Collabysis

Collabysis collects public data and analyzes it on your computer, for researchers who must show their work.

Download for Windows Or analyze a CSV in your browser
Gross job gains rate by geography group, all 270 collected rows46810121416Gross job gains, percent of employmentU.S. totaln = 5States and DCn = 255Territoriesn = 10Each dot is one collected row. Vertical marks show group medians: U.S. total, 5 rows, median 5.7 percent; States and DC, 255 rows, median 5.8 percent; Territories, 10 rows, median 5.85 percent.
Figure 1. All 270 rows Collabysis collected from two public BLS tables, by area group. The group medians lie within 0.15 points of each other, and the checked test finds no clear difference.

One dataset, from collection to report

  1. 1. Collect

    Collected from two public BLS pages, with robots.txt checked and each row tied to its source record.

    Table 1. Five of the 270 collected rows, December 2025
    AreaJob gainsJob lossesSource record
    California1,048,011935,872weblink:69c281e83977c73f
    Texas686,824590,952weblink:b8ace2a9453c2c78
    North Carolina251,507212,979weblink:f0aabcfd2d4d614a
    Tennessee176,311146,925weblink:cbd0799fe73f945b
    Pennsylvania276,274251,356weblink:c73d7a2deee128f9

    270 joined rows across five three-month periods and 54 areas, collected 2026-09-24 from BLS Table 6 and BLS Table 7. The tables hold aggregate data only. Download the CSV, codebook, manifest, or summary (HTML, Markdown).

  2. 2. Prepare

    Joined on period and area, with every net change checked against gains minus losses: 270 of 270 verified.

    Table 2. The same rows after preparation
    AreaNet changeGains rate, %Losses rate, %Check
    California112,1396.86.1verified
    Texas95,8725.64.9verified
    North Carolina38,5286.05.1verified
    Tennessee29,3866.25.2verified
    Pennsylvania24,9185.14.7verified
  3. 3. Analyze

    Tested with its assumptions checked. When a check fails, the verdict says so and reports a suitable test.

    The values within the groups are not normally distributed and the groups have unequal spread, so the standard ANOVA is not used as the finding. The rank-based Kruskal-Wallis test finds no clear difference between the groups.

    Standard ANOVA
    F(2, 267) = 3.61, p = .028. Set aside: the values are not normally distributed (p < .001) and the spread is unequal (p < .001).
    Kruskal-Wallis
    H(2) = 0.13, p = .94. Reported. Figure 1 shows the rows behind it.
    Also flagged
    54 areas repeat across five periods, so the rows are not fully independent.
  4. 4. Report

    Exported as a report you can send, with the same verdict and both tests.

    Excerpt of the exported report: the one-way ANOVA heading, the verdict sentence, three failed assumption checks, and Table 1 with the Kruskal-Wallis and ANOVA results.

    An unedited excerpt of the report the Windows app exported for this test, including its calculation-status labels.

Collabysis and this derivative are independent and are not affiliated with, sponsored by, or endorsed by the U.S. Bureau of Labor Statistics. See BLS copyright information and BLS terms.

How Collabysis compares

CollabysisSocial-listening suitesGeneral AI assistants
Where the data comes fromCollects public pages and open data itself, or opens your file. Each collected record keeps its source link, retrieval time, and method.Brandwatch states it covers conversations from 100 million unique sites[1]. Sprout Social states it pulls social data from a variety of sources[5].Typically text or files you paste or upload; some search the web.
Who computes each numberThe statistics engine on your computer, from your rows. Code that Collabysis AI writes is shown before it runs and labeled as AI-authored.Brandwatch states it uses AI to spot insights[1]. Sprout Social states its AI Assist surfaces conversation patterns and sentiment shifts[5].Typically a language model, which can state a number it did not compute unless it runs code.
Assumption checksRun with the common tests. When one fails, the verdict says so and reports a suitable test (step 3).Typically not described on the public product pages.Typically only when you ask.
A record another person can rerunA reproducibility pack with the data hash, method, parameters, checks, and runnable code. The run script needs a Collabysis Python runtime, and the Windows app is not that runtime.Brandwatch lists Excel, PPT, PDF, and API sharing[1] and an API that exports analysis results[4]. Typically no runnable code with the method.Typically a chat transcript.
Where your data staysOn your computer. Collabysis AI is off until you turn it on, and then it sends only the fields the privacy page lists.Brandwatch states that for one product (abbreviated BCR on the page) raw data ingestion, indexing, analysis, and storage take place in AWS in the US, and that analysis results and customer metadata are stored in colocation facilities in the UK[3]. Sprout Social states that its services are hosted through third-party data centers[7].Typically on the provider's servers.
CostThe browser workbench is free. The Windows app runs with a trial key (30 days when issued by Collabysis, 7 days with a subscription trial at checkout), then a paid key.Brandwatch lists plans from standard to enterprise and offers a free demo[2]. Sprout Social lists per-seat pricing, annual billing, and listening as an add-on[6].Typically a free tier and paid plans.

Social-listening suites: for example Brandwatch and Sprout Social. General AI assistants name no product, and each of their cells says “typically”. Bracketed numbers link to the public pages listed below.

Sources

  1. Consumer intelligence product page, Brandwatch. https://www.brandwatch.com/suite/consumer-intelligence/. Read 10 October 2026.
  2. Pricing page, Brandwatch. https://www.brandwatch.com/pricing/. Read 10 October 2026.
  3. Information security page, Brandwatch. https://www.brandwatch.com/legal/information-security/. Read 10 October 2026.
  4. APIs page, Brandwatch. https://www.brandwatch.com/products/apis/. Read 10 October 2026.
  5. Social listening article, Sprout Social. https://sproutsocial.com/insights/social-media-listening/. Read 10 October 2026.
  6. Pricing page, Sprout Social. https://sproutsocial.com/pricing/. Read 10 October 2026.
  7. Security page, Sprout Social. https://sproutsocial.com/security/. Read 10 October 2026.

Collabysis column: Figure 1, Tables 1 and 2, step 3, and the report excerpt on this page; the dataset manifest; the privacy and data flow page; the license. Product pages change; check a vendor's current documentation before relying on a cell.

Download for Windows

Windows 11, 64-bit. Version 3.24.0, 152.0 MiB. The early-access installer is unsigned, so Windows may show an Unknown publisher warning; verify its published SHA-256 checksum before you run it.

Download for Windows Portable ZIP Redeem activation code Checksums and install notes
License and release terms

Version 3.24.0 is licensed under the Collabysis Software License Agreement, version 1.3. LICENSE.txt and THIRD_PARTY_NOTICES.txt are inside the ZIP. Analyses and collections run with a trial key (30 days when issued by Collabysis, 7 days when started with a subscription trial at checkout) or a paid key valid for the term shown at purchase. Opening existing files, exporting existing results, viewing settings, and checking for updates work without a key. This build stops running analyses and collections on 2027-03-04; this page always offers the current build. Release packaging rejects .env files, private-key files, and detected credentials. Data collected from third-party sites keeps its original rights and terms.

Who owns the results?

Version 3.24.0 is licensed under the Collabysis Software License Agreement, version 1.3; LICENSE.txt is inside the ZIP. The hosted browser workbench is free and needs no key. Analysis and collection in the downloaded Windows app need a trial or paid key. Results, charts, generated code, and reproducibility packages created with the software belong to the person who created them and may be published without restriction or attribution under this Agreement. Third-party source material included in an output remains subject to its own rights and terms. Third-party components keep their own licenses. See the third-party notices.

What permitted public content can Collabysis read?

The Windows app reads permitted public content through source APIs and direct pages. Direct-page extraction accepts readable text, HTML tables, repeated page records, JSON-LD, and microdata; optional bounded discovery can propose page links, feeds, and sitemaps. Full-depth web search requests up to 20 results per query, and page extraction defaults to 20 URLs and can be raised to 50. Requests are paced per site, check robots.txt by default, honor Retry-After, and use bounded retries. A CAPTCHA, verification page, login wall, or persistent access block stops that site. Private or login-gated content is outside the collection path, and Collabysis does not evade access controls. Provider APIs remain subject to provider terms, your API key, quota, and requested bounds. Instagram and TikTok are collected through a provider under your own key, after you accept its notice. Facebook and WhatsApp are not collectable.

Back to the home page

Browser workbench

Analyze a CSV on this device.

Move through Data, Question, Methodology, Results, and Export. Your session stays in this browser.

1 of 5
Step 1

Data

Nothing is uploaded. If you choose to remember it, the file stays in this browser's storage until you clear it. Remembered files expire after 30 days. CSV rows stay on this device while the workbench is open.

Data source

Need to collect data first? The Windows app can collect permitted public sources and export a receipt-backed CSV.

Choose a CSV to build its local profile. No dataset is loaded automatically.

Column roles, missing values, and examples
Column Detected role Missing Example values Role override

Role overrides guide engine recommendations, diagnostics, and eligible column controls.

Start over?

This clears the current browser workbench session, including any remembered CSV and result. Choose Cancel to keep the current data and settings.