| `pull_requests` | PullRequestEvent | 2,468 | 2.3% | PR lifecycle events |
-| `stars` | WatchEvent | 11,931 | 11.0% | Repository stars |
-| `forks` | ForkEvent | 2,912 | 2.7% | Repository forks |
-| `creates` | CreateEvent | 21,881 | 20.1% | Branch/tag/repo creation |
-| `deletes` | DeleteEvent | 1,713 | 1.6% | Branch/tag deletion |
-| `commit_comments` | CommitCommentEvent | 1,420 | 1.3% | Comments on commits |
-| `wiki_pages` | GollumEvent | 3,560 | 3.3% | Wiki page edits |
-| `members` | MemberEvent | 808 | 0.7% | Collaborator additions |
-| `public_events` | PublicEvent | 89 | 0.1% | Repo made public |
+| `pushes` | PushEvent | 80,381 | 53.8% | Git pushes with commits |
+| `issues` | IssuesEvent | 5,991 | 4.0% | Issue lifecycle events |
+| `pull_requests` | PullRequestEvent | 3,477 | 2.3% | PR lifecycle events |
+| `stars` | WatchEvent | 16,898 | 11.3% | Repository stars |
+| `forks` | ForkEvent | 4,090 | 2.7% | Repository forks |
+| `creates` | CreateEvent | 28,340 | 19.0% | Branch/tag/repo creation |
+| `deletes` | DeleteEvent | 2,425 | 1.6% | Branch/tag deletion |
+| `commit_comments` | CommitCommentEvent | 1,938 | 1.3% | Comments on commits |
+| `wiki_pages` | GollumEvent | 4,710 | 3.1% | Wiki page edits |
+| `members` | MemberEvent | 1,151 | 0.8% | Collaborator additions |
+| `public_events` | PublicEvent | 132 | 0.1% | Repo made public |
 
 ## How it's built
 
-The pipeline processes each hourly GH Archive file in a single pass:
+The pipeline has two modes that work together:
+
+**Archive mode** processes historical GH Archive hourly dumps in a single pass per file: download the `.json.gz`, decompress and parse each JSON line, route by event type to one of 16 handlers, flatten nested JSON into typed columns, write to Parquet with Zstd compression, and publish daily to HuggingFace.
 
-1. **Download** the hourly `.json.gz` file from `data.gharchive.org` (typically 100 to 150 MB compressed per hour)
-2. **Decompress** and parse each line as a JSON event
-3. **Route** each event by its `type` field to one of 16 handlers
-4. **Flatten** nested JSON into typed columns. For example, `issue.user.login` becomes `issue_user_login` and `pull_request.head.repo.name` becomes `pr_head_repo_name`
-5. **Write** to Parquet with Zstd compression, partitioned as `data/TABLE/YYYY/MM/DD.parquet`
-6. **Publish** daily to HuggingFace with auto-generated stats
+**Live mode** captures events directly from the GitHub Events API in near-real-time. Multiple API tokens poll concurrently with adaptive pagination (up to 300 events per cycle). Events are deduplicated by ID, bucketed into 5-minute blocks by their `created_at` timestamp, and written as Parquet files. Each block is pushed to HuggingFace immediately after writing. On each hour boundary, the corresponding GH Archive file is downloaded and merged into the typed daily tables for complete coverage.
 
 All scalar fields are fully flattened into typed columns. Variable-length arrays (commits, labels, assets, topics, assignees) are stored as native Parquet LIST columns — no JSON strings. All `*_at` timestamp fields use the Parquet TIMESTAMP type (UTC microsecond precision), so DuckDB, pandas, Spark, and the HuggingFace viewer all read them as native datetimes.
 
@@ -512,10 +544,10 @@ No events are filtered. Every public event captured by GH Archive appears in the
 
 ## Known limitations
 
-- **Full coverage starts 2015-01-01.** Events from 2011-02-12 to 2014-12-31 are included but parsed from the deprecated Timeline API (see above).
+- **Full coverage starts 2015-01-01.** Events from 2011-02-12 to 2014-12-31 are included but parsed from the deprecated Timeline API format, which has less detail for some event types.
 - **Bot activity.** A significant fraction of events (especially pushes and issues) are generated by bots such as Dependabot, Renovate, and CI systems. No bot filtering is applied.
 - **Event lag.** GH Archive captures events with a small delay (roughly minutes). Events during GitHub outages may be missing.
-- **Coverage starts 2015-01-01 for full data.** Events from 2011-02-12 to 2014-12-31 are parsed from the deprecated Timeline API. IssuesEvent and IssueCommentEvent from that period contain only integer IDs (no title, body, or state) because the old API did not include full objects in event payloads.
+- **Pre-2015 limitations.** IssuesEvent and IssueCommentEvent from 2012-2014 contain only integer IDs (no title, body, or state) because the old API did not include full objects in event payloads.
 
 ## Personal information
 

data/commit_comments/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b65152134467f7e6e46388ab6a18f513b6036b588ac1af3189f39199c14e068
+size 29826

data/creates/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5054e84971d660c40fb517e4e5e04a139baaa6a47bfc81dfb7bf24b4d56aa68
+size 144847

data/deletes/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b09af1092f50a51fc00ddeeaadd41a978cd93c407d306b23c717f6aa7aec0666
+size 20561

data/forks/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:895c5f8ddacb2087f234455756577ecc682ce1b00851f7497514be4472d73e1d
+size 49989

data/issues/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e45c6c0b03ef3dc9a8f4df0569f66a9560e57b42be838be165e6d2f1a7d0c7ef
+size 67530

data/members/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:115e91c520baae87d98276395fbaf437479af0974434e5dcfcf3072112f9cbd7
+size 15316

data/public_events/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65dda3d945db39b2a7c9a0e6cd35100fd4e3a30143240b690d273baff199a71f
+size 4270

data/pull_requests/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8bda2597a7af361ba7f8963a04d0f346c6b83024f220e96da2af7dcecf90315
+size 67216

data/pushes/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0021704e9f5eb379a3873e6bb8dd8ac39689f2e3d6cddd66caa40de52bc8ced
+size 3713701

data/stars/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4c25f17ed30c8af1a062519430e0ae05f403562e614863637b0958eb7d530d5d
+size 148133

data/wiki_pages/2011/02/15.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d4e3fa5e76d9c70e7de58f1edd9e064dbf47967bcfdb5d2dea9be848b932d2a
+size 90832

stats.csv

@@ -1,4 +1,5 @@
 date,total_events,parse_errors,pushes,issues,issue_comments,pull_requests,pr_reviews,pr_review_comments,stars,forks,creates,deletes,releases,commit_comments,wiki_pages,members,public_events,discussions,bytes_read,dur_seconds,parquet_bytes,dur_download_s,dur_process_s,dur_commit_s
 2011-02-12,30100,0,16694,1161,0,799,0,0,3517,930,4842,466,0,410,1029,228,24,0,5912248,9.9,3288866,5.9,9.9,31.8
 2011-02-13,37082,0,18277,1332,0,667,0,0,3692,822,10106,455,0,451,1035,209,36,0,6432532,12.1,3596553,15.3,12.1,35.5
-2011-02-14,41481,0,22797,1620,0,1002,0,0,4722,1160,6933,792,0,559,1496,371,29,0,8283118,14.1,4520871,5.8,14.1,0.0
+2011-02-14,41481,0,22797,1620,0,1002,0,0,4722,1160,6933,792,0,559,1496,371,29,0,8283118,14.1,4520871,5.8,14.1,20.3
+2011-02-15,40870,0,22613,1878,0,1009,0,0,4967,1178,6459,712,0,518,1150,343,43,0,8224882,15.1,4352221,19.4,15.1,0.0