Where we are
The instrument already has five layers: the decision a screen serves, the contract behind its data, the platform it runs on, the promise it makes about freshness. Tonight adds the layer most reviewers reach for first and, on its own, trust too much - the rendered surface. Six defects belong to this session alone, more than to any other, because a screen can be perfectly decided, contracted, platformed and promised, and still lie to the eye on every load.
★ What you walk out with tonight. A lie factor you compute from axis bounds and bar heights, not one you eyeball. Few's 13 pitfalls run as a real audit list against four real charts. And the measured proof that a seven-rule reporting standard, applied after Tufte and Few are already switched on, catches nothing new.
LiveAn audit question, not a redesign brief3 min▶
A reviewer signing off a dashboard is not being asked to make it prettier. They are being asked one narrower question: does this screen represent the data accurately, and would a reasonable reader draw the right conclusion from it. That question has a small, checkable set of answers - an axis either starts where the data says it should or it does not, a colour either carries one meaning everywhere or it does not - and this session trains exactly that check.
Choosing which chart type fits which comparison, decluttering a busy panel, or laying out a dashboard so the eye lands in the right order are real skills with their own depth. They are not what a sign-off checks. A reviewer who tries to redesign every screen they audit will get invited to fewer reviews.
Tideline's retention chart passed three design reviews. Nobody in any of them measured the axis. All three reviewers said the chart "looked fine" - which is a design judgement, not an integrity check, and it is exactly the judgement that let a 6.5-to-1 overstatement ship three times.
Self-studyWhere this goes deeper3 min▶
Two sibling courses own the ground this session deliberately leaves alone, and are worth the follow-up once the audit habit is in place:
- Chart choice and encoding craft - which mark and channel fit which comparison, and how to build the muscle for picking well the first time: learn-dataviz-with-phoebe.
- Dashboard build and layout - panel arrangement, information hierarchy, and the tooling choices behind a screen that scans well: learn-business-intelligence-with-phoebe.
This session's job is narrower and comes first: before a screen is redesigned, someone has to be able to say precisely what is wrong with it, and prove the claim with a computable number.
The lie factor is arithmetic, not an opinion
6 min liveEdward Tufte's three books never published graphical integrity as a numbered list - that is worth saying plainly before teaching it, because every secondary source that claims a fixed count is quoting a list Tufte never wrote. What he did leave behind is a small set of concepts precise enough to compute, and one of them turns "this chart feels misleading" into a number.
Honesty check. These concepts come from The Visual Display of Quantitative Information (1983), Envisioning Information (1990) and Visual Explanations (1997). Tufte never published them as a canonical numbered list, which is why every secondary source gives a different count. Treat what follows as the concepts themselves, not a list with a fixed length.
LiveFive concepts worth carrying4 min▶
- Data-ink ratio - the share of ink on a chart that encodes actual data, versus grids, borders and decoration that do not.
- Chartjunk - visual elements that add neither information nor clarity: gradients, drop shadows, decorative icons standing in for data.
- Small multiples - the same chart form repeated across a series, so comparison happens by eye movement rather than by memory.
- The lie factor - the size of the effect shown in the graphic, divided by the size of the effect in the data. An honest chart scores close to 1.
- Graphical integrity - the umbrella claim that a representation of numbers should be at least as faithful to those numbers as the numbers themselves are.
Of the five, the lie factor is the one you can compute on the spot with nothing but a ruler and the axis labels, which is why it carries the rest of this part.
LiveComputing 6.5 to 1 from the rendered bars4 min▶
Tideline's retention chart sets its axis from 40 to 54 rather than 0 to 54. Real retention moved from 47.1 percent in March to 52.0 percent in August - a genuine 4.9-point, 10.4 percent gain. But because the axis starts at 40, the bar for March renders at 46 pixels tall and the bar for August at 77 pixels tall, inside a 90-pixel plot area.
The lie factor is the shown effect over the real effect:
- Shown - (77 minus 46) over 46 equals 67 percent apparent growth
- Real - (52.0 minus 47.1) over 47.1 equals 10.4 percent actual growth
- Lie factor - 67 divided by 10.4 equals about 6.5 to 1
That is the exact number the bench in session 9 computes from this chart's rendered geometry, not a rounded estimate. Anyone with the axis bounds and a ruler can reproduce it in under a minute, which is what makes the lie factor the most defensible finding you can bring into a review meeting.
The fix costs nothing to argue for once the number exists: move the axis minimum to zero, or add a break mark if the chart genuinely needs to show a narrow band. Either way, the lie factor drops toward 1 and the conversation moves from "it looks steep" to "it now matches the data".
Few's 13 is the one list built for exactly this
6 min liveStephen Few's white paper is, of the whole canon, the single most directly applicable list to a dashboard sign-off. It was written from Few's own consulting practice, published by Perceptual Edge in February 2006, and folded into Information Dashboard Design (O'Reilly, 2006) the same year. Unlike Tufte or Norman, Few numbered his list himself. All thirteen belong in an audit, in order.
| # | Pitfall | Fires on Tideline |
|---|---|---|
| 1 | Exceeding the boundaries of a single screen | Yes - D10, 64px below the fold |
| 2 | Supplying inadequate context for the data | No, on this screen |
| 3 | Displaying excessive detail or precision | Yes - D2, MRR to four decimal places |
| 4 | Expressing measures indirectly | Yes - D6, a count where the rate is the measure |
| 5 | Choosing inappropriate media of display | Yes - D5, a gauge for one number |
| 6 | Introducing meaningless variety | No, on this screen |
| 7 | Using poorly designed display media | No, on this screen |
| 8 | Encoding quantitative data inaccurately | Yes - D4, the truncated axis, lie factor 6.5 to 1 |
| 9 | Arranging the data poorly | No, on this screen |
| 10 | Ineffectively highlighting what's important | No, on this screen |
| 11 | Cluttering the screen with useless decoration | No, on this screen |
| 12 | Misusing or overusing color | Yes - D7, 13 distinct fill colours |
| 13 | Designing an unappealing visual display | No, on this screen |
What the eye reads before it thinks, and one lens that adds nothing
6 min liveTwo more lenses round out the surface layer. One tells you which visual channels a reader decodes accurately without conscious effort. The other tells you why elements on a screen read as one group or several. A third lens, IBCS, gets tested here too - and on Tideline it earns its keep by catching precisely nothing new.
LivePreattentive attributes: length and position win4 min▶
Nielsen Norman Group's dashboard guidance, published 2017-06-18 by Page Laubheimer, is direct about which visual channels carry quantitative meaning accurately and which do not. Length and two-dimensional position are read fast and read correctly. Colour and shape are read fast too, but only reliably as secondary reinforcement - a way to flag which bar matters, never the primary way to convey the bar's value. The guidance follows through into a concrete preference: bar, line and scatter charts over pie, donut, treemap, gauge and 3D.
Tideline's gauge encodes one number using angle and colour together - exactly the combination NN/g's guidance flags as weak. A single labelled number, or a bar against a target line, would let a reader extract the same value without decoding an angle first.
Self-studyGestalt grouping: five mechanics that decide what reads as one thing3 min▶
Honesty check. Max Wertheimer's 1923 paper is the foundational source for these principles and is not practically fetchable today, so what follows travels through modern textbook summaries, which is also why the count varies by source. Treat these as grouping mechanics, not a fixed numbered law.
- Proximity - elements placed close together read as one group. A KPI and its trend line belong touching, not on opposite sides of the panel.
- Similarity - elements sharing colour, shape or size read as related, whether or not that relation is real. Thirteen distinct fills on one screen means the eye tries and fails to group anything.
- Closure - the eye completes a familiar shape from incomplete information, useful for sparklines and gauges alike, and a reason a gauge feels informative even when it is not.
- Continuity - the eye follows a line or alignment across a break. A trend line reads as continuous even across a gap in the data, which is exactly why gaps must be marked, not left implicit.
- Figure-ground - the eye separates a foreground subject from its background. A panel with no visual lead has no figure, so nothing steps forward against the rest - Tideline's identical-weight panel layout is this failure by name.
LiveIBCS: seven rules, zero new findings4 min▶
Honesty check. The IBCS Association's primary standards page renders its content via JavaScript rather than static text, so the letter-by-letter wording below travels through two secondary sources that agree with each other. Treat the acronym as reliable and the exact phrasing as close paraphrase.
IBCS Standards v1.2 compresses reporting discipline into seven letters, SUCCESS: Say - convey a message rather than just show numbers. Unify - use one consistent semantic notation across every report. Condense - maximise information density without excess. Check - verify visual and mathematical integrity. Express - choose the proper visualisation type for the message. Simplify - remove anything that does not aid understanding. Structure - organise content logically.
Run against Tideline on top of Tufte and Few, IBCS moves the catch count from 11 to 11. Every violation IBCS would flag - the truncated axis under Check, the colour overuse under Simplify, the wrong chart type under Express - was already caught by Tufte's lie factor or Few's pitfalls before IBCS was ever switched on.
That is not a mark against IBCS, which is a serious, widely adopted standard across European enterprise finance reporting. It is what overlap looks like in practice: three well-regarded lenses converging on the same twenty pixels because they are all, in the end, describing the same honest chart from different vocabularies.
Audit Tideline's four charts
★ 16 min · everyone buildsWork through Tideline's revenue panel, retention chart, health gauge and signups line in order. Use only a straightedge, the axis labels already on the screen, and the tables above - no tooling required for tonight's version.
Compute the lie factor by hand. Retention moved from 47.1 to 52.0 across the chart, a real 10.4 percent gain. The chart's axis runs 40 to 54. Measure or estimate the bar heights for the first and last bars as drawn, take (tall minus short) over short for the shown effect, divide by 10.4 percent for the real effect, and confirm you land near 6.5 to 1.
Run Few's 13 against all four charts. Go pitfall by pitfall down the table in Part 2 and mark yes or no for the revenue panel, the retention chart, the gauge and the signups line. Do not stop at the first hit per chart - a single panel can commit more than one pitfall.
Tally the fires. You should land on six: exceeding the screen, excess precision, indirect measure, wrong media, inaccurate encoding, overusing colour. If you found a seventh or an eighth, write down which pitfall and which chart - it may be a real finding this session's defect list did not enumerate.
Pick the one fix that removes the most reader error. Of the six, which single change would most reduce the chance a reader draws the wrong conclusion from this screen on a normal Monday? Argue for the axis fix or against it - both answers are defensible, but only one of them is defensible with a number attached.
The exercise works on any dashboard, not only Tideline's. The axis-bounds trick and the 13-pitfall pass both transfer directly - what will not transfer is the specific lie factor of 6.5, which is Tideline's number and nobody else's.
Before session 7
20 min- Find one chart in active use on your team where you are not certain the axis starts where it should. Compute its lie factor using the method from Part 1. If it is above roughly 1.5, that is a finding, not a style preference.
- Run Few's 13 against your team's single most-viewed dashboard. Write down every pitfall that fires and which panel it fires on.
- Pick one panel on that dashboard and name, out loud to a colleague, which Gestalt mechanic is failing - proximity, similarity, closure, continuity or figure-ground. If you cannot name one, the panel may not have a grouping problem at all, which is itself worth knowing.
Three questions
Retention moved from 47.1 to 52.0, a real 10.4 percent gain. The truncated axis renders that as a 67 percent apparent rise. What is the lie factor, and what does a value near 1 mean?
67 divided by 10.4 is about 6.5. The lie factor is the shown effect over the real effect, and an honest chart scores close to 1 because what is drawn matches what happened.
Adding IBCS's SUCCESS rules on top of Tufte and Few moves Tideline's catch count from 11 to what, and why?
It stays at 11. That is not a weakness in IBCS, a serious enterprise reporting standard - it is what overlap looks like when three lenses describe the same honest-or-dishonest chart from different vocabularies.
NN/g's dashboard guidance ranks length and 2D position above colour and shape for quantitative encoding. What follows for a gauge showing one number?
NN/g names gauges among the media it advises against for exactly this reason: the primary read is an angle, reinforced by colour, and both are less accurate than length or position for conveying a single quantity.
What this session teaches, and where it came from
Every fact on this page, and its verification tier, is recorded in the course's source map.
Session 6 cheat sheet · pin this
A screen can pass every integrity check in this session and still fail its reader the moment they touch it - a filter that cannot be undone, a query that runs for forty seconds with no state to show for it, a status row that only a sighted, colour-normal reader can parse. Session 7 turns five interaction lenses on Tideline's filter, its long-running query, and the red-and-green row that fails roughly one man in twelve.