Analytical

Honeycomb · Trace / cart checkout September 11, 2026 Honeycomb sandbox, public dataset (signed out) dense · light · desktop-web Toggle hotspots

Seventy-one spans over 3.288 seconds for one checkout, and the shape gives the answer away before you read a single duration. Two thirds of the way down, getDiscounts runs for 2.576s—more than three quarters of the whole request —and underneath it nine visible SELECT spans step down and to the right in a staircase, each starting after the last one finished. The badge on the parent says 19. Nineteen queries in a loop, run one at a time, and the waterfall says so by its outline rather than by any number. That is the shape worth learning: siblings overlapping means concurrency, siblings in a staircase means something that should have been one query. The panel top right is the other good idea here—it plots the distribution of this span's duration across the whole dataset and marks where this particular trace fell, so you can see whether you are looking at a normal request or the tail before you start optimising.

Trace waterfall Indentation is causality, length is duration, horizontal position is when it started. Six levels deep here.
Trace waterfall The staircase: nineteen SELECTs one after another inside getDiscounts. Batch the query, don't add a machine.
Percentile summary This span's duration against the whole distribution, with this trace marked—so you know if you're looking at the tail.
Detail on demand Selecting a span fills the right pane with its fields. The waterfall never moves while you read.
Overview then detail A minimap of all 71 spans above the list, so the shape of the whole trace is visible before you scroll it.
Categorical series palette Five services, five hues, and the name in a column beside every one. Colour is never carrying it alone.

Tableau Public · Cohort Retention Analysis September 11, 2026 Tableau Public embed view; workbook published by Tyran Christian medium · light · desktop-web Toggle hotspots

The triangle, exactly as the entry describes it: thirteen monthly cohorts down the side, thirteen periods since joining across the top, and each row one cell shorter than the last because younger cohorts have had less time. Reading down column two gives the comparison the grid exists for—37%, 22%, 19%, 15%, then a slow climb back to 24%—which no single averaged retention number could show. Two things undercut it. The first column is 100% for every cohort by definition, and it takes the darkest step on the ramp, so the one column carrying no information anchors the scale and compresses every real value between 2% and 50% into what is left. And look along the staircase edge: 9%, 4%, 3%, 2%, 4%, 2%, 4%, 2%, 3%, 4%. Every row's final cell falls off a cliff relative to its neighbour, because that period is still in progress and is being drawn as though it were complete.

Cohort grid One row per cohort, one column per period since joining. The triangle is the shape, not a rendering accident.
Cohort grid 50% then 9%. The last cell in every row is a partial period drawn as a finished one.
Sequential and diverging scales Column one is 100% for everyone and takes the darkest step, so a constant sets the top of the ramp.
Ratio and rate Cohort dates, and no cohort size. A 100% first cell could be ten people or ten thousand.

Tableau Public · Emergency Department / Clinical Dashboard September 11, 2026 Tableau Public embed view; emergency department patient flow workbook dense · light · desktop-web Toggle hotspots

Worth recording partly for what it is not. I went looking for a bed map, and what the public web has instead is this: analysis about an emergency department rather than the board the department actually runs on. There is no row per bed, no occupancy, no waiting-for column. Track boards live inside the patient record and never leave it, which is why that pattern has no example here and probably never will. What this does have is the punch card the calendar-heatmap entry names as the better answer for anything with a daily rather than weekly shape: weekday down the side, hour of day across the top, and the busy band from roughly ten to twenty-one is legible instantly without anyone labelling it. Its ramp is the problem—a blue-to-orange diverging scale on a patient count, which has no meaningful centre, so the midpoint sits wherever the data happened to average. The treemap beneath it degenerates into a mosaic of unlabelled slivers about a third of the way across.

Calendar heatmap The punch-card variant: hour of day against weekday. The busy band reads in a second, from the layout alone.
Small multiples Twelve month panels on one shared y-axis, so the seasonal fall from 265 in May to 52 in December is comparable across all of them.
Sequential and diverging scales A diverging blue-orange ramp on wait time, which has no meaningful centre, so the midpoint is wherever the mean fell.
Target and progress Each unit against a median reference line, green below and red above. A target marker doing the work of a threshold.
Filter bar One dropdown, full width, showing its selected value rather than a count. Everything below is scoped to it.

Tableau Public · Energy flows in the Regional scenario September 11, 2026 Tableau Public embed view; workbook published by Chia Yu Lin medium · light · desktop-web Toggle hotspots

A Sankey doing the job it was invented for, published by an analyst rather than designed by Tableau. The entry argues that the form comes from engineering, where what enters a node leaves it, and that conservation is what makes band width readable as quantity—and that product analytics breaks it because people leave and nothing accounts for them. Here conservation holds and is stated: primary supply 546 TWh at the bottom left, final demand 476 TWh at the bottom right, and the missing 70 TWh has its own destination node called Conversion losses. Nothing disappears off the edge of the diagram. Colour is doing identity rather than status—blue for electricity, green for hydrogen, teal for biomass, orange for heat—and it stays consistent across all three columns, so a carrier can be traced from supply to end use without a legend. The one oddity is that the author has exposed the layout parameters as live controls, including a squish ratio printed to nine decimal places.

Sankey and path Three columns of nodes, link width as volume, and crossings kept to the few places where a carrier genuinely switches rank.
Sankey and path Conversion losses as an explicit destination. This is the node product analytics leaves out, and the reason its widths stop adding up.
Ratio and rate 546 TWh in, 476 TWh out, both stated. The diagram's own arithmetic is checkable from the page.
Categorical series palette Five carriers, five hues, held constant across every column so a band can be followed end to end.
Dashboard builder Curve type, whitespace and squish ratio exposed as reader-facing controls. The last one reads 0.484057971.

Tableau Public · User Funnel & Conversion Rates September 11, 2026 Tableau Public embed view; workbook published by Tetiana Berezhna medium · light · desktop-web Toggle hotspots

Another published workbook rather than anything Tableau designed, and it makes both of the mistakes the funnel entry names. It shows one number per step, not two. The labels down the right read CVR 100%, 74%, 55%, 37%, 23%, 7%, and every one of those is overall conversion—the share of the original 8,460 still present. Step conversion is missing, which matters because the worst step in this funnel is invisible: 1,948 people start a trial and 577 pay, so that step converts at 30%, and nothing on the chart says 30. You have to divide two numbers printed four inches apart. The second is the rendering. It is drawn as a taper, so quantity is encoded as the width of a trapezoid, and the eye compares areas rather than lengths. Plain horizontal bars would have been easier to read and easier to label. The panel on the left has a third problem: a 20% conversion rate sits above a bar of 10 registrations, next to a 26% rate over 3,673, at the same size and in the same grey.

Funnel Drawn as a taper, so the count is the width of a trapezoid. Length would have read more accurately for nothing.
Funnel Overall conversion only. The trial-to-payment step converts at 30% and no number on the chart says so.
Ratio and rate A 20% rate over 10 registrations drawn the same size as a 26% rate over 3,673.
Header KPI strip Three counts, boxed and centred, with no delta and no base. The first and last are the funnel's own endpoints.
Tabs as genres Seven worksheet tabs across the top, named after the data rather than the question—Users : Registration, CVR to Start of Trial.

Grafana · Examples / Dashboard Variables September 10, 2026 Grafana Play (signed out; no version string exposed) medium · dark · desktop-web Toggle hotspots

Six selectors across the top and the page is unusually honest about what they cost. The middle panel says there is a hidden variable called bestfood, currently set to pizza, which has no control anywhere on screen—state affecting the page that a viewer cannot see or change. The overview panel says that setting Instance and then changing the Prometheus source will clear your selection, because the two variables are chained and the second one invalidates the first. Both of those are correct behaviour and both are the reason parameterised pages get distrusted. What the top row gets right is the display: Instance and CPU Usage Type show their selected values as chips with an × rather than as a count, so what is filtering the page is legible without opening anything.

Template variable Six selectors bound to every panel's query, with the chosen values carried in the URL.
Filter bar Values as removable chips rather than a dropdown saying Instance (3).
Explain this metric Names a hidden variable set to pizza. Prose is doing the work a control should.
Categorical series palette Twelve-plus hostnames differing only in the middle digits, with twelve-plus colours to match.
Data table Four columns visible, the fourth cut mid-word, with no horizontal scroll offered.

Grafana · Examples / Gauge September 10, 2026 Grafana Play (signed out; no version string exposed) medium · dark · desktop-web Toggle hotspots

Nine gauges, and the page makes Few's argument against them without meaning to. The panel at bottom left holds seven arcs reading 30.6, 30.8, 30.8, 30.9, 30.9, 31.0 and 31.3 GB. Seven squares, seven needles, to say that seven disks are all about the same. The same seven values as a bullet graph would be seven short bars in a fraction of the height, and the one that was different would be obvious rather than requiring you to read each label. What the gauges do carry is a bounded range where the endpoints mean something, which is the case the pattern survives: these are percentages and capacities, not rates. The panel beside them is where it stops working—three circular gauges at 40.8, 20 and 48.6 GB, all with a red arc, so a value and another value two and a half times larger get the same verdict.

Gauge and dial Seven arcs to report that seven disks agree. This is the area argument in one panel.
Sparkline A sparkline inside the arc: the same series encoded twice in one square.
Sequential and diverging scales The arc runs pink through green with no perceptual order, so position on it means nothing without the number.
Semantic status color 40.8, 20 and 48.6 all arc red. The threshold fires for everything, so it says nothing.
Threshold line and region The coloured band around the rim is the threshold, drawn as a region rather than a line.

Grafana · Examples / Geomap September 10, 2026 Grafana Play (signed out; no version string exposed) medium · dark · desktop-web Toggle hotspots

Four maps of the same flight data, and the base map does more damage than any of the data settings. On the three dark panels the markers are small green dots on dark grey, which at this size is close to the contrast floor; on the satellite panel they are small green dots on terrain that is already green, brown and blue, and they effectively disappear. The base layer is supposed to be context and recede, and imagery is the loudest option available. The density layer along the bottom is the honest one: it stops pretending individual points resolve, and the clusters on both coasts read immediately. Its radius is doing a great deal of unlabelled work, though, and the blobs merge into clouds at a setting nobody can see. Three of the four panels also carry a legend that says "Layer 1", which is a legend occupying a corner and naming nothing.

Geo map with markers Markers as small dots. Green on dark grey at four pixels is near the contrast floor.
Geo map with markers Satellite imagery under the same markers, competing with the data instead of receding.
Sequential and diverging scales Density, with a ramp whose radius parameter changes the apparent story and isn't shown.
Legend and series toggle A legend reading 'Layer 1'. It holds a corner of the panel and identifies nothing.

Grafana · Examples / Histogram September 10, 2026 Grafana Play (signed out; no version string exposed) medium · dark · desktop-web Toggle hotspots

This page accidentally contains the cleanest demonstration of the bucket problem I have found in a shipped product. The panel top right is the source series, a line oscillating between 24 and 32. Directly under it, the same series on automatic buckets: roughly twenty-four bars, and it is plainly bimodal, with one hump around 27 and a taller one at 30.4. To the left of that, labelled "Force timeseries into bucket size=3", is the identical data at a bucket width of three. Four bars. One hump. The second mode has not been smoothed or de-emphasised, it is simply gone, and nothing about the chart indicates that a decision was made. Neither panel is wrong. One of them answers "is this one population or two" and the other cannot, and the only difference between them is a number in a field.

Time series The source series, before any bucketing. Two modes are not visible here either.
Histogram and distribution Automatic buckets on that series: clearly bimodal, humps at roughly 27 and 30.4.
Histogram and distribution The same data at bucket size 3. Four bars, one hump, and no sign that anything was lost.
Histogram and distribution A different series at the default count, showing a long left shoulder and a small second mode at 50.5.
Sequential and diverging scales A gradient fill down each bar, encoding nothing. The height already carried the count.

Grafana · Examples / Stats, light theme September 10, 2026 Grafana Play (signed out; ?theme=light) dense · light · desktop-web Toggle hotspots

The same dashboard as the dark Stats capture, same data, same palette, one URL parameter apart. Put them side by side and the palette turns out to have been designed on charcoal. On the dark version the value colours carry because they are brighter than the ground—reds, oranges and greens against near-black. Here they have to carry by being darker than the ground instead, and the same hues arrive as pale pink and pale salmon. Look at the Color value panel: six numbers at 30-odd pixels, and 93.5, 88.7, 93.4, 81.6, 77.6 and 87.9 are all light pink on white. The column on the right is worse, because 63.8 GB lands on pale orange. Nothing about the theme switch is broken, and nothing was chosen for this ground. That's the whole argument for designing the dark case first and deriving the other one: the constrained case should set the lightness range, and here the unconstrained one did.

Dark-first data palette The same six values as the dark capture. Bright-on-charcoal became pale-on-white, and the contrast went with it.
Dark-first data palette 63.8 GB in pale orange on white. On the dark version the identical colour was the readable one.
Semantic status color Whole-tile fills survive the switch, because the text sits on the colour rather than being the colour.
Grayscale with alerts Ninety green rectangles, identical in both themes. Nothing here can read as abnormal on either ground.
Sparkline The area fills under each value drop to near-white here, so the shape that was legible on charcoal is a smudge.

Grafana · Examples / Stats September 10, 2026 Grafana Play (signed out; no version string exposed) dense · dark · desktop-web Toggle hotspots

A showcase rather than a working dashboard, which is what makes it useful: it is Grafana demonstrating every visual option the stat panel has, in one place, with nothing else competing. Two things are worth reading it for. The first is that across roughly sixty tiles here, not one carries a comparison. Every option on display is about how the number looks—background colour, value colour, orientation, text mode, grid packing—and none is about what the number should be measured against. The panel type that most needs a delta and a base is being shown off without either. The second is the colour. Every single tile on this page is coloured, and the "No text" panel is ninety-odd green rectangles carrying no label and no value at all, which is colour as the sole carrier of meaning at its purest. Both of those are reasonable for a catalogue of options and neither survives being copied onto a real page.

KPI tile Label, value, unit and a sparkline behind it. No delta, no base, no window.
Dark-first data palette The dark half of the pair. The same six values in the light capture arrive as pale pink on white.
Sparkline A column of them, each scaled to its own series, which is the comparison Tufte warns the arrangement invites.
Semantic status color Whole-tile background colour: the loudest channel available, spent on all six.
Grayscale with alerts Seven rows, seven different colours, no resting state. Nothing here can read as abnormal.
KPI tile Thirty numbers with no labels—the value with nothing attached to it.
Semantic status color No text at all. Colour is the only carrier, which is the WCAG failure stated as a feature.
Header KPI strip Six tiles sharing one anatomy, which is the arrangement that reads as a row rather than as six things.

Grafana · Examples / Table September 10, 2026 Grafana Play (signed out; no version string exposed) dense · dark · desktop-web Toggle hotspots

Seven tables demonstrating what a cell can hold, and the range is the point: a threshold colour on the whole cell, a continuous gradient down a column, an image, a bar rendered inside the cell, a pill, and a markdown block carrying three metrics at once. Two of them stop working at this size. The bar gauge column on the right is squeezed to a sliver about six pixels wide, so the encoding that was supposed to make magnitude visible carries nothing, and the catch-phrase column beside it truncates mid-word with no way to see the rest. The bottom-left table is the one to look at hardest. It is sorted, it has column headers you can click, and the pager under it reads page 1 of 72. Any sort applied there is a sort of the page unless the query re-runs, and nothing on screen tells you which of those is happening.

Data table Threshold colour filling the cell, which makes a column scannable without being read.
Sequential and diverging scales A continuous ramp down a sorted column: the table doing a heatmap's job.
Data table In-cell bar gauge clipped to a sliver, and text truncated mid-word with no escape to the full value.
Semantic status color Value mappings turning raw levels into a closed set, each with its word in the cell.
Delta indicator Pills reading up, down and down fast: direction with no magnitude and no base. Also page 1 of 72.

Honeycomb · Query / HEATMAP(duration_ms) September 10, 2026 Honeycomb sandbox, public dataset (signed out; no version string exposed) medium · light · desktop-web Toggle hotspots

My heatmap entry cites Honeycomb as the argument for log-scale y-buckets, so it's worth recording that this is Honeycomb's own sandbox rendering the same chart on a linear axis. The result is the failure the argument warns about: the ticks run 0 to 3500 evenly, the dense band where almost every request actually lives is squashed into the bottom sixth of the panel, and the top half is mostly empty. What survives anyway is the thing a percentile can't tell you. The solid band under a second doesn't move across the whole window, while from about 07:00 a separate purple tail climbs to 3000ms and keeps going. Two populations, one of them fine and one of them deteriorating. A p95 line over this data would have risen and said nothing about which. The other thing worth stealing: the footer reports elapsed query time and that it examined 11,710,335 rows, so the panel tells you what it cost and how much it looked at. Cookie banner and a no-signup onboarding modal were removed to take the shot; nothing of the product's own UI was.

Heatmap Linear y-axis, so the band holding most requests is six percent of the panel height and the empty top half gets the rest.
Dashboard builder No panel to configure—the query is the page. WHERE trace.parent_id does-not-exist is how you say root spans only.
Tabs as genres Five readings of one result: overview, BubbleUp, correlations, traces, raw. Each tab is a different question, not a different subject.
Search across panels The schema is the navigation, and it opens with a filter box rather than a tree you're expected to browse.
Time-range picker Absolute, with the granularity stated beside it, and arrows that step to the previous window rather than retyping it.
Data source badge Elapsed query time and 11,710,335 rows examined. The panel reporting its own cost and scope, which almost nothing else does.
Freshness indicator When data last arrived, not when the page last ran. The one of the three ages that actually matters.

Kibana · Dashboards / list September 10, 2026 Elastic demo environment (guest session) medium · light · desktop-web Toggle hotspots

The estate, rather than a dashboard. Twenty rows a page and fourteen pages, so something close to 280 dashboards, and sixteen of the twenty on this first page are called "[Metrics Kubernetes] something"—Cronjobs, StatefulSets, Volumes, Pods, Deployments, Proxy, DaemonSets, Jobs, Nodes. They were generated by an integration rather than designed, they all landed on the same day, and they will sit in this list forever. That's the shape every mature dashboard estate takes, and it's why search stops being a convenience at this scale: nobody is browsing to page nine. The list does two things well. Each row carries a one-line description under the title, which is the difference between a name and an answer. And the bracket prefix is doing the work a folder would, so the generated ones sort together and stay out of the way of the four a person actually made.

Multi-page dashboard A set of peers with no landing page. Fourteen pages of them, and nothing says which is the one the team uses.
Search across panels Past about fifty dashboards this is the navigation and the list below it is a filing system nobody browses.
Semantic grouping Tags and a bracket prefix standing in for folders, which is what keeps 250 generated pages out of the way.
Data table Name, description, last updated, actions. The description column is what makes this a list you can read rather than scan.

Kibana · Discover / filebeat logs September 10, 2026 Elastic demo environment (guest session; no version string exposed) dense · light · desktop-web Toggle hotspots

Two panes: 315 fields down the left, 13,637 documents on the right, and a volume histogram over both. The left pane is the good half—it opens with a search box rather than a tree, which is the only sane way to navigate that many fields. The right pane is where it falls over. The documents table ships with two columns, a timestamp and "Summary", and Summary is every field on the record concatenated into one cell: agent.ephemeral_id, agent.id, agent.name, agent.type, agent.version, cloud.account.id, cloud.availability_zone, and on for three wrapped lines per row. It is technically complete and it cannot be scanned, so the first thing anyone does here is pick columns—which is to say the default view's job is to make you configure it. Underneath, the pager reads 100 rows per page across 137 pages, and the sort control sits above a table showing the first of them.

Two-pane list and detail Fields on the left, records on the right, both on screen at once. The oldest working shape for a queue there is.
Detail on demand An expander on every row. Inline rather than a side panel, so it pushes the rest of the list down.
Data table One column called Summary holding every field on the record. Complete, and unreadable at a glance.
Search across panels 315 fields, so the sidebar opens with a filter box. Past a certain count a tree is a filing system nobody browses.
Filter bar A query language in the bar rather than chips. Powerful, and it hides what's applied from anyone who didn't type it.
Time-range picker Relative by default, with the refresh control beside it rather than buried in a settings menu.
Histogram and distribution Volume over the window at a 30-second auto interval. The final bar is the bucket still filling, and it always reads low.

Kibana · Dashboards / [Flights] Global Flight Dashboard September 10, 2026 Elastic demo environment, sample flight data (guest session) dense · light · desktop-web Toggle hotspots

Two things on this page are worth arguing with. The first is the table on the right, sorted by delay rate: Chicago/Rockford 100%, Syracuse 100%, Birmingham 75%. A hundred percent of flights delayed is either a catastrophe or one flight, and nothing in the table says which, because the denominator isn't a column. The cells are on a red ramp, so the two rows that are almost certainly a sample of one are the loudest thing in the panel. The second is the tile row: Delayed 25.2%, then beside it "Delayed vs 1 week earlier—76.9%". Seventy-six point nine percent of what? It could be last week's rate, it could be this week as a proportion of last week, it could be the change. Three different numbers, one label, and the tile picks whichever the query returned. What the page gets right is the filtering: a KQL bar for people who know the syntax and three declared controls underneath for people who don't, both visible at once.

Ratio and rate Delay rates up to 100% with no denominator anywhere. One flight and a thousand flights render identically.
Compare periods "vs 1 week earlier, 76.9%"—the comparison base is named and the expression isn't.
Share and embed Share, export and full-screen in the header. Whether the range and filters travel with them is the whole question.
Filter bar Declared controls under the query bar: two pickers and a price range. Both mechanisms on screen at once.
Panel grid Twelve columns, and the biggest panel is a table rather than the headline chart. Size isn't carrying priority here.
Stacked composition Stacked to 100%, so the total is discarded on purpose and only the mix of delay types remains.
Annotation Event markers along the top of the series, numbered and grouped, on the data's own axis.
Header KPI strip Five tiles in three different sizes and two different layouts, so the row reads as five things.

Tableau Public · Shopify Customer Journey September 10, 2026 Tableau Public embed view; workbook published by Lovelytics medium · light · desktop-web Toggle hotspots

Tableau Public is the product; the design decisions here are the author's. This workbook was published by Lovelytics, so read it as what a competent analyst builds in Tableau rather than as how Tableau thinks dashboards should look. What is instructive is that it carries three separate lines of small-caps instruction—"click on metric to filter dashboard", "hover on a province to view breakdown by top 10 cities", "click on bar to view the second product purchased". Every interaction on the page needed a label, because none of them announces itself. That is the honest cost of cross-filtering: it is powerful and it is invisible until someone tells you it is there. Two other things. The tile block is nine values in a three-by-three grid, which is past the point where a strip has a reading order—the eye has to be told where to start and isn't. And the chart titled "Total sales per month" is plotting seven days, on a y-axis that begins at 500K, so a roughly twenty-five percent spread draws as a mountain range.

Cross-filter The tiles are the filter. It needed a line of instruction above it, because nothing about a number says it is clickable.
Header KPI strip Nine values in a grid rather than four to six in a row, so there is no privileged place for the eye to start.
Time series Titled per month, plotting seven days, on an axis starting at 500K. A 25% spread rendered as a cliff.
Hover detail The breakdown by city exists only on hover, so it is unavailable on touch and invisible in this screenshot.
Geo map with markers A choropleth of raw sales, unnormalised, so California and Texas lead partly by being large and populous.
Ranked list 490 against 30 for second place, so every bar below the first is a sliver and the ordering is all you get.

Grafana · Grafana Heatmaps September 9, 2026 Grafana Play (signed out; no version string exposed) dense · dark · desktop-web Toggle hotspots

Grafana's own teaching dashboard for the heatmap panel, and it teaches well because it holds the data constant and varies exactly one thing. The top-left panel is the source series. The eight heatmaps are all those same numbers under different y-bucket scales: linear, log2, log10, each with and without a split, two of them clamped to a 700-15k range. Read across the grid and the argument makes itself. The linear version crushes almost everything into a band near the bottom of the axis; the log versions spread the same distribution across the full height and the dense region moves. Same data, different readings, and nothing on any single panel tells you which bucketing you're looking at except the title someone remembered to write.

Time series The source series. Every heatmap on the page is built from this.
Heatmap Linear y buckets, which is the default and the one that hides the low end.
Small multiples Nine panels, one variable changed. The comparison is the content.
Sequential and diverging scales Every panel carries its own ramp legend, 1 to 30, so the colour is readable per panel but not across them.

Analytical

Patterns seen in this genre 48

Captures