One-time downloads · $0.99 USD. No subscription.

Perunlight

Step-by-step guide

CohortCrosswalk

Review and document exact category mappings between successive CSV/TSV releases while preserving the original cells.

Use this for changing category codes in nonclinical open aggregate data. You supply the category meaning; the app records release-specific decisions, coverage, source identity and missing reasons for a reproducible handoff.

Get CohortCrosswalk · $0.99

See it in use

Align codes in two annual parks tables

A screenshot walkthrough with fictional practice data. Select a step to explore.

1. Start with

Align codes in two annual parks tables

Two fictional three-row parks tables use different free, paid and unknown codes.

Sample files and contents

Fictional parks-2023.csv and parks-2024.csv, three rows each. 2023 uses 01, 02 and 99; 2024 uses FREE, PAID and UNK.

Practice files and guides are free. The complete app is included only where the product page says so.

2. In the app

CohortCrosswalk 1.0.1 shows original 2024 park codes beside reviewed categories, statuses and missing reasons.
Actual Chrome on Linux. The preview shows the first five rows of the selected release; this example has three. Enlarge image ↗

Try it in three steps

  1. Load the two CSVs and inspect unresolved 2024 UNK.
  2. Save an explicit missing decision with reason “not recorded”.
  3. Download the ZIP and compare its output CSVs.
Show every click and setting
  1. Try the parks example and inspect the one unresolved 2024 UNK code.
  2. Choose Explicit missing, reason not recorded, then Save decision.
  3. Inspect the 2024 handoff preview: original UNK, appended missing status and not recorded reason.
  4. Download the complete ZIP and compare output CSVs with the expected files.

3. The result

Inspect the result

All six original rows stay in order. Mapped statuses are appended while 99 and UNK remain unchanged in their source columns.

The actual CohortCrosswalk report exported from the supplied fictional parks example.
The complete ZIP includes original columns, appended results, a decision ledger, coverage and a hash manifest. Enlarge image ↗
What to check in the practice result

All six original rows remain in order. 01 and FREE map to free; 02 and PAID map to paid. 99 and UNK remain unchanged in their source columns with an appended missing status and reason. Coverage becomes 6/6, with four mapped and two missing rows.

This fictional example is shown in the working interface and in the supplied result. Review the guide before using your own files.

See the workflow at a glance

Choose a short explanation, then follow the worked example below. Check the requirements and compatibility before using your own files.

Keep the original code beside your reviewed meaning.

Last year's 01 is this year's FREE, but an unknown code still needs a decision. CohortCrosswalk's fictional parks tables preserve all six original rows while appending reviewed categories and missing reasons. Resolve UNK as explicitly missing, then inspect the exported data and decision record. A coverage count of 6/6 means decisions were recorded; it does not prove semantic equivalence. Intended for nonclinical open aggregate CSV/TSV data, the tool does not perform joins, statistics or automatic matching. The guide walks through the review in Linux Chrome.

View the illustrated card
Keep the original code beside your reviewed meaning. — the same explanation written above, with a labelled illustrative panel.
Download the full-size PNG card

Before you start

  • Google Chrome on Linux. Verified in Chrome 152 on Ubuntu 22.04; other browsers and operating systems are unverified.
  • Extract the entire buyer ZIP and open app/index.html. Keep its neighbouring app files together; no server or installation is needed.
  • UTF-8 CSV/TSV files with unique nonempty headers, comma or tab separators and consistent row lengths. Combined inputs must fit the documented byte, row, cell and rule limits.
  • Definitions and source notes sufficient for you to decide how each code maps. The app does not infer equivalence or validate scientific meaning.

Your first result

  1. Open the app and try the parks example

    Extract the entire ZIP, open app/index.html, and choose Try the parks example. Expect 5/6 codes covered and one row needing review.

  2. Resolve the deliberate unknown

    Filter Needs review. For release 2024, concept access, exact code UNK, choose Explicit missing, enter not recorded as the reason, then Save decision. Expect 6/6 codes and zero rows needing review. A missing reason is shown beside the decision and focused for correction. After saving the final decision, use Inspect handoff preview to continue.

  3. Download and compare the result

    Before downloading, inspect Review & handoff → Inspect the appended result. Choose release 2024: original UNK remains visible, its appended status is missing, and its reason is not recorded. This read-only preview shows the first five rows; complete export still includes every row. Then choose Download complete ZIP, extract it, and open report.html. Compare the two harmonized CSVs with the supplied expected files: six rows, four mapped values and two explicit missing values.

  4. Connect your own releases

    Add each UTF-8 table with a unique release ID, comma or tab delimiter and a source note. Inspect the preview before accepting it. Add a shared concept and assign the intended source column in every active release.

  5. Save each exact-code decision

    Choose a target category or an explicit missing reason for every observed code. Save each edited decision. If several codes map to one category within a release, record a rationale and acknowledge information loss for every contributor. Templates supply suggestions; they do not approve the destination automatically.

  6. Keep a project and the originals

    Save the project JSON before closing: there is no autosave. To restore it, reselect every original file with the same filename and bytes. A changed file needs a new release and fresh review. Complete export requires all decisions; a draft download contains only a clearly marked HTML report. To remove the app, delete the extracted folder while retaining source files, projects and outputs you need.

A worked example

Two annual parks tables with different codes

Input: Fictional parks-2023.csv and parks-2024.csv, three rows each. 2023 uses 01, 02 and 99; 2024 uses FREE, PAID and UNK.

  1. Try the parks example and inspect the one unresolved 2024 UNK code.
  2. Choose Explicit missing, reason not recorded, then Save decision.
  3. Inspect the 2024 handoff preview: original UNK, appended missing status and not recorded reason.
  4. Download the complete ZIP and compare output CSVs with the expected files.

Expected result: All six original rows remain in order. 01 and FREE map to free; 02 and PAID map to paid. 99 and UNK remain unchanged in their source columns with an appended missing status and reason. Coverage becomes 6/6, with four mapped and two missing rows.

Troubleshooting

Complete export disabled

Save the edited card, finish column assignments, then filter Needs review and supply missing reasons and all merge acknowledgements.

Filename or hash mismatch

Reselect the unchanged named original. Treat changed data as a new release with fresh review.

One-column or malformed preview

Confirm comma/tab delimiter and UTF-8 CSV/TSV quoting; repair a separate copy if needed.

Size limit exceeded

Use a smaller aggregate subset or fewer concepts; no silent truncation occurs.

Compatibility & limits

  • Only Google Chrome 152 on Linux was tested. Other browsers and operating systems are unverified.
  • UTF-8 CSV or TSV, with comma or tab separators, only. No XLSX, DTA, SAV, record joins, statistics, fuzzy matching or automatic category equivalence.
  • Up to 10 releases and 10 concepts, 5 MiB cumulative input, 20,000 data rows, 200,000 source cells including headers, and 100 columns per file. All limits apply together.
  • Up to 16 KiB per cell, 5,000 observed decisions, 4 MiB saved project and 25 MiB generated file payloads. Larger inputs are refused rather than truncated.
  • No autosave. Keep the saved project and exact original filenames and bytes. Changed inputs require fresh review; project files do not contain the complete raw tables.
  • You decide what categories mean. Coverage counts recorded decisions and does not certify semantic equivalence, scientific correctness or fitness for clinical use. Intended for nonclinical open aggregate data.
  • CSV output preserves original strings, including formula-like text. Import every CSV column as text when using a spreadsheet; output is not spreadsheet sanitization.
  • Merging different source codes requires a reason and explicit information-loss acknowledgement. Reused templates provide suggestions and reset destination approvals.
  • The read-only handoff preview shows the first five data rows per release. It is not a full-table editor; complete export contains all supported rows.

Need a hand?

Tell us the product, host application and version, what you tried, and the exact error. Contact support with a fictional example; keep private customer files and passwords out of your message.