Keep a Complex Chart’s Argument When You Add an Accessible Reading Route

The most dangerous alternative description of a chart is the one that is factually correct.

Here is a line that could sit under a real bar chart in a real operations deck: Average turnaround fell from 6.0 days to 5.1 days, a 15% improvement. Every figure checks. The percentage is right. If you were reviewing the text alternative rather than the chart, you would approve it and move on. And the chart's argument — the thing the slide was built to make the committee decide — would be gone.

The problem is not accuracy. It is that a compressed description answers the question what does this chart say while the chart was answering what should we do, and what could make that wrong. Those are different jobs, and only one of them survives compression by default.

What follows is a method for the second job. The example is invented for the exercise; nothing below reports a test, a client result, or a real dataset.

Find the evidence a summary must not erase

Before you shorten anything, write the chart's claim in full sentences at whatever length it takes. Not the headline. The whole argument, including the part that makes you uncomfortable.

Three questions get you there.

What comparison does the chart actually support? Between periods, regions, cohorts, before-and-after a change. Name both sides.

Which series complicates it? The one that moves the other way, or fails to move at all when the story requires movement.

What uncertainty would change the decision? An unresolved causal question, a category that means something different this period, a segment small enough that its rate is noisy.

Now the example. A fictional company tracked average turnaround — days from job request to completion — for three regions across two periods. The deck argues for extending a dispatch pilot from North to South. Here is the underlying data, which is what the table in the finished deck would carry:

The aggregate improves: 30,000 job-days across 5,000 jobs in period 1, 25,500 across the same 5,000 in period 2. Volume held steady, so nobody can accuse the chart of comparing different-sized things. That is what makes the favorable summary so persuasive, and so incomplete.

The favorable summary is: Average turnaround improved from 6.0 to 5.1 days, a 15% reduction, with total job volume unchanged.

The argument-preserving summary is longer, and every clause earns its place: Average turnaround improved from 6.0 to 5.1 days across a stable 5,000 jobs. North absorbed 80% of that volume, up from 40%, and North's own turnaround improved from 5.0 to 4.5 days. South's turnaround worsened from 7.0 to 9.0 days, on a much smaller base. Roughly 0.7 of the 0.9-day gain comes from the shift in regional mix rather than from work getting faster.

That last sentence is just arithmetic, and it is the part the deck most needs. If period 2's job volumes had carried period 1's turnaround rates, the average would have been 26,500 job-days over 5,000 jobs — 5.3 days. The mix shift alone accounts for 0.7 of the 0.9-day improvement. The remaining 0.2 comes from the rates themselves: North's half-day gain contributes −0.4 days at its new 80% share, South's two-day loss contributes +0.2 days at its 10% share, netting −0.2.

Notice what this does to the decision. It does not destroy the pilot's case. North really did get faster, by half a day, on four times the volume — that is genuine evidence. What it resizes is the effect. A 0.9-day improvement sounds like a company-wide transformation. A 0.2-day improvement concentrated in one region, with another region sliding two days in the wrong direction, is a smaller and more conditional claim. The exception is not decoration on the argument. It is most of the argument.

One thing the chart does not support at all: that the dispatch pilot caused any of this. North's improvement is consistent with the pilot working. It is equally consistent with volume growth letting North specialize, or with the pilot being deployed in North precisely because North was already improving. South's deterioration is consistent with the policy harming it, with only the hardest jobs remaining in South after the easy ones moved, or with a reporting change. The chart shows association. The uncertainty is not a footnote to the decision; it is a condition of it.

Compare overview-plus-detail with a staged explanation

Two structures are usually on the table once you know what must survive.

Overview plus detail puts a short identifier and a brief orienting claim in the primary reading position, then hands the reader a structured account and the table. In practice that means an alternative description of a sentence or two that names the comparison and its qualification, followed by a fuller prose account and the numbers themselves. The reader can stop after the overview and still hold the right conclusion.

A staged explanation introduces the series in an order that builds: North first, then the volume shift, then South. For a mixed audience this can be genuinely better teaching, because the reader meets one idea at a time instead of a dense relational block.

The failure modes are not symmetrical, and that is the point of comparing them.

Overview-plus-detail fails by burial. The overview gets written as the flattering version — turnaround improved 15% — and the qualification sits three paragraphs down or behind a link. A reader who stops early, which is most readers, holds the wrong belief with full confidence that they read the alternative. Every element of the accessible route is present and correct. The route still transfers the wrong thing.

Staged explanation fails by stranding. If North is one slide and South is another, the reader may never hold them in the same moment, and the joint comparison — the thing that changes the decision — never appears anywhere. This is worse than burial, because the reader cannot go looking for a relationship they do not know exists.

Navigation effort, repetition, and the risk of never reaching the exception pull in different directions. Overview-plus-detail costs one extra navigation step for the full account. Staging costs repetition if you restore the comparison at each stage, and risks loss if you do not.

The rule that resolves it: a staged explanation must restore the joint comparison before it ends. Introducing series one at a time is a sequence, not an argument. The final stage has to place North, South, and the aggregate together, with the volumes visible, or the route has simply distributed the omission across more pages.

For this deck, overview-plus-detail is the structure I would choose, for a reason specific to the decision. The committee is being asked to extend a pilot, which means they need the size of the effect, and size is a relationship between the mix shift and the rate changes. A staged sequence would make them hold three partially-explained pictures in working memory to get there. If the audience instead needed to understand the causal chain of a process change, staging would likely win, provided the last stage returned to the combined view.

Neither route is sufficient by default. Both are choices with costs, and the costs differ.

Keep the same meaning across every route

The accessible route is a second rendering of one dataset, and every divergence between the renderings is a new error introduced by the accessibility work. Check the following against a single source.

Units. Turnaround measured in calendar days or business days? The chart's axis and the table's column header must say the same thing.

Periods. If the chart labels the x-axis "Q1" and "Q2" and the table says "first half," someone has just invented a discrepancy.

Definitions. Does turnaround mean request-to-completion or request-to-first-contact? Does it count jobs that were reopened? Is the reported figure a mean or a median? A useful discipline is to write the definition once and paste the identical sentence into the alternative description, the structured account, and the table caption. Different wording in three places is how meanings drift.

Populations. The table's "All regions" row must be computed from the same three regions the chart plots. If a fourth region was folded into the chart's aggregate but omitted from the table, the table is now misleading in a way the chart was not.

Missing values. A blank cell is not zero. If a region reported no jobs in a period, say so; do not let an empty cell render as 0.0 days, which reads as instant service. If the chart's dashed line indicates an estimate, the table needs a note that says estimate.

Precision you did not compute. Do not add "statistically significant" to a chart whose source never ran a test. Do not add a confidence interval to tidy up an unruly point. The alternative route is not an invitation to make the data more decisive than it was.

There is a subtler consistency problem that tables make worse rather than better. A table grants access to entries. It does not, by itself, deliver an argument. If a reader receives only the table above, they can see that South went from 7.0 to 9.0 and that the aggregate went from 6.0 to 5.1, and they may reasonably conclude that the company's performance is contradictory or that the aggregate is untrustworthy. Both numbers are present. The relationship between them is not. The fix is not to hide the table; it is to state the consequential relationship in prose adjacent to it — the one sentence about mix and rates doing the work. Exposure to the numbers and explanation of their relationship are two different services, and the route needs both.

Test retrieval of the argument, not the presence of text

The question that certifies nothing is: is there a text alternative? The question that matters is: can the main claim and its limitation both be found in the actual delivered format?

That means opening the thing that ships — the slide file, the exported PDF, the page — and working through it the way a reader would, rather than reviewing the source document you wrote.

Inspect the surfaces that break first:

Headings and labels. A link called "Learn more" tells a reader nothing and gives no reason to follow it. "Full figures by region and period" does.

Reading order. On a slide, a text box added after the chart is often announced in insertion order rather than visual order. If your alternative description was added last, a reader may encounter it after the bullet points, the footnote, and the page number — a sequence that no sighted reader ever experiences.

Access to the fuller account. The structured description and the table need to be reachable without a step a reader has no reason to take.

The overview's content. Read only the short description. Does it carry the claim and the limitation? In this example, an overview that says "average turnaround improved 15%, but South worsened, the gain is mostly a volume shift, and the chart does not show that the pilot caused it" lets a reader who stops after one sentence hold the right conclusion. An overview that says "average turnaround improved 15%" does not, no matter how good everything downstream is.

Then compare failures. If the overview-plus-detail route loses the exception at the overview and the staged route loses it at the transition between stages, you have two different problems to fix, not one. Revision should follow the failure: rewrite the overview, or add the comparison stage, depending on which route is actually in use.

This work needs appropriate accessibility expertise and, ideally, readers who use assistive technology, because the behavior of a specific format is not something you can reason your way to from a specification. The W3C Web Accessibility Initiative's tutorial on complex images — guidance scoped to images on the web, which I am treating as a design principle rather than a certificate for any particular format — is useful here for one reason: it asks for an equivalent account of the image's important information, including scales, values, relationships, and trends, rather than a label identifying what kind of chart it is. That principle is what makes the favorable summary disqualifying. It is also, on its own, not evidence that your slide, your table, or your staged sequence delivers an equivalent account. That can only be checked against the artifact.

I have not built this chart, exported this deck, or run this test. The dataset above is invented; the arithmetic is verified against it, and the turnaround figures, the job counts, the ×0.15 reduction, and the 0.7/0.2 mix-and-rate split all reconcile. Everything about how the routes perform in a real presentation file remains unperformed and unverified.

What to choose, and what the first sentence carries

The route I would ship for this case is an overview linked to a prose account and the table, with the overview written to survive on its own.

Its first summary reads roughly: Average turnaround improved from 6.0 to 5.1 days across a stable 5,000 jobs, but the gain is mostly a volume shift toward North, which now carries 80% of work, South's turnaround worsened from 7.0 to 9.0 days, and the chart shows association, not proof that the pilot caused the change.

Four things are in that sentence. The favorable headline, so the reader is not misled about the aggregate. The exception, so the reader knows the improvement is not uniform. The mechanism, so the reader knows how much of the gain survives if the mix stops shifting. The limit of the evidence, so the reader does not treat the improvement as grounds to extend the pilot everywhere without further check. What remains for the fuller account is the per-region rates, the job counts that make the aggregate legible, the mix-versus-rate decomposition, and the competing explanations — specialization at higher volume, or a pilot deployed in North because North was already improving — that keep the causal question open.

If you take one habit from this, take the ordering: write the qualifying version first, at whatever length, and compress from there. Compression applied to a summary that already dropped the exception can only produce a shorter version of the wrong argument. Compression applied to the whole argument produces an overview that happens to be short and still decides correctly — which is what the reader needed all along.

Frequently asked questions

Why is an accurate summary like 'turnaround improved 15%' not enough?

It answers what the chart says while the chart was answering what should we do and what could make that wrong. In the invented data, the aggregate improves from 6.0 to 5.1 days, but North absorbed 80% of volume, South worsened from 7.0 to 9.0 days, roughly 0.7 of the 0.9-day gain comes from the mix shift, and the chart shows association rather than causation. A compressed summary can erase those decision-relevant parts.

What should an accessible reading route preserve?

It should preserve the chart's main claim and its limitation: the comparison, the complicating series, and the uncertainty that would change the decision. Units, periods, definitions, populations, and missing values must stay consistent across the chart, the alternative description, and the table. Do not add statistical significance or confidence intervals the source never computed.

What is the difference between overview-plus-detail and a staged explanation?

Overview plus detail puts a short identifier and orienting claim in the primary reading position, then hands the reader a structured account and the table. It fails by burial if the qualification sits down the page or behind a link. A staged explanation introduces series in order; it fails by stranding if the reader never holds the joint comparison. A staged explanation must restore the joint comparison before it ends. For the pilot decision, overview-plus-detail is chosen because the committee needs the size of the effect.

How should the accessible route be tested?

Test retrieval of the argument in the actual delivered format, not just whether text exists in the source. Open the shipped slide file, PDF, or page; read only the short overview to see whether the claim and its limitation are both there; inspect headings and labels, reading order, and access to the fuller account and table. Reviewing the source document you wrote is not the same as checking what ships.

What are the limits of the example and the guidance?

The dataset is invented, though the arithmetic is verified against it and the figures reconcile. The author has not built the chart, exported the deck, or run the test. The W3C WAI complex images tutorial is scoped to images on the web and is treated here as a design principle, not a certificate for any particular format. The work needs accessibility expertise and, ideally, readers who use assistive technology.

Further reading

• Make a Company Deck Readable Without Flattening the Design — Read next for the deck's own design pass: keeping a qualifier in the same visual unit as its claim, fixing stored reading order, and inspecting the exported file.
• Chart, Table, or Diagram: Choose the Form That Answers the Slide’s Question — Start here when deciding whether the slide's question needs a chart, a table, or a diagram, since the form cannot repair what the records never established.
• Describe a Treatment’s Visual Argument for a Nonvisual Reader — Use when the route must describe a visual argument rather than list what is in the frame; it offers a reconstruction test for nonvisual readers.