Public data index / Developer-tool experiments

Inspect a large data file without uploading it

This is the evidence and purchase contract for a proposed JetBrains plugin that profiles CSV, newline-delimited JSON, and Parquet files inside an IntelliJ-based IDE. It is aimed at a data engineering lead who needs a fast first answer about shape, nulls, ranges, and suspicious columns before opening a notebook or moving a file to another service.

Measured 2026-08-09: the bounded reference profiler read 1,073,741,824 bytes across 1,048,576 synthetic CSV rows, checked four fields per row, and returned the frozen expected bounds. This is a reference-core measurement, not a claim about an unbuilt plugin.

The full-read rule. The paid build cannot pass by loading only a convenient sample; the benchmark reads the entire one-gibibyte fixture. A future plugin release must reproduce the public acceptance suite on the packaged build. A preview that quietly samples the first rows cannot earn a passing performance claim.

$249 one-time founding-team pilot

One data engineering team receives a private JetBrains-compatible beta license, CSV, newline-delimited JSON, and Parquet profiling, the public one-gibibyte acceptance suite, and twelve months of compatibility updates. Delivery is within ten business days after one non-sensitive fixture shape and the supported IDE are accepted. This one-time pilot has no automatic renewal to cancel.

Request the $249 pilot scope

The request takes no payment, creates no subscription or contract, and does not activate a JetBrains Marketplace listing.

1,073,741,824 bytesThe synthetic CSV fixture was read from beginning to end.

1,048,576 rowsEvery fixed-length row entered the field checks.

849.09 MiB/sOne host run, including SHA-256 and field validation.

24.97 MiBObserved process peak RSS for the reference scan.

What the benchmark proves—and what it does not

The receipt is a real full-file scan, not a generated estimate. The script writes an exact one-gibibyte synthetic CSV fixture made of 1,048,576 one-kibibyte rows. It then reopens the fixture, reads every byte, verifies four fields on every row, parses the integer column, counts empty cells, computes minimum and maximum, and calculates SHA-256 while timing the scan. The run passes only when byte count, row count, column count, null count, and numeric bounds all equal the frozen expectations. The temporary fixture is deleted after the receipt is written.

The measured implementation is a bounded Python reference profiler, not an IntelliJ plugin. It establishes that the full-read acceptance shape is practical on this host and gives the future build a reproducible target. It does not establish IDE responsiveness, JVM memory, Windows performance, Parquet decoding speed, malformed-file behavior, or performance on a buyer's data. JSON and Parquet remain mandatory acceptance cases before a paid Marketplace listing can claim support. If the benchmark script cannot open, scan, validate, or write its receipt, it exits nonzero; the page returns 503 when either evidence file is missing or malformed rather than converting an unreadable probe into a zero or a pass.

Download the machine-readable benchmark receipt. Its SHA-256 covers the entire generated fixture, while the published limits preserve the distinction between one synthetic observation and product performance.

The proposed inspection contract

Included in the pilot

Local-only opening and profiling for CSV, newline-delimited JSON, and Parquet files.
Row and column counts, inferred scalar types, null counts, distinct-count estimates, numeric bounds, and a bounded value preview.
A cancellable background task with progress and an explicit incomplete result when the user stops it.
File size, format, scan mode, bytes read, and whether each metric is exact or estimated beside every result.
The public benchmark script and frozen receipt, plus a packaged-build reproduction report before delivery.
Twelve months of compatibility updates for the accepted JetBrains IDE family.

Explicitly excluded

No upload, telemetry, external API, model call, cloud processing, or automatic sharing of file contents.
No database import, schema migration, data cleaning, write-back, transformation pipeline, or claim that inferred types are authoritative.
No support for one giant JSON array in the pilot; JSON means newline-delimited records so scanning remains bounded.
No promise that compressed Parquet bytes equal decoded memory or that a one-gibibyte file is the largest safe input.
No legal, compliance, privacy, quality, or production-readiness conclusion about a dataset.
No Marketplace availability or purchase button until the plugin and its packaged-build benchmark both exist.

The UI would label each statistic exact, estimated, incomplete, or unknown. Row counts can be exact after a full scan. Distinct counts may be estimated using a bounded sketch. A cancelled scan never reports the partial count as the file total. A malformed record includes its byte or row location when known, but the plugin does not rewrite the source file.

Marketplace evidence observed on 9 August 2026

The public JetBrains Marketplace API supplied the five rows below. The counts prove that developers install data-file tools inside JetBrains IDEs and that both free and freemium distribution exist. They do not reveal paid users, renewals, revenue, buyer identity, or conversion for this proposed plugin. That boundary matters because a large free download counter is evidence for the job, not evidence that a buyer will pay for another implementation.

Comparable	Downloads	Model	Fetched response hash
CSV Editor	13,407,363	FREE	`d9c4a55598f3...`
Big Data File Viewer	2,302,704	FREE	`852b42b08f39...`
Parquet Viewer - Query, Inspect & More	18,169	FREE	`100bac325874...`
Omni Viewer	38,676	FREEMIUM	`b33431ed8c83...`
Data Preprocessor	543	FREEMIUM	`519813e8a939...`

The strongest free incumbents are not incidental: CSV Editor was observed at 13,407,363 downloads and JetBrains' Big Data File Viewer at 2,302,704. Parquet Viewer was observed at 18,169 downloads. Omni Viewer, a freemium multi-format product that includes a free Parquet viewer, was observed at 38,676. Data Preprocessor, a freemium CSV, Excel, and JSON tool with profiling features, was observed at 543. A generic table viewer would enter a crowded, free market; the only credible wedge here is a full-read, local-only inspection contract whose exact-versus-estimated labels and packaged benchmark are visible before purchase.

JetBrains' official Marketplace documentation says the platform supports evaluation licenses, personal and commercial subscriptions, and monthly or yearly billing. It also says paid IntelliJ plugins use platform-side licensing communication. Those facts establish a real distribution and billing rail; they do not authorize account creation, agreement acceptance, or publication under a vendor identity. Download the dated market-probe receipt and source URLs.

Three steps from evidence to a paid plugin

Review the evidence. Inspect the benchmark receipt, its generator, and the Marketplace comparison. Decide whether full-read honesty solves a real gap that the free viewers do not.
Request a pilot scope. Name the JetBrains IDE, operating system, file formats, largest non-sensitive size, and the statistics that decide your first troubleshooting step. Do not upload a file, customer data, credentials, or logs.
Build before listing. A qualified request authorizes a private beta build, not a sales claim. The packaged plugin must pass CSV, newline-delimited JSON, and Parquet correctness fixtures plus the one-gibibyte full-read test before any Marketplace listing can advertise those formats.

Demand-test state: no JetBrains Marketplace listing, packaged plugin, checkout, paid license, or customer fixture exists. The public page is a scoped offer backed by a reproducible reference receipt. At the seven-day checkpoint, page reach and product-specific pilot requests must be counted separately; our own fetches and crawlers are never demand.

Request the founding-team pilot