NEWS

---

kitJoin 0.7.4

• Uses kitDesign 0.6.1: plots and tables inside closed sections are drawn when the section is opened (no more squashed plots). Nothing else changes.

kitJoin 0.7.3

• Uses kitDesign 0.6.0 (the same version as kitsUI and kitverse). Nothing changes on screen.

kitJoin 0.7.2

Fixes

• Firefox with some add-ons (for example Google PDF Viewer and other "document viewer" add-ons): clicking a download button left the page and the download was "Not Found". The add-on takes over download links on the page. Download buttons now come from kitDesign 0.5.2, where they are buttons, not links, so there is nothing for an add-on to take over. The buttons look the same.

kitJoin 0.7.1

Fixes

• ⑧ 格式轉換: CSV.GZ → CSV now starts the CSV with a BOM when the text is UTF-8, so Windows Excel shows Chinese correctly (kitJoin's own .csv.gz files have no BOM, so converting them back never had one).
• File names with a time stamp (for example long_20261031_194357) use Taiwan time; on the cloud they were in UTC.
• Re-running ② after ③ or ④ had a result no longer prints Error in if: missing value in the R console (the preview line read an emptied result).
• The cloud version and run.bat sourced R/ twice at start-up and warned that the folder looks like an R package. R/_disable_autoload.R now turns shiny's autoload off.

kitJoin 0.7.0

Changes you can see

• The app is now called 「KIT 資料處理工具」 (KIT data processing tool) instead of 「KIT 資料串接工具」 (KIT data join tool): it does more than joining (stacking, attrition, missing values, validation, kits response files). The package name stays kitJoin.
• The version line in the header starts with the package name: "kitJoin v0.7.0 · 2026-11-01" (kitDesign 0.5.0).

Under the hood

• Uses kitDesign 0.5.0 (Imports: kitDesign (>= 0.5.0), Remotes: jiewenTsai/[email protected]); manifest.json regenerated.

kitJoin 0.6.0

Changes you can see

• ⑥ kits 作答檔: step 2 only picks the item columns. Domain and scoring are set in step 3's item table, which is shown open and can be downloaded at any time (kits_item_settings.csv). The default domain is the item group's name without a trailing _, . or -.
• ⑦ The reshape diagram shows wide → long first, with a double arrow: both directions work.
• 「用 R 怎麼做」: code boxes stay inside the page, and every code line fits in 70 columns.
• Danger buttons are salmon, as the KIT design language says.

Under the hood

• kitJoin now uses kitDesign (Imports: kitDesign (>= 0.4.0), installed from GitHub jiewenTsai/kitDesign) for the page, the shared components and the styles. Its own copies (kit_brite.R, code_wrap.R, kit_theme.R, the logo) are gone, about 2,000 lines. Every output file is byte-identical to 0.5.0's.
• The download button (R/download.R) and the cloud-version check and footer moved into kitDesign 0.4.0 (kit_download_button(), kit_is_cloud(), kit_cloud_footer()); R/cloud.R keeps only kitJoin's text. Same file names, headers and bytes as before.
• The six run tabs (join, stack, attrition, missing values, reshape, file conversion) use kitDesign's kit_run_tab_ui() / kit_run_tab_server() for the page skeleton and the status flow (ready line, run with progress, status box, notification, scroll, stale after a change, result details, 「用 R 怎麼做」); each tab keeps only its own settings, run and result. The ⑥ kits tab uses the server part for its run and status. Same page HTML, ids, statuses and output files as before.
• run.bat installs or upgrades the packages that are missing or older than DESCRIPTION asks for (an old bslib used to break the app on the first screen), including the GitHub ones (kitDesign).

kitJoin 0.5.0

Generality fixes found by running every tab on another panel survey (TDCYP, five waves of SPSS files whose ID column, item names and missing-value codes differ from KIT's).

Wrong results that were reported as success

• ⑥ no longer writes suspected missing-value codes as scores without saying so. Values labelled 遺漏值 / 跳答 / 不適用 / 拒答 (and similar), ②'s code list, values of 7777 and above, and values clearly apart from the item's range (e.g. 95–99 next to 1–5) are listed per item in 「3. 題目設定」 and in the result, which then shows 「完成，但請注意」 with a link to ②: turn them into NA in 「② 遺漏值轉 NA」 first, then come back to ⑥. ⑥ itself never turns values into NA (the data are assumed correct): if you go on, these values are written unchanged as scores, and shifting or reversing the scores leaves them alone. The gap warning no longer suggests 「合併」 for values that look like codes. The same warning covers codes in the sex, wave_m and age_m columns. For KIT data nothing changes: values of 7777 and above are still written unchanged, with the new warning.
• ⑦ wide → long no longer copies the first column's value labels onto every item. When value or variable labels differ between the stacked columns they are dropped with a visible note.
• ⑤ stops when most baseline IDs are duplicated (e.g. a birth year chosen as the ID) instead of reporting 「完成」; ③ asks for confirmation when a join would multiply the rows and refuses joins that are far too large. ID choices show each column's share of unique values.
• ④ a column named w (e.g. a survey weight) no longer overwrites the wave column.
• Numeric IDs of six or more digits are no longer turned into scientific notation (1e+05) when compared with text IDs or written to subj_id / visit_id. Text IDs are trimmed before matching.
• ② can convert per variable from SPSS value labels: only values whose label says missing (遺漏、跳答、跳題、不適用、拒答、未作答、missing、refused, skip…) become NA, so a real weight of 99 kg stays. 不知道 / 不清楚 / "don't know" are not in the default keywords (they can be a valid answer); add them in ② when you want them turned into NA. ⑥ uses the same rule when it looks for suspected codes. This mode is chosen automatically when the labels mark missing values and the KIT codes are absent or at least ten times rarer (TDCYP has a few dozen real values that happen to equal 9999); when labels mark more cells but not that many more, ② only suggests it. The code list is still available.
• ② results are cleared when the codes, excluded columns or text matching change, and ③–⑦ say that ② must be run again instead of silently using the raw data.
• SPSS files saved as Big5 without a proper encoding record are re-read as CP950 (① can also be set by hand).

Crashes and broken files

• SAV downloads are read back after writing; when haven cannot read them, problematic text value labels are cleaned and the file is rewritten, and the affected columns are named on screen (also in ⑧'s notes).
• Columns whose value labels were all missing codes no longer break the preview in ②–⑤.
• 「用 R 怎麼做？」 code uses data_1, data_2… and runs (and matches the app) with numeric or Chinese wave names, quotes, blank or mixed-type IDs and column names with spaces; this is now checked by tests that evaluate the generated code.

Data with other naming schemes

• ③ ⑤ each file can use its own ID column(s) (e.g. w1_id, w2_id, w3id); they are renamed to one key before matching. Files that share the ID name work as before.
• ⑥ 「2. 對應欄位」: subj_id, the wave label, wave_m, age_m and sex can be chosen per file.
• ⑥ common output item IDs across waves: an item_id_out column in the item table or settings file, or the rule 「欄名去掉 ^w[0-9]+」. Two columns of one file mapping to the same ID are refused; merged IDs are listed, with a warning when their variable labels differ.
• ⑥ reverse-item naming rule: trailing r (default, as before), leading r, or off.
• ⑥ wave_m is called 「時間點」 with a selectable unit (months by default). Negative or non-numeric values and visit_id collisions caused by _ are refused.
• ⑥ labelled cohort / sex columns can be written as codes or labels; missing codes in them are flagged. An item that is text in one file is skipped only in that file. The item table / settings parser accepts BOM, SAV label columns, Y/N and 1=0,2=1, and stops with an explanation when it cannot read a row; items with no answers need no domain.
• Other: empty or header-only files are errors; semicolon and tab separated CSV files are detected; uploading the same file twice is flagged; wave names that differ only in case are refused; ID detection also recognises w3id, 學號 and 編號.
• Code and monospace text use JetBrains Mono (Google Fonts, loaded by the browser) instead of M PLUS Code Latin.

kitJoin 0.4.3

• Downloads now always carry the file name and extension, including names with Chinese characters, in every browser. The app serves each file itself: the download URL ends with the file name, the response has an RFC 6266 Content-Disposition (filename plus filename*=UTF-8''..., no redirect), and the button has download="<name>". This covers every download (②③④⑤⑦ CSV / CSV.GZ / SAV, ⑥ response file and item settings, ⑧). Before, some browsers saved e.g. missing-download_sav without .sav.

kitJoin 0.4.2

• ⑥ 「2. 對應欄位」: the wave label, wave_m, sex and age_m rows can now come from a column, be typed per file (「自己填」), or be left out (optional rows only). Typed values get one input per file in section 1 (the wave label is prefilled with the wave name from ①); wave_m and age_m must be numbers. Typed values are written exactly like a column holding that value, and visit_id uses the typed wave label. This covers one file per wave with no wave columns. The allowed sources are defined in KITS_MAP_ROWS.
• ⑥ section 3 is now 「題目設定」: one row per candidate item column with include/exclude, domain, observed raw scores, a raw → score map (shown as a two-row table: 原分數 / 計分), the resulting scores and alerts. Bulk actions on selected rows: set domain, shift, reverse (max − score; max defaults to the item's current maximum), merge (「原分數 4、5 → 計分 3」), restore, exclude/include; selection helpers 全選 / 全不選 / name filter. The regex rules box and the separate 「進階調整」 steps are gone.
• ⑥ item settings can be downloaded (item_id, include, domain, map_raw_to_score written as 1=0;2=1) and uploaded again for the next wave; a plain item_id + domain table still works (with the r fallback). Maps that were changed and miss a newly observed raw score block the run until fixed.
• ⑥ suggests (never applies by itself) "select all and shift −m" when every included item starts at the same m > 0. At run time, final scores must be integers ≥ 0.
• ⑥ 「用 R 怎麼做？」 reads the settings file and writes only changed maps, grouped: \(x) x - 1, \(x) 4 - x, or case_when().
• ⑥ 加上「開啟 kitsUI」按鈕 (in the tab guide and next to the downloads; opens kitsUI in a new tab).
• ⑥ top callout: several files are merged into one response file; upload only the file you need for a single wave or domain.
• ⑥ 「3. 題目設定」 stays fast with 150–300 items: the table is built as one HTML string (same markup as before), and the name filter hides rows in the browser instead of redrawing the table. The filter is now case-insensitive.
• ⑦ wide → long: 「要轉成長格式的欄（cols）」 is click-only. The matches() regex box is gone. Pick the columns to convert, or pick the columns to keep (cols = !c(id, sex)) and convert the rest. 全選 / 清除 and name-group buttons (「lan 開頭（10 欄）」) add whole groups; a live line says how many columns convert and how many stay. The R code writes !c(...), starts_with() (only when it selects exactly the picked columns) or the names.
• ⑦ looks at the file first and says whether it looks long (an ID repeats) or wide (one row per ID with groups of item columns). The ID column can be changed. When the chosen direction contradicts the guess, an orange note appears by the run button (it does not block or switch direction).
• ⑧ downloads keep the file name for non-ASCII names: 第一波.csv.gz → 第一波.sav. Before, the browser saved convert-download_file without the .sav extension (Shiny sends no file name for non-ASCII names; the button now carries it in its download attribute).
• Code and monospace text use M PLUS Code Latin (Google Fonts, loaded by the browser).
• README: the ⑥ section is complete again (part of it had moved into the ⑦ and ⑧ sections), with FAQ entries for "every item is one domain" and "no wave column".

kitJoin 0.4.1

• ②（選用）遺漏值轉 NA: its texts now say that ③–⑦ (not only ③④⑤) use the processed data, and the result has its own 「下載處理後的檔案」 section (CSV, CSV.GZ, SAV), so a user with one file can fix missing codes and download without going further.
• ① 上傳檔案: with one file, the next-step box lists what works with one file (②, ⑥, ⑦) and says ③④⑤ need at least two; the box for two or more files also links to ⑦.
• ⑥ no longer has its own missing-code step; use ②. ⑥'s adjustments are now 1 shift scores by -1 and 2 reverse scoring. When item, wave_m or age_m columns still hold values of 7777 or more and ② was not applied, ⑥ shows a warning with a link to ②, and the "shift by 1" suggestion ignores those values. The 「用 R 怎麼做？」 code notes when ② was applied.

kitJoin 0.4.0

New features

• New tab "⑥ 轉成 kits 作答檔" turns wide item data (one row per child and wave, one column per item) into the long response file that kits and kitsUI read. The output always has the nine kits columns in fixed order and type (visit_id, subj_id, cohort, sex, wave_m, age_m, item_id, item_score, domain); unmapped optional columns are written empty.
  • Cohort is typed per file or taken from a column (never guessed from the file name).
  • visit_id is built as {cohort}_{subj_id}_{wave label}, falling back to wave_m (with an on-screen note) when no wave-label column is mapped.
  • Domains come from an uploaded item table (item_id, domain) and/or typed regex rules; r-suffixed items missing from the table use their base item. Columns without a domain must be excluded or assigned one.
  • Opt-in adjustments run in a fixed order: missing codes to NA, shift scores by -1, reverse scoring (max - x on the 0-based scale). Shifting to 1-based data is suggested, never ticked automatically.
  • Checks block duplicates (visit x item), empty required columns, visit ids shared by two cohorts, non-integer or negative scores; warnings flag items without a 0, score gaps, likely missing codes, and non-anchor domain names.
  • Memory-light: all checks and adjustments run on the wide data, and the long file is streamed to .csv.gz in chunks at download, so ~10M long rows peak around 0.6 GB.
  • Downloads one merged .csv.gz (or .csv); the "用 R 怎麼做？" panel shows tidyverse code with only the ticked steps.
• New tab "⑦ 長寬互轉" reshapes one uploaded file with tidyr::pivot_wider() or pivot_longer(). Controls map 1:1 to tidyr arguments with tidyr defaults.
  • Wider: duplicated id x names_from blocks and lists the rows (no values_fn); id columns that vary within an id are flagged; new names may not collide with id columns; SPSS renames are announced.
  • A "套用 kitsUI 分數表預設" button fills (never runs) a preset for kitsUI score tables, and an empty-cell table separates "no row" from "row present but empty".
  • The "用 R 怎麼做？" panel prints the pivot_*() call with non-default arguments only.
• Shared pivot_longer() control module (mod_pivot_longer_ui(), mod_pivot_longer_server(), pivot_longer_code(), pivot_longer_run()), used by ⑥ and ⑦: each control maps to one tidyr argument, defaults follow tidyr, and the generated code lists only non-default arguments.
• New tab "⑧ 檔案格式轉換" (file format conversion): a standalone converter that takes one CSV, CSV.GZ or SPSS .sav file and converts it to one of the other two formats. CSV <-> CSV.GZ is streamed in chunks without parsing (UTF-8 bytes are kept as is; BIG5 is re-encoded to UTF-8 line by line), so large files are not held in memory. Conversions involving .sav use haven; IDs with leading zeros stay text, invalid SPSS names are renamed, and SPSS value labels are written as codes unless "數值標籤轉成文字" is ticked.
• Every result download area gains a "下載 CSV.GZ" button (readr::write_csv() to a gzip-compressed file).
• "① 上傳檔案" now tells users who pick a .csv.gz file to convert it in ⑧ first.
• tidyr and vctrs are now imported.

Bug fixes

• ① 上傳檔案: when several files were uploaded at once, only the last file's wave name, move and remove buttons worked; the others looked edited but the server kept t1, t2.