Skip to content

What's New

A plain-language, timelined summary of new and meaningfully improved capabilities across wafermap and tsmap — for people using the app, or building on the library, not for contributors. Only genuinely user/adopter-facing capabilities are listed here: internal refactors, dependency bumps, docs-only fixes, and process/tooling changes are skipped, even when they're a real fix worth having. For the complete technical record of everything that shipped, see each project's own changelog: wafermap · tsmap.

Covers everything since both projects moved to the wafertools GitHub org and @wafertools npm scope (2026-08-01) — see each changelog directly for anything earlier.


2026-11-06 — What divides yield at a glance, groups in What stands out, and wafers that open on what you clicked

What divides yield at a glance. Each wafer field is one row: the worst and best group as two bars on a 0 to 100% scale, a sentence giving the gap in points, and a confidence of Clear, Possible or Not clear in place of the p-values. Fields that group the wafers the same way (a tester and the handler docked to it) are one row, and a field hidden by another (a weak probe card on wafers that also differ by recipe) is read after allowing for it.

Groups in What stands out. What stands out compares groups of wafers on yield and on the weakest ring as well as on tests, so a probe card or a recipe that costs yield is named with its size in points and the fail bin it is in. Groups that differ clearly but on too few wafers to confirm are listed under Possible, not confirmed.

A wafer opens on what you clicked. From any chart or table, a wafer opens on the test it is about, on its bins for a yield, and otherwise in the mode the maps are in. A row of the Dies table opens its wafer with the die ringed, on the test whose column you clicked. Right-click opens the same menu a click does on every Insights chart.

Worth a look follows the filter. Under a filter, Worth a look shows the items What stands out lists for what the filter keeps, and each Chart link in the Summary panel opens its card.

In tsmap. The desktop app tells you when a newer release is out, with Not now and Skip this version; nothing is downloaded. The sample lot is a full lot of 25 wafers tested in five sessions, so every Group by field moves the data.

For wafermap adopters. All of the above comes with the Insights tab and the Summary panel, with nothing to wire.

See: tsmap v0.2.3, tsmap v0.2.2, wafermap v0.40.1, wafermap v0.40.0

2026-11-01 — More Insights charts, signed colour scales and a My plots | Examples switch

More Insights charts. In tsmap, Tests by wafer (or by group) at the top of Distributions shows the tests whose wafers differ most, as a grid of z-scores. Yield by site appears when a lot was tested on more than one site, and Re-probed dies, by ring and wafer, when a lot has re-probed dies. The box plot can be drawn by radius, the value histogram as a cumulative curve, and the test scatter, histogram and wafer-to-wafer trend have log scales.

Correlation. On the correlation matrix, Fails together shows which tests fail on the same dies, and Within groups takes each group's mean off first, so a difference between groups is not read as a link between tests.

Comparing groups. With Group by on, the comparison tables shade each group's mean among its peers, and the Data tab's Statistics view opens with one table of every test for every group, with Ppk and the share inside limits. A control chart can colour its points by a wafer field, so a pattern that belongs to a group shows as a colour.

One colour language. Every signed scale is blue below and orange above: the correlation matrix, z-score heatmaps and the difference map, so one colour never means high on one chart and low on the next. Heatmaps have a line between cells, a frame and a tick for each row and column.

Overview and data. The Overview leads with Worth a look, then what divides yield, then the standing cards, and the Wafers table puts the numbers first. A plot card's ⋯ menu can save the numbers behind it as CSV or copy them to paste into a spreadsheet.

My plots | Examples. Examples shows one plot of each chart type the lot can show. Add to my plots keeps one; Add all keeps the rest, with an Undo. The Filter window and the plot editor each have a simple JSON editor: a filter or a plot can be written, pasted and copied as text, with suggestions for the keys, the lot's fields and tests, and the values a field takes.

For wafermap adopters. The Insights tab carries all of the above, and a plots file written with mark and encoding is read as chart and fields. A chart or field name the library does not know is reported with the valid choices and the nearest match.

Support. The support page offers GitHub Sponsors as well as Ko-fi.

See: tsmap v0.2.1, wafermap v0.39.0

2026-10-28 — Bin tables, setup files and settings export, and richer Insights cards

Edit bins, limits and derived tests in tables. In tsmap, Setup ▾ → Bin definitions… lists every bin with its die count, name, pass flag and colour, and Load… reads a definitions file in as edits you can check before applying. Edit limits… and Derived tests… in the test selector do the same for limits, units, type and computed tests, with each expression checked as you type. --bins applies a definitions file from the command line.

One setup file, and exportable settings. Setup ▾ → Setup file… saves the wafer diameter, edge exclusion, reticle, retest policy and value filter as one CSV, and --setup applies it for a launch. Preferences → Export or import settings… moves saved layouts, plots, filters and preferences between machines.

More on the Insights cards. Ring and quadrant yield by wafer heatmaps, a Group, then wafer view, a bin mix pareto, a pass-rate grid by group and a control chart on the wafer-to-wafer trend. The number keys 1 to 5 set the columns of a gallery or of the Insights charts.

For wafermap adopters. checkDerivedExpression checks a derived test's expression without building a map, so a host's own editor agrees with the build. The Summary panel docks on the right or the left only (placement no longer takes 'top' or 'bottom').

See: tsmap v0.2.0, wafermap v0.38.0

2026-10-22 — Learn tsmap, a choice of retest result, and part average testing on the map

Learn tsmap. Thirteen short lessons, from what a wafer map shows to opening data from a link, each ending with a lab in the web app and a slide to check your results against. They are slide decks that work at your own pace, with screenshots and one idea per slide. Lessons 00 to 06 are the core path; the rest are there for when you need them. A glossary explains the terms, and the sample lots the labs use are on the sample files page. (Learn tsmap, glossary)

Choose which result counts for a retested die. Setup ▾ → Retests… keeps the last, first, best or worst result when a die position is probed more than once, and --retests sets it from the command line. (Retests)

Part average testing on the map. Plot mode now offers PAT verdict, colouring each die within the PAT limits, an outlier or failing, on a single map and on every wafer of a gallery. (Part average testing)

More of the file's facts to group by. The MIR code fields and the equipment an SDR names (handler, probe card, load board and so on) are read from STDF and ATDF files, and are Group by choices when they vary between wafers.

2026-10-20 — A far bigger chart builder, part average testing, group comparisons, and one filter for every view

Build your own charts, and many more of them. The My plots tab now draws control charts (with the points that break a Western Electric rule marked), capability goal plots, heatmaps, probability plots, panels (one small chart per value of a field), nested categories, paretos and stacked or 100% bars, several tests in one chart rescaled to share an axis, a measured value cut into ranges, fit lines with their R², histogram shares and what-if limits, and reference lines and bands on any axis. Add examples picks the tests and split that have the most to show in your lot, and right-clicking a wafer's mark charts that wafer on its own. (The Insights tab, Chart gallery)

Part average testing (PAT). A PAT card on the Overview shows, per wafer, the dies within PAT limits set from that wafer's own spread of each test, the outliers that pass every limit but sit far from their neighbours, and the tests that drive them. The Dies table has a PAT column, plots can colour, divide or panel by PAT verdict, and Map the verdicts draws a map of each wafer by verdict. (Part average testing)

Comparing groups, and what divides yield. With Group by on, a Comparison by table compares yield and every test across the groups with each wafer as one value, so a few wafers are never made to look like thousands of dies, with a ranked q value, the size of the shift and a Difference map of where on the wafer the groups differ. What divides yield does the same for every wafer field at once, so you start with the field that separates the wafers most clearly. (Comparing groups of wafers)

What stands out, drawn as charts. The items What stands out lists beside the map now appear on the Insights Overview as Worth a look, each drawn as the chart that shows it, ranked together: a region's loss, a failing bin, a test that shifts in a region, a trend across the wafers, a spread too wide for its limits, and a wafer field whose groups differ (a corner lot's split). Show on map takes you back to the map showing that item. Findings also find a cluster of a single failing bin where other bins' fails sit beside it. (The Insights tab)

The Insights charts are plots you can edit. Yield, bin pareto, pass rate, box plot, histogram, trend and scatter are drawn by the same machinery as your own plots, so Edit as plot in a chart's ⋯ menu opens any of them as a plot of your own to change and keep. Every chart has an About this chart line, and the charts name the section of the user guide that explains them, which opens beside the app when you click it. (Plots)

One filter for the maps, the gallery and Insights. Filter…, the last button in the toolbar row, keeps only the wafers and dies that meet conditions you set: one split, bin 1, a ring, a range of a test, wafers below a yield. A wafer condition removes the wafer's card; a die condition draws the dies it drops faded on every map. The Summary panel, the findings, the report and the tables are analysed again over what is kept, and a line above the view always states the filter. Filters can be saved by name and moved as a small JSON file you can write by hand or generate; in tsmap they are remembered for the next lot. (Filtering wafers and dies, demo)

Explained, and a reference. The user guide has a new chapter, The statistics, explained, saying for each figure what question it answers and how it is worked out, and a Chart reference describes every chart, map and table: what it shows, why you would use it, how to read it and what to watch for. (The statistics, explained, Chart reference)


2026-10-13 — Test-value findings offered where they are missing, findings that say what they measured, and bzip2 and European CSV files

Test-value findings, offered where they are missing. Findings about a test reading high or low in one region take the longest to compute, so wafermap now decides when to run them: automatically when the data is small enough to take about a second, and otherwise the Findings panel says they are not included, with the number of tests and dies, roughly how long they would take, and an Analyse button. For a lot, one click analyses every wafer. Findings about tests failing their limits are always included. In tsmap the Show test-value findings menu item is gone: the offer is in the panel you are already reading. In wafermap the analysis now runs by default within that budget (enableTestValueAnalysis: 'auto'), and onStatsSummaryChange reports a re-run. (Findings panel, tsmap guide)

Findings that say what they measured. A test finding names its statistic: "CONT_V mean" or "CONT_V limit fail rate", with every rate in percentage points, so a rate difference is never read as a shift in the test's values. The report's links stay in the report, with a Back button to return to where you were, and its bar figures read on any bin colour.

Your map preferences are remembered. tsmap now keeps every map preference between sessions (overlays, legend position, number format, the Insights chart axis settings) as well as the colour schemes. A rotation, the compact layout and log scale are not kept: they describe the data on screen, and the next lot opens without them. (What tsmap remembers)

bzip2 and European CSV files. .bz2 files open like .gz, and a ;-separated CSV with decimal commas (0,25) reads its values as numbers. Dense wafers drawn small no longer show light bands, test limits are shown in the digits the tester stored (1.6, not 1.600000023841858), and large lots open and analyse faster. (Supported file formats)


2026-10-09 — Exclude clamp values, a Summary button, and fewer false patterns

Exclude clamp values. A tester that runs out of range records its clamp value, not a reading, and one such value stretches the colour scale and drags the mean, σ and Cpk. A test can now carry validity limits, the range a real measurement lies in, and a value outside them counts as missing in the map, statistics and charts and is counted wherever a population is shown. Bins and the tester's verdict are unchanged. In tsmap, Setup ▾ → Exclude values outside limits… chooses which limit set a value must lie inside, and the new VALID-LOT-09.csv sample has clamp values to try it on. In wafermap it is the valueFilter option. (Example, tsmap guide)

A Summary button, and rows that show their dies. The Summary panel opens from a labelled Summary button above the map, in place of the toolbar icon, and closes from its own header. Every row in "What stands out" shows its dies on the map when clicked, and the Findings section's Kind and Region dropdowns and Detail modal are gone: its severity chips are the controls. (Summary panel)

Fewer false patterns. Scratches are recognised at any angle and donuts by their radial profile. A cluster is judged by its size against random placement, so purely random failures rarely read as one (3–5% of wafers at 2–10% fail rates), and a region is compared only with the regions that are not themselves deviant. Quadrants now hold equal shares, which can move a quadrant yield by up to two points on a wafer of a few hundred dies. (Pattern detection)

Smaller changes. Soft-bin-only data opens in soft-bin mode, and the reports link each item to its evidence.


2026-10-06 — A Reticle setting in tsmap, reticle cells and shots, and pass/fail tests as plot fields

Set the reticle, and see what repeats. In tsmap, Setup ▾ → Reticle… (or --reticle 4x3) takes the stepper field's size in dies. Every wafer then draws the field grid and is checked for failures that repeat at the same place in every field. In wafermap, a die's Reticle cell (its place inside the mask) and Reticle shot (where on the wafer the mask was placed for it) join ring and quadrant as categories to colour, compare and split yield by, and the Overview gains a Reticle cell yield grid: a cell that is low across every wafer points at the mask, not the wafer. (Plots, tsmap Reticle)

Pass/fail tests are plot fields. A functional test, or a derived test that gives a verdict, is a Pass or Fail category under Verdicts, so "Vth for the dies that pass the scan chain and those that fail it" is a box with two boxes. A derived test can also read a die's own dieX(), dieY(), hbin(), sbin() and site(). (Derived tests)

One name for each wafer attribute. Group by, the plot fields, the Wafers table, the header strip and the reports name an attribute the same way, and a host can set the names with the new attributes option. In tsmap, reference fields such as the program revision and tester details no longer clutter Group by and the plot fields.

Build your own plots, and keep them. The new Plot tab in Insights is a chart builder. + New plot opens a plot beside a large copy of its chart: choose scatter, histogram, box, bar or line, the field for X, Y and colour (any test, a die's position, a wafer's yield, or a lot field such as split, slot or temperature), and the titles, limits, log scale and direction of each axis. Yield by temperature or by split is a bar of pooled yield, a test per wafer is a box, a scatter can be coloured on a gradient by a third value, and one mark per wafer combines the dies' values by mean, median or another measure. Every change is drawn at once, titles write themselves until you type your own, and each plot states how many wafers and dies it shows and how many it left out for a missing value. A plot is a recipe with no lot in it, so it is kept for the next lot, offered in the right-click menu over any selection of dies, and can be exported and imported as a file to share. Delete all plots… clears the list after asking you to confirm (it says how many, and points to Export plots… for a copy), and Undo brings it back for ten seconds. Add examples draws one plot of each type from the lot to start from, and a title you typed is checked against its plot, so one that still names a test you have since swapped out is flagged. In tsmap your plots are remembered between sessions. A plot that needs a test the open lot lacks is kept, dimmed, with the reason, and it checks a test's name as well as its number, so a plot saved against another test program is never drawn against the wrong measurement. (Plots guide, screenshot, plot editor, Insights example, in tsmap, tsmap screenshot)

A Data tab in Insights. The last Insights tab shows the wafers in scope as tables, one at a time: Statistics (the test-values and functional-tests tables), Dies (one row per die with a column per test, sortable, and smooth to scroll at hundreds of thousands of dies) and Wafers (one row per wafer: metadata, die counts, yield and test means). Each has Export CSV, which writes the table as shown, and Copy. Click a die's row to see that die ringed on its map. A very large export is written to disk in pieces, so the page stays responsive. Exported numbers carry full precision rather than the screen's four figures. (Data tables guide, screenshot, in tsmap)

A sweep remembers what each test was called. A sweep opened on a lot from another test program that reuses the same numbers is not drawn as if it were the original: the card names the mismatch, leaves the test undrawn and does not measure the crossing or widths from it. (Sweep cards)

A plotted test's limits are drawn, and yield can be split by ring, quadrant, bin or die position. A plot of a test that has limits carries them as dashed lines, short for test limits and long for spec limits, each labelled with its value, and a plot's editor dims a choice that would leave a plot that cannot be drawn, with the reason. (Plots)

A chart saved as an image says what it is, and a printed chart says the same. Its title, the settings of its controls, its population and count, and its colour key travel with it, and the controls that only act on the screen are left off the printed page. (Exporting a chart)

Right-click, or click a mark, to chart and tabulate exactly those dies. The drilldown menu opens on a selection, a wafer, a gallery card or a chart mark, even on a map with nothing to chart, and offers the charts the dies support, your saved plots, New plot… and New sweep… drawn over just those dies, and tables: Dies, Test statistics and, for dies on several wafers, Wafers, the Data tab's tables over those dies with Export CSV and Copy. A saved file is named for the selection and carries a Wafer column. The Summary panel's "View die list" is now Data tables and opens the same tables over the wafer or lot, with Insights on or off. (Charting dies and wafers, screenshot)

  • A bin in the legend. Right-click a bin on a map's legend, or on the gallery's lot legend, to open the menu on that bin's dies, on that wafer or across the lot, without selecting them. With several bins filtered in, Only bin N and All filtered bins switch between them. (Bin legend filter)
  • A finding in the Summary panel. Right-click a finding to show it and open the menu on every die it selects, across all the wafers of a gallery, or on its wafers when it names no dies. (Summary panel)
  • A mark on an Insights chart. A bar of the bin pareto, a pass-rate row (the dies failing that test), a bar of the value histogram, a ring or quadrant of the yield diagrams, a level of a sweep curve, a drag across the wafer-to-wafer trend, and a plot's point, bar or box each open the menu on exactly the dies they count, and a Wafers-table row opens its wafer. (What a click does)
  • Plots and sweeps from a selection. The right-click Value histogram offers Edit as new plot, a saved sweep opens in its editor over just the selected dies, and a new plot or sweep is kept only if you add it. (Plots, Sweep cards)

Compare the same place across a lot. In a gallery, Select on every wafer applies a box or click made on any card at the same die positions on every card, so an edge ring, a scratch zone or a reticle corner is read lot-wide, and the right-click menu then charts and tabulates those dies across all the wafers. Ctrl/Cmd+click a card's header to pick whole wafers. In the Insights scatter, hover a point to name its wafer and die, click it to open that wafer, or drag a rectangle to select dies across wafers. (Select on every wafer)

Insights is on by default, with a Maps | Insights switch that stays in the same place in both views, and a slim Summary tab on the edge of the map to open the Summary panel when it is closed. The metadata strip fits as many of a field's values as the width allows. (Upgrading)

In tsmap, the column mapping and file filter read rows from throughout a file (the first five, then up to twenty spread through the rest), so a column that changes partway through shows as varying; the filter says which values come from a sample. In tsmap the desktop window also reopens where you left it, and a small lot imports every test without stopping at the test selector (Setup ▾ → Tests… filters afterwards). (column mapping, file filter screenshot)

For developers building on wafermap: insights.plots and onPlotsChange keep the My plots tab's list (with readPlotsFile and writePlotsFile for the file, and onPickPlotsFile for a native import dialog), insights.enabled defaults to true (pass { enabled: false } to opt out), onSaveText can receive a Blob for a very large table, and a map's right-click menu is always the drilldown menu. The steps are in Upgrading.

Two changes can move numbers you have seen before. Standard deviation (σ) now divides by n−1 everywhere. The Test Values table, the per-test statistics, the lot trend, lot drift and the report all use the sample standard deviation, the same definition the capability indices (Cp, Cpk) already used, so a σ in a table now matches the σ behind a capability figure. Dividing by n−1 is the usual choice when the dies you measured are a sample of what the process makes; it makes σ slightly larger, by about 2.6% for 20 dies, 0.5% for 100 dies, and not visibly at lot scale. (Upgrading) Yield follows your pass bins everywhere. If you state pass bins such as 3 and 5, every chart, table and report, including those opened from a right-click and the gallery's stacked views, now judges by them, as the map does, and ring counts follow the same rule. (Upgrading)

See: tsmap v0.1.44, wafermap v0.34.0


2026-10-02 — Multi-project wafers fill the map, a Summary and reports that lead with what matters, and lot drift

Compact layout for multi-project wafers. On a multi-project wafer each reticle holds only a few of one product's dies, so the map is mostly empty and every die is tiny. When the occupied columns and rows repeat at a regular pitch, the Overlays menu offers Compact layout: the empty rows and columns disappear and each group of dies is outlined, so your product fills the map. Nothing is left out. The legend, yield and findings are the same as on the wafer view, and hover text and axis labels still give the original die coordinates. In a gallery every card shares one layout, so wafers compare cell for cell, and the cards are only as tall as their maps need. Random missing dies do not repeat, so a wafer that merely has holes is not offered it. (Compact layout, example, in tsmap)

Layout diagnostics, for when a layout is not recognised. A Layout diagnostics row in the same menu shows what the detector saw as counts and scores only — no die positions, bins, test values or wafer names — with Copy and Save as file, so someone can report why a layout was or was not recognised without sharing their data.

Axis labels on request, and the XY indicator when you turn the wafer. The die coordinates along the edges appear when you zoom; an Axis labels row in the Overlays menu shows them all the time, and on a compact layout they mark each group of dies. The first rotate or flip switches the XY indicator on, so you can see which way X and Y now run; it stays on until you switch it off.

A wafer opened in its own window keeps its size when you open the Summary panel. The window widens by the panel's width instead of squeezing the map.

The Summary panel opens with "What stands out". A headline, then up to three items ranked by the dies each one costs — a region whose pass rate fell, a fail bin that rose, a wafer well below the lot, a test outside its limits — with the fail bins behind a region's shortfall named. Click a name to see it on the map. The HTML reports open with the same section and have a new layout: severity is three dots and a word, so it reads in a black-and-white print, bars sit behind yields and bins, and a table row is tinted only when it is one or more points below its reference. A click on a finding in a report opened from the panel shows it on the map. The panel and the reports now count bins, order findings and read yields the same way, and the weakest severity is called "Minor". (Summary panel, example)

Lot findings come from the lot. A spatial pattern that recurs on the same dies across wafers is found by stacking the lot's wafers and reported once, on every wafer it shows on. Adjacent regions that carry one signal are one finding ("Sectors E–N", "Rings 1–2"). Outlier wafers follow one rule, shared by the findings and the Wafer Yield list. For lots of five or more wafers, a yield or a test mean that drifts across the wafers is listed as a Watch line. (lot findings example)

In tsmap, the column mapping dialog opens ready to confirm. Columns that are recalculated on load (Ring, Quadrant, Edge excluded, a derived test's column) start as "— ignore —", numeric run conditions such as test temperature, operator, program, recipe and slot come in as Display info rather than tests, when two columns look like the same role the stronger match keeps it, the dialog says when a saved mapping is in use, and more X/Y names are recognised (chip_x, position_x, X Coordinate, Y Row and their equivalents). (column mapping, screenshot)

For developers building on wafermap: viewOptions: { compact: true } applies the compact layout whether or not it would be offered, and viewOptions: { showAxes } is the axis labels choice. sectorCount is 4, 8 or 16 (the compass names cover 16 bearings; other values are corrected to the default with an 'analysis-option-corrected' warning). Drift findings have ids drift:yield and drift:test:<number>. The report builders load when a report is opened, which takes about 10 KB gzipped out of what rendering a map downloads. A new Multi-project wafers example shows the compact layout.

See: tsmap v0.1.43, wafermap v0.33.0


2026-09-27 — Selections you can see, several bins at once, and lot findings that count every wafer

Selected dies stand out, whatever the colours. Selecting dies — a click, a dragged box, or a finding clicked in the Summary panel — now fades the rest of the wafer and outlines the selection, so the selected dies are the only ones in full colour. A finding's pattern stays readable inside its outline, and a ring-shaped region is unmistakably the ring, not the centre. Clicking the only selected die again clears the selection. (Box select)

Maps open ready to select. Dragging on a map now draws a selection box straight away; hold Space and drag, or choose Pan in the toolbar, to pan instead.

Show several bins at once. Ctrl/Cmd+click in the legend adds or removes bins (or metadata values), so related fails can be viewed together; a plain click still shows just one. Clicking a finding about a bin filters the legend to that bin, so the legend always names what the finding describes. (Bin legend filter)

Lot findings count every wafer that shows a pattern. A regional pattern — a bin, the yield, a limit failure or a test value that differs at the edge, in a quadrant, a sector and so on — is now tested on all the wafers' data together, and reported as, for example, "higher on 8/8 wafers, all wafers' data combined". Patterns confined to one region (a bin found only at the edge), yield losses of a few points, and edge rings made of scattered fails are now reported too, and edge-ring and edge-local wafers count as one edge pattern. (lot-level findings, example)

For developers building on wafermap: highlightBin and highlightMetadataValue accept a list as well as a single value, and a lot finding about a region carries each counted wafer's dies in highlight.dieKeysByWafer.

See: tsmap v0.1.42, wafermap v0.32.0


2026-09-26 — Large lots load far faster and use far less memory

The browser build opens lots of about 60 million test values (dies × selected tests) — for example 1.2 million dies at 50 tests each — where it previously stopped at around 200,000 dies regardless of test count. A file past that limit now fails with a message saying so and what to do, instead of a parser error or a crashed tab. Selecting fewer tests in the test selector raises how many dies fit, since a filtered parse only holds the tests kept. (parser performance)

Every load and analysis is faster, and every lot takes much less memory. A lot now moves from the parser to the map as columns — one array per test rather than one object per die — with no copying in between. On a 341 MB, 266,325-die, 51-test file: the desktop app goes from starting a parse to a finished gallery in about 15 seconds, where it took 36; the app's memory use with the gallery open falls from about 4.1 GB to about 1.2 GB. In the browser, the same file parses in about 6 seconds and holds about 330 MB, where it used to crash the tab outright. Regional test-value analysis (the Analyse findings) is two to three times faster, and now runs automatically on lots estimated to take under a second, with an Analyse button and a time estimate offered above that. (Performance)

For developers building on wafermap: test values and verdicts are now read from per-test columns rather than held as an object on every die — die.testValues/die.testPass still work, but each read now builds a fresh snapshot rather than returning a live object, so assigning to either throws. results can also be passed as columns directly (DieColumns) by a host that already holds its data that way.

For developers building on @wafertools/testdata-parser: every parse function now returns a compact columnar buffer instead of one JavaScript object per die, with a small decoder (decodeParsed/decodeColumns, shipped in the package) to read it back. This is the whole basis of the memory and speed improvements above; see the package's own docs for the exact shape.

See: tsmap v0.1.42, wafermap v0.32.0


2026-09-25 — Spec limits on the charts, files in different units combine, and a slimmer API

For developers building on wafermap — 0.31.0 is a breaking release: (Upgrading: removed exports)

  • The exports deprecated in 0.30 are removed — the low-level drawing pipeline, the chart-data and region builders, and a set of internal helpers — along with buildWaferMap's unused second argument and the showPartialDies/includePartial options. Each has a replacement, listed on the new Upgrading page; most integrations use none of them.
  • Values outside the STDF V4 ranges are now treated as missing by buildWaferMap, as 0.30.4 announced, with the input-values-outside-stdf warning saying what was left out; waferConfig.orientation accepts 0, 90, 180 or 270.
  • downloadFilename is a prefix for every saved file, CSVs and charts included, rather than the name of the map's PNG alone.
  • Each release now has a GitHub Release with its changelog notes.
  • The data layer is about 49 KB gzipped, down from about 61 KB.

If you have hand-written definitions files: lsl/usl columns are now read as spec limits. They used to be read as test limits. The log says so whenever a file uses them. To keep a file's values as the limits dies are judged by, rename those columns to lo_limit/hi_limit. Files saved by tsmap are unaffected.

Spec limits sit beside test limits on the charts. The Insights box plot, histogram, wafer-to-wafer trend and scatter now draw a test's spec limits (LSL/USL) as well as its test limits (Lo limit / Hi limit), each labelled and dashed differently so the two are never confused. When a test has both, a Limits choice switches all four charts together between both (the default), either kind, or none. Gridlines are lighter so the limits stand out, and the scatter marks a limit that falls outside its plotted range at the edge, as the other charts already did. The wafer map itself still judges pass/fail by the test limits. (Spec limits)

Spec limits can come from your own files. A test-definitions file sets spec limits with lsl/usl (or lo_spec/hi_spec) columns, beside test limits in lo_limit/hi_limit, and a long-format CSV, JSON or Parquet file can map columns to either kind. Process capability is measured against them, and Save definitions writes both. Every column name now means exactly one kind of limit, and a name that doesn't say which (min, max, lower, upper) is ignored with a warning rather than guessed. (Test definitions file)

A test recorded in mV by one file and V by another now stays in the lot. tsmap used to treat the two as different measurements and leave the test out of every view that compares wafers. It now converts the later file's values and limits to the unit the first file uses, and says so in the log. Only a difference of SI prefix on the same unit is converted; anything else is still kept apart. A test-definitions file whose limits are written in another prefix is converted the same way. (Opening several files)

Every format now reads values to the STDF ranges. CSV, JSON and Parquet join STDF and ATDF: a bin, position or site number the STDF format cannot hold, or a reading that is not a finite number, is treated as missing and reported, rather than plotted as if it were ordinary data.

New warnings are visible without opening the log. tsmap's Log button shows how many warnings and errors have arrived since you last looked, such as "▲ 3 new warnings".

See: tsmap v0.1.41, wafermap v0.31.0


2026-09-25 — Spec limits, correct retest matching, and test data read to the STDF spec

Spec limits and test limits are now shown and used separately. A test can carry both the limits it was judged against on the tester (Lo limit / Hi limit) and the process's own specification limits (LSL/USL) — the two are frequently different, and conflating them has always been a source of confusion. Process capability (Cpk and friends) now measures against a test's spec limits when it has them, and against its test limits otherwise, and says which one it used. Every chart, table and report labels the two apart. (Spec limits)

A result exactly on a limit is judged the way the file says, not assumed. STDF and ATDF each record whether a value equal to a test's limit counts as a pass or a fail; that rule is now read and applied everywhere a die's pass/fail status is shown — the map, the yield figures, the Summary panel, every chart.

Retests are matched correctly even when a file gives no die position. Some testers record retests by part ID rather than by X/Y, and previously such a lot had no reliable way to say which record superseded which. Retests are now matched by part ID within a wafer when the file has no position for them, so retest handling (first/best/worst/last) resolves correctly instead of treating every attempt as a separate die. (Handling retests, example)

Test data is read exactly to the STDF/ATDF specification. Bins, coordinates, test results and part IDs are all read to the value ranges and validity rules the formats define — including which recorded results the tester itself marked as usable — so what you see matches the tester's own judgement of the data. A value outside what the spec allows is reported rather than plotted as if it were ordinary.

Derived tests and sweeps can be cleared, and tsmap warns when they won't fit. The test selector has a Remove derived tests action, and Setup ▾ → Sweeps… has Clear — both previously had to be redefined from scratch to remove. After loading a file, tsmap now also warns when none of your derived tests could be computed, or when a sweep names none of the lot's tests, with a button to remove just the sweeps that don't apply. (derived tests in tsmap)

For developers building on wafermap:

  • TestDef.specLow/specHigh — spec limits, read separately from limitLow/limitHigh. TestCapability.limitBasis says which pair a given test's capability was measured against.
  • TestDef.limitLowInclusive/limitHighInclusive — whether a value equal to a limit passes, read from the file and honoured by every pass/fail surface.
  • DieResult.supersedes — set when the tester marked a record as replacing an earlier one; it always wins, whatever retestPolicy says.
  • DieResult.partId/Die.partId now accept text as well as numbers, as STDF and ATDF define them.

See: tsmap v0.1.40, wafermap v0.30.4


2026-09-23 — Tests computed from other tests, response curves, and charts of just the dies you pick

You can now define a test as a formula of other tests, and it works everywhere a measured test does. A switching window (t[1202] - t[1212]), a leakage shift, a ratio, a log of a current, "did every step of this sweep pass" — each becomes a test you can plot on the map, give spec limits, and see in every chart, table and report. Anything computed this way is marked † wherever it appears, so it is never mistaken for something the tester measured, and a die missing one of the inputs shows as no data rather than a misleading zero. In tsmap, add an expression column to your test definitions file. (Derived tests example, in tsmap)

Sweeps: read a run of tests as a curve. Many test programs measure one thing at a series of conditions — voltages, temperatures, bake times, cycle counts, resistance thresholds — and record each step as its own test. Declare them as a sweep and Insights draws them as a response curve (the median with a p10–p90 band), and measures a pair of curves against each other: where they cross, and how far apart they are at the levels you choose. The swept value can be given directly or read straight from the test names (LEVEL_12K is 12 kΩ), steps that grow by multiples can go on a log axis, and the axis can carry a unit. In tsmap, load a sweeps file from Setup ▾ → Sweeps…, or pass --sweeps on the command line. (Sweeps example, in tsmap)

Right-click to chart just the dies you care about. Select a cluster or a scratch on a map and right-click — or right-click a wafer in a gallery, or one wafer's bar in a chart — for a histogram, process capability or any sweep drawn from only that population. The chart says exactly how many dies it used and from which wafer, so it cannot be mistaken for the whole lot. The same menu is on the map's new Chart button, and on the Menu key. (Charting dies and wafers)

An expanded chart now uses its window. Charts opened in their own window used to stay the size they were in the grid; each now grows in the way that suits it — plots fill the window, lists of wafers or bins show every row, and circles and matrices grow to fit.

tsmap's file dialogs now remember where you were. Opening data, saving images, exporting, loading definitions and saving filters each reopen in the folder you last used for that task — on Linux too, where they used to start at your home folder every time.

For developers building on wafermap:

  • derivedTests on buildWaferMap — a TestDef plus an expression. Parsed to a typed tree and walked, with no eval and no expression engine, so a set is safe to share as JSON.
  • insights.sweeps — sweep definitions with test ranges ('1200..1230'), xValues or xFromName (a {x} placeholder pattern, deliberately not a regex), xUnit and xScale: 'log'.
  • Drilldown needs no wiring and has no option: a map takes over right-click only when there is something to chart, so a host's own context menu still works on bins-only maps.

See: tsmap v0.1.39, wafermap v0.30.3


2026-09-21 — Big lots load far faster, and tsmap warns before a crash

Loading a large gallery is now up to 20× faster, with no single pause long enough to make the page look frozen. A gallery's shared legend and Summary panel used to rebuild from scratch after every wafer arrived, so a progressive load got slower the more wafers it had — 182 seconds for 50 wafers of 8,000 dies each. The same lot now loads its cards in about 8 seconds, with no single pause longer than half a second. A lot's Summary panel, which used to lock up the browser for over 15 seconds on the largest lots, now fills in piece by piece instead of freezing.

tsmap warns you before a web-browser load that would crash the tab, instead of just crashing. Above roughly 200,000 dies the browser build can run out of memory outright; it now checks for that specific risk — separately from its existing "this may be slow" warning, which didn't cover it — and tells you before you commit to the load. The desktop app was never affected. (tsmap on the web)

tsmap's loading screen is now one clear, consistent indicator for every stage of opening a file — scanning, parsing, analysing, rendering — instead of a small spinner that vanished on narrower windows and didn't always agree with the rest of the screen about whether something was still happening. A new "Log phase timings" option in the Help menu can record exactly how long each stage took, useful if you need to report a slow load.

A very large wafer or lot could fail outright with an obscure "Maximum call stack size exceeded" error. Anything at or above roughly 100,000 dies could hit this in several places — now fixed everywhere it could occur.

Correlation charts on very large lots no longer hang. Above 25,000 dies, the correlation panel now computes from an evenly-spread sample of the lot rather than all of it, and says so plainly when it has. The numbers you see are unaffected either way — the sample is large enough that the values wouldn't change. (Performance)

Smaller fixes:

  • A single bad test value no longer turns that whole test's statistics into "NaN" in the Summary and Insights panels — it's now correctly left out instead.
  • Switching between Insights tabs and back no longer re-runs the whole analysis a second time.

For developers building on wafermap:

  • renderWaferGallery can now report its own loading progress. onItemResolved(resolved, total) fires as each card is built, and onItemsResolved() fires once the whole gallery — including its shared legend, colours and Summary panel — has actually settled. tsmap's new loading indicator (above) is built on this.
  • A sampled correlation matrix says so in its data, via CorrelationMatrix.sample, not just in the panel — for hosts reading matrices directly.

See: tsmap v0.1.38, wafermap v0.30.2


2026-09-17 — A security fix, exports named for their data, choosing a test on a phone

A security fix: please update. This release fixes a security issue in how names from a data file are displayed. It affects earlier versions of both tsmap and wafermap, so update, particularly if you open files from sources you don't control.

Saved images and CSVs are named for the data they came from, for example LOT123_W05_hard-bin.png or LOT123_25-wafers_yield-by-wafer.png. Exports used to have fixed names such as dies.csv, so the same export from two wafers saved as dies.csv and dies (1).csv, with nothing to say which wafer each held. Anything the data doesn't have is left out rather than guessed: a wafer with no ID gets no wafer part in the name.

You can choose a test from the plot-mode menu on a phone or tablet. With more than six tests, "Test Value ▶" opens a list, and that list only opened when a mouse hovered over it. A tap opened it and closed it again straight away. It now opens with a tap, and with Enter or Space from the keyboard, where the arrow keys move through the tests.

Ring, quadrant and reticle lines are clear on high-resolution screens. Map lines and markers were drawn at a fixed number of screen pixels, so on the high-DPI displays most laptops now have they came out at half their intended width or less and were hard to see. They now look the same at every display scale. Nothing changes on a standard display.

A detached gallery card no longer shows two sets of window buttons. In the tsmap desktop app a detached card opens in a window drawn inside the app. Maximized, its header sat just under the app's title bar, and its minimize button looked like the real one but only shrank the card. That button is now Collapse, with a different icon, and it is hidden while the window is maximized.

Smaller fixes:

  • tsmap's Filter files dialog no longer shows a "selected" count in its title that never changed. The count above the table is the live one.
  • A gallery with a fixed number of columns uses the full width instead of leaving most of each row empty.
  • Changing the bin colours in an expanded gallery card recolours that card.
  • A card put back into the gallery matches the other cards again instead of keeping the view it had while expanded.
  • Outlined buttons in the Summary panel have a visible border in dark themes.
  • A lot report analyses its wafers when no summary was prepared beforehand, instead of reporting empty wafers.

For developers building on wafermap:

  • Five of the exports deprecated in 0.30.0 are staying after review, because each is the only supported way to do something an app needs: visibleFindings, openReportModal, metadataDisplayValue, getReticleCell and renderFindingsReportHtml. They no longer log a notice.
  • The rest now have replacements. Analysis results include process capability (Cp, Cpk, Pp, Ppk), pass rates by the tester's recorded verdict, ring and quadrant yield and the spatial pattern. There are also new report functions that take built maps, getBinColors() on both controllers, and buildWaferMap({ layout: true }) for a die layout with no test data. Every remaining deprecation notice names what to use instead.
  • downloadFilename becomes a file-name prefix in 0.31.0. Until then it still names the map and gallery PNG, and logs a notice once.
  • Text bins and test values raise a warning. A CSV parser gives every field as text, and a bin of "1" is not pass bin 1: yield read 0% with nothing to say why.

See: tsmap v0.1.37, wafermap v0.30.1


2026-09-15 — A bin keeps its colour in every lot; your pass bins count everywhere; tsmap works offline

A bin is now the same colour in every lot. The 2026-09-11 release coloured bins by how many dies each held: the biggest failing bin took the first fail colour. That kept colours distinct within one view, but it meant a colour stood for "the largest failure here" rather than a particular bin. Bin 7 could be red in one lot and brown in the next, and filtering a gallery could recolour bins. Wafer-map tools across the industry tie colour to the bin number, so engineers can learn a program's colours and compare screenshots by eye. wafermap does that again. It still follows pass and fail: a passing bin never takes a fail colour, a failing bin 1 is never green, and colours from your bin definitions file still win. Hard bin 5 and soft bin 5 are different colours, so the two maps can't be mistaken for each other. Screenshots of bin maps will change colour. (how bin colours are assigned, example)

Pass bins from your file now count everywhere, not just in the yield figure. tsmap reads each wafer's pass bins from the file. Until now they reached only the headline yield number, and everything else assumed bin 1 was the only pass: findings, yield statistics, bin colours, the Summary panel and report, region yield, the Insights yield charts and the gallery strip. On a program where bins 1 and 2 both pass, bin 2 showed as a failure beside a yield that counted it as a pass. Every surface now uses the pass bins in your file. In a lot that mixes test programs, each wafer is judged by its own pass bins, and a warning names any bin that passes on one wafer but fails on another. (Pass bins and yield)

Soft-bin tables and yield are right too. They had been judged against hard pass-bin numbers, so the gallery's soft-bin yield could read close to 0%. Soft bins are now judged by their own results.

Bin legends list passing bins first, then failing bins from most dies to fewest, the same order as the Summary panel and the Insights pareto. The map legend and gallery legend used to sort by bin number, so one lot was listed two different ways on one screen.

Insights now works for files without die coordinates. On a file with no X/Y positions, the "No die position data" panel stayed on top of Insights and hid most of the charts. The same happened on a file where only some dies have coordinates. (Dies with no reported position)

The tsmap browser app now works offline, and Chrome, Chromium and Edge can install it as an app with its own window and launcher entry. It is aimed at platforms tsmap ships no installer for, such as RHEL. Firefox on the desktop can't install web apps, but the offline caching still works there. The whole app is cached on first visit, including the file parser, so an offline tsmap can still open files. The user guide's text works offline; its screenshots load the first time you view them online. Updates are never applied behind your back: tsmap asks first, because reloading discards anything you have loaded. None of this changes the desktop app. (installing tsmap as an app)

Wafer configuration details are in plain language. The info row used to show raw STDF codes such as "Wf Flat: D · Pos X: R". It now reads "Wafer Flat: Bottom · X Increases: Right", with units on every size.

Opening a single-wafer file no longer stops to ask for a wafer label when the file's lot ID already identifies the wafer.

For developers building on wafermap: 78 exports are deprecated and will be removed in 0.31.0. They are the low-level drawing pipeline (buildView, toCanvas and related functions), the chart-data builders the Insights tab now draws for you, and helpers that were exported by accident. Each still works in 0.30.0 and logs one console notice the first time you call it, saying what to use instead. If you depend on one with no replacement, open an issue before 0.31.0. The passBins options on the analysis and render functions are removed: give pass bins to buildWaferMap and everything downstream uses them. The changelog lists every name.

See: tsmap v0.1.36, wafermap v0.30.0


2026-09-12 — A clearer default colour scale for value and stacked maps

Value maps and stacked maps now use the Viridis colour scale by default, and they will look different. The old scale ran blue → cyan → yellow → red. That is a rainbow, and rainbows have a specific problem for reading data: their brightness does not rise steadily. Cyan and yellow are both near maximum brightness while blue and red are much darker, so two genuinely different readings could look equally intense — and because hue changes fast around cyan and yellow, the map drew ring-shaped boundaries that no process step put there. Engineers told us they could not judge how far apart two colours were. Viridis brightens steadily from one end to the other, so the colour you see corresponds to the value you have, and a dense grid of dies shows small differences honestly. (Colour schemes example)

On a stacked map it also puts the emphasis the right way round. The healthy bulk of the wafer — where nothing failed — now sits back as dark ground, and the edge ring, scratch or cluster you are actually hunting for is the bright thing on it. Previously that was inverted: the good area glowed and the defects were dark specks.

If you preferred the old look, it is still there. Pick Jet from the Colour scheme menu — it is the same rainbow family, now honestly labelled as one, alongside a note that rainbow scales trade read accuracy for familiarity.

Every scale now runs the same direction: dark for low, bright for high. Four of them — Cividis, Plasma, Inferno and Greyscale — had been running inverted, which meant switching colour scheme silently flipped which end of your wafer looked "hot". They now match their published definitions, so a scale you recognise from elsewhere reads the way you expect.

New Reverse gradient option. Some parameters have their interesting end at the bottom, and monochrome print has its own convention that more ink means more. One tick box in the same Colour scheme menu flips whichever scale you have chosen, including one your own application registered. tsmap remembers the setting across restarts, alongside your bin and value colour choices. (Custom colour schemes)

Also new: the Mako colour scale, a blue-to-pale-green alternative with a little more separation at the top end than Viridis.

See: tsmap v0.1.35, wafermap v0.29.0


2026-09-11 — Bin colours you can trust; STDF and ATDF together; host tsmap on your own intranet

Bin colours now follow pass and fail, and different bins no longer share a colour. wafermap used to pick a bin's colour from its bin number, so two bins could be drawn in the same colour and bin 1 came out green even when your pass bins said it failed. Colour now follows the pass bins you actually use — passing bins take green pass shades, failing bins the rest, ranked by how many dies they hold — and the palettes were re-chosen by measurement so every pair stays distinguishable, including a colour-blind-safe set. When a lot has more bins than a palette can separate, the map says which bins share a colour rather than letting you assume they differ. Bin maps therefore look different from before; the change is deliberate. Bin and value colours are now separate choices, and tsmap remembers both across sessions; a bin definitions file can also set a bin's colour. (how bin colours are assigned, example)

Two parser fixes change numbers you may have relied on. STDF wafer geometry (the WCR record: wafer size, die size, orientation) was read two bytes out of place, so any file that carried it had its geometry misread; and ATDF files were read in STDF's field order rather than ATDF's own. Both are now read by their specifications. Separately, a file holding several lots, test temperatures or programs now loads as it should: CSV/JSON/Parquet rows used to be grouped by wafer ID alone, merging every lot's W01 into one map, and a concatenated STDF with several lot records labelled every wafer with the last lot. Each wafer now carries its own lot.

STDF and ATDF load together, since they are the same records in binary and text. The file filter got easier to read too: columns lead with name, lot, wafer count and when each file was Tested; columns with nothing to show for the files in view are hidden; Parquet files report their wafer count; and when a scan mixes kinds of file, one click chooses which kind to show. (Scanning a folder)

tsmap now ships as a static bundle you can host yourself. Every release carries tsmap-<version>-web.zip alongside the desktop installers: unpack it into any web server's document root, browse to it, and that is the whole procedure — no installer, no server-side component, and no internet access needed at any point, installation included. Files are still parsed in the browser, so nothing reaches your server either; it only ever serves the application. (hosting tsmap yourself)

This is the answer for a site that cannot reach the public internet, or whose policy says lot data must not touch an external origin. It is also the only dependable offline option in a browser: the hosted app keeps working without a network once loaded, but only because the browser cached it — a hard refresh, an evicted cache or a different profile all send it looking for our servers again. A copy on your own machine has no such dependency. Two things catch people out and are documented: .wasm must be served as application/wasm (nginx, Apache and Python get this right; IIS needs a MIME entry added), and it must be served over http/https rather than opened as a file.

Reload the same test definitions in one click. If you follow a product or test-program series, you were re-navigating the file picker for the same definitions file on every dataset. Load definitions now has a ▾ listing the files you recently loaded or saved, wired into the test selector and the Setup ▾ definitions dialogs. On the desktop the file is re-read from disk, so you always get its current contents and are told when it has changed. In a browser there is no path to re-read — a browser picker hands a page contents, never a location — so the entry is the copy taken when you last used it. That is fine for a selection or a set of renames and needs care with spec limits, so the menu says so, each row dates its copy, and applying one logs a warning naming the file when it sets limits. Re-pick through Choose a file… if the file may have moved on. (definitions files)

The file and setup menus are tidier. Recent files moved into Open files ▾ beside Choose files… and Scan a folder… — recents are a way of answering which file, which is the question that caret exists for — so the separate Recent button is gone. The Lot ▾ menu is now Setup ▾, grouped into Tests & bins, Wafers and Analysis, because no single word describes six unrelated dialogs. And its two overlapping test entries are one: Filter tests… and Test definitions… read and wrote the same file to the same effect, differing only in whether the data was re-parsed — a consequence of what you changed, not a choice worth putting to you. Setup ▾ → Tests… now covers selecting, renaming, limits, units, type and save/load together.

You can see and clear what tsmap remembers. Help → Reset saved settings… lists what is actually stored on your machine — theme, recent files, saved column mappings, wafer splits, diameter overrides — explains each one in a line, and forgets only what you tick. Until now a wrong saved column mapping was reapplied on every load of that layout with no way to undo it short of browser developer tools. Nothing here ever leaves your machine, and loaded wafer data is never stored. (What tsmap remembers)

Two fixes worth naming. The last-used directory was never remembered on Windows — it had been resolving its state file from a variable Windows does not set for a GUI application, so the picker silently always opened at the default; it now asks the OS for the right location on all three platforms. Your settings carry over this update even though the desktop app's internal identifier changed with the move to the wafertools organisation. And tsmap's About dialog now names the engine underneath (wafermap 0.28.0), reported by the bundle itself rather than a manifest — when a map looks wrong, which version drew it is the first thing worth knowing.

wafermap makes that version publicly readable (WMAP_VERSION) so any application embedding it can do the same, and fixes two things adopters hit: analyzeWaferMap raised an "option corrected" advisory for an option that was never set — the most ordinary thing a host writes, forwarding an optional value that happens to be undefined — and buildWaferMap with an empty testDefs array silently lost every value plot mode and test column. In a single-wafer view, clicking a box in the test-value distribution now opens the map on that test, as it already did in a gallery. Its documentation site also gained a light/dark theme, a table of contents that follows your scroll on the long API reference, and hover definitions for domain terms like STDF, PTR and Cpk.

See: tsmap v0.1.34, wafermap v0.28.0 — a breaking release for applications built on wafermap: see its changelog for the colour API changes.

2026-09-09 — Spec limits stay with the file they came from; scan a folder; test-value findings run on their own

If you load several files at once, this release fixes a correctness bug worth knowing about. tsmap merged every file's test definitions into a single list — last one wins — and then plotted every wafer against it. A test number identifies a test within one test program, so across files it is not an identity at all: test 1001 is a threshold voltage in millivolts in one of tsmap's own sample lots and a leakage current in nanoamps in another. Loading them together silently discarded all but one definition, then labelled, scaled and capability-scored every lot against the survivor — box plots of nanoamp readings titled vth_n_mV with that test's spec line drawn across them, a process-capability panel reporting Ppk −1489, and drilled-in maps showing every die out of spec. Nothing warned, because by the time the data reached the map every wafer carried the same already-wrong list.

Definitions now stay with the file that declared them — wafermap gained per-wafer test definitions to make that possible — and where two files genuinely disagree about a test number you are told at scan time, before you choose, rather than after the load. The imported data was never wrong: tests are selected by number. What was wrong was the name, the units and the limits it was shown against. If you have ever loaded more than one lot into tsmap at once, this is the release to move to.

Test-value findings now run by themselves when they are cheap. The regional analysis of parametric values lived behind a menu item, off by default — so a reader looking at the Findings panel got no hint that a whole category of finding had been skipped, and nothing would ever send them into the menu to discover it. tsmap now estimates the cost: under about a second it simply runs the analysis and notes it in the log; over that, the Findings panel itself carries a row saying what is missing, how much data it would cover and roughly how long it would take, with an Analyse button. The offer and its price arrive together, where you are already looking. (Value findings)

Point tsmap at a folder instead of picking files. Open files and Add files each gained a ▾ offering Choose files… or Scan a folder…, and dropping a folder onto the window now scans it (previously that failed outright). A folder picker asks where to look, which the file-triage table cannot answer — so the two steps stop asking you the same question twice. Subfolders are opt-in and only offered when there are any. (Scanning a folder)

wafermap adds the chart a split experiment is actually run to produce — a per-test pass-rate chart broken out by group — plus a wafer-to-wafer trend chart, population tiles on the Insights overview, split comparison in the lot summary report, and a dedicated Insights example (the library's largest feature surface previously had no example of its own). Geometry checking gained its missing half: a supplied wafer diameter that is too large for the dies used to pass in complete silence, and is now reported. The old inferred-pitch advisory is gone — it fired on every build that supplied a diameter without a die pitch, which in tsmap left a permanent red banner on screen with nothing available to dismiss it.

Smaller things: an About tsmap dialog (the app previously stated its version only in the log and its licence nowhere), eight more colour themes taking the picker from 8 to 16, interface zoom with Ctrl/Cmd +/−/0, and resizable columns in the file-filter table. Since the last entry, both projects also fixed the in-app user guide rendering as plain light — or as pale, near-illegible text on white — under every dark theme.

See: tsmap v0.1.33, wafermap v0.27.0 · earlier: tsmap v0.1.32, wafermap v0.26.1

2026-08-28 — Bin definitions, wafer diameter & edge-exclusion overrides, reports open in-app

tsmap can now read, save and load bin definitions — human-readable names for hard/soft bin numbers, plus which ones count as pass. STDF/ATDF already carry this in their own HBR/SBR records; CSV/JSON/Parquet, which have no such record at all, can now load definitions from a CSV via a new Load bin definitions… button in the column mapping step, and any format can save/load them afterwards from Lot ▾ → Bin definitions…. (Bin definitions)

A new Lot ▾ → Diameter & edge exclusion… dialog lets you override the wafer size tsmap infers from the data and add an edge-exclusion band — dies within it are dropped from yield and shown dimmed on the map — useful when the data is too sparse for a reliable automatic guess. Both are also settable at launch via --wafer-diameter/--edge-exclusion. A new Help → Definitions file formats… dialog saves a filled-in example of any of tsmap's three definitions-file types (tests, splits, bins) without needing to load data first. (Diameter and edge exclusion, definitions files)

wafermap's summary and lot reports now open inside the app instead of a separate browser tab — closing a longstanding issue where the report could open successfully but sit hidden in an unfocused, sometimes-minimized window, looking like the button had silently done nothing. The new in-app view has its own Print/Save-as-PDF button, plus an "Open as full page" link for anyone who wants a real separate page. The built-in help guide picked up the same in-app Print/PDF button, a combined table of contents that no longer reads oddly across a host app's own guide content, and a find-in-page search box for the one case where a browser's own Ctrl+F can't reach it (an embedded desktop app like tsmap).

CSV exports (die lists, test/bin/split tables) are now protected against formula injection — a cell value starting with =, +, -, or @, sourced from metadata this library didn't originate (an operator's free-text note, a MES field), used to be read as a live formula by Excel/Sheets/LibreOffice on open; it's now safely escaped.

Since this entry: the Lot ▾ menu named above became Setup ▾ in tsmap 0.1.34, and its two test entries merged into one Tests… — so the paths here read Setup ▾ → Bin definitions… and Setup ▾ → Diameter & edge exclusion… on current versions.

See: tsmap v0.1.31, wafermap v0.26.0

2026-08-23 — A live demo of launching tsmap from your own page; dark themes readable again

If you're wiring tsmap into a data-selection tool of your own, there's now a live demo of the launch-from-a-link flow: pick a sample dataset and get real, working links for both the browser build (?dataUrl=) and the desktop app (tsmap://open?url=...), plus the equivalent command line. Previously this was described in the docs but there was nothing to click. (opening data from a URL)

tsmap's three dark themes (Dark, Nord, Solarized Dark) had borders and secondary text that measured as low as 1.04:1 against their own backgrounds — effectively invisible rather than deliberately subtle, with the empty-state "Supports STDF, ATDF…" subtitle the most visible casualty. Every one of those values has been re-picked per theme and checked against each surface it actually renders on. Light themes were unaffected. tsmap also picked up a new app and tab icon — a wafer glyph in place of the old placeholder.

Two loading fixes: a gzipped sample file could fail to open in the browser build with "The compressed data was not valid" (the server had already decompressed it, and tsmap decompressed it a second time); and a sparsely-positioned wafer whose dies all landed in one row or column could render visibly stretched, non-square dies — cosmetic only, die counts, bins and yield were always correct.

See: tsmap v0.1.28, wafermap v0.23.1

2026-08-16 — Filter large batches before loading; wafer maps handle data with no position info

tsmap can now scan a large batch of files — a whole directory of lots, say — for their lot/wafer metadata before loading any of them, and show the results in a sortable, filterable table. Pick just the files you actually want from many, rather than loading everything and sorting it out afterward. The scan reads only header-level information (lot ID, part type, tester, wafer count, dates, etc), so it stays fast even across a large batch. A filter can be saved and reloaded, and the last one you used is remembered automatically. (file filter screenshot)

Both wafermap and tsmap now handle wafer test data that has no reported die x/y positions at all — a wafer-number-only test log, or a lot where some dies report a position and others don't. This used to be silently dropped; it's now kept, with the wafer shown as a bin breakdown or value histogram instead of a fabricated map (which would risk being misread as real spatial data), plus a full exportable die list. Yield, bin counts, and per-test statistics are unaffected either way — only spatial analysis (rings, quadrants, clustering) doesn't apply to a die with no position. (in wafermap, in tsmap)

Also: a pass over keyboard and screen-reader accessibility across wafermap's summary panel, chart legends, and die-list tables.

See: tsmap v0.1.27, wafermap v0.23.0

2026-08-15 — Parquet support, load data straight from a URL, native file associations

tsmap now opens .parquet files alongside STDF/ATDF/CSV/JSON, through the same column mapping used for CSV/JSON — useful if your data pipeline already lands in a columnar format rather than raw test-log files. (column mapping)

If your data lives behind an API or on a server rather than on your own filesystem, tsmap can now fetch and load it directly: tsmap --url <url> --url-format <format> on desktop, or a ?dataUrl=&dataFormat= link for the browser build — no need to download a file by hand first. A caller application (your own data-selection tool, a script) can hand tsmap a URL and have it open immediately. Header-based authentication is supported on desktop for APIs that need it. A tsmap://open?url=... link lets a web page launch the desktop app directly with data to load, and on Windows/macOS/Linux tsmap can now register itself as the default handler for .stdf/.atdf/.parquet files, so double-clicking one in a file manager just opens it. (opening data from a URL, file associations)

See: tsmap v0.1.26

2026-08-09 — A full accessibility pass; test numbers stay stable when you reorder columns

tsmap had a full UI and accessibility review: keyboard focus is now visible everywhere, modal dialogs trap Tab and restore focus correctly, every theme now meets WCAG AA contrast for primary buttons, and several dialogs gained proper Escape/Enter/backdrop-click handling that some had been missing.

For CSV/JSON data, test numbers are now a stable identity derived from the test's own name or source column, rather than the order it happened to appear in the file — so reordering or adding/removing columns no longer silently reshuffles which test is which. Loading a previously-saved test selection now recovers a renumbered test by matching its name instead of losing track of it.

See: tsmap v0.1.25

2026-08-04 to 2026-08-07 — wafermap tells you when it's had to skip something; a downloadable offline examples package

wafermap now surfaces its own advisories directly in the UI — a warning indicator appears whenever, for example, it had to guess at wafer geometry from the data, or capped analysis at 250 tests and skipped the rest. Previously some of these were invisible unless you went looking; now there's always a visible signal when something might be off, in both the standalone map view and tsmap, which picked this up automatically. Findings that used to repeat the same fact several times (a hard-bin row, its identical soft-bin twin, and a yield row all restating one edge failure) now collapse into one.

For anyone evaluating or integrating wafermap itself, every example now ships as a downloadable, offline-capable archive — unzip and open, no build step, no network required, which matters on a locked-down fab network.

See: tsmap v0.1.24, wafermap v0.22.0

2026-08-02 — More accurate edge-die yield

tsmap's displayed yield numbers are now more accurate for wafers where the physical geometry has to be inferred from the data rather than read from an explicit map file: dies at the wafer edge that were actually tested were sometimes wrongly treated as falling outside the wafer and dropped from yield and statistics entirely. Every affected die was a genuine tested site, so recovering them means displayed yield can shift slightly, and is now correct.

See: tsmap v0.1.23