RowCoveAccount ↗Inspect columnar data without a cloud upload
A feature export, training split or partitioned market history often needs a focused inspection before a larger analysis. RowCove lets you choose a local file or dataset folder, review the detected reader and begin with useful records.
Free Preview needs no account and returns a read-only first page of up to 100 records where the reader permits. It can help establish the source structure and types. Pro adds supported query and output workflows; a complete folder-wide aggregate is a separate requirement, not something inferred from a visible page.
Use the schema to choose columns
Parquet's columnar layout makes the required fields part of the access decision. Project the columns that answer the question rather than decoding unrelated text, features or wide nested structures. Inspect exact values and returned types before comparing them with another source.
For market data, verify timestamp units, timezone and price precision. For training features, check identifiers, split meaning and nested values. These example choices use synthetic mock data only.
| Question | Selected fields | Mock scope |
|---|---|---|
| Inspect a price record | symbol, timestamp, close | DEMO-A · one day |
| Check a training example | record_id, text, split | mock_001 · demo split |
| Compare feature types | record_id, feature_score | synthetic features |
Scroll horizontally to inspect every field. Keyboard: focus the table and use the arrow keys.
Select partitions before expanding the scan
Open a supported Parquet dataset folder to discover its shards and recognized Hive-style partitions. A folder layout such as split=demo or year=2026/month=01 can expose partition values that help scope the task. Only recognized predicates can support pruning; not every arbitrary filter avoids reading a file.
- Choose the intended dataset root and version.
- Review partitions and select the smallest useful scope.
- Project the fields required for the result.
- Browse bounded pages and preserve any continuation cursor.
- Keep source-file provenance with findings.
Schema differences across shards need investigation. Bounded discovery does not establish complete schema reconciliation, global ordering, an exact total row count or a global aggregate. Follow the partitioned Parquet and Arrow tutorial for the scope and continuation rules.
Match performance expectations to the operation
Column selection and supported partition pruning can reduce unnecessary data access and decoding. They do not make every full scan or sort instantaneous. Storage, compression, nested values, partition layout and the selected task still affect latency and resource use.
The published 20 GB CSV comparison measures CSV, not Parquet throughput. Use the performance model and test a representative local source before expanding a job. For very large collections, the TB-scale access planning guide covers fields, scan scope, memory and disk budgets.
Continue with a deliberate local workflow
The Parquet reader supports schema inspection, projection, bounded filtering/sorting and supported new-artifact workflows. For a supported bounded source, use Pro to stage cell edits and Save copy to a new destination. Parquet and Arrow edit-copy currently require a source no larger than 128 MiB, with at most 200,000 rows and 512 columns. Inspection and query support do not make a larger source editable. Dataset-folder browsing is read-only; open an eligible individual shard for a correction. Check the columnar editing limits and steps before buying Pro for this task. Arrow IPC and Feather have their own columnar readers; a CSV viewer workflow and a JSONL record workflow have different access costs and semantics.
Use authorized local MCP tools when you want an assistant to request a small result. Dataset processing stays local, while returned tool results are shared with the connected client. Review format-specific capabilities and the introductory offer and standard Pro renewal terms before committing to a larger workflow.
Try the first page on your own data.
Start on Windows or macOS: source information and up to 100 records, read-only. Review reader capabilities before planning a full-file Pro workflow.
Version 0.1.1 · Windows x64 · macOS Apple Silicon · Install guide →
Reader capabilities →