> For the complete documentation index, see [llms.txt](https://docs.beye.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.beye.ai/release-notes/october-6-2026.md).

# October 6, 2026

### October 6, 2026

Dashboards built by an analysis can now explain themselves, in written commentary and in a note on every figure. PDF reports have a new layout, and questions about large dashboards no longer fail.

#### New

**Written commentary on dashboards**

A dashboard produced by an analysis can now include written sections alongside its figures, not just charts and tables.

The analysis can write the commentary itself, describing what the numbers show in prose, with the figures formatted the way a reader expects rather than as raw values.

**How each figure was calculated**

Every figure on a dashboard built by an analysis now carries a short note explaining how it was worked out.

Hover a figure to read it. Notes are written as the dashboard is built, so the explanation comes from the analysis that produced the number rather than being added afterwards.

Notes can include formatting and mathematical notation, so a formula reads as a formula.

#### Improved

**Redesigned PDF reports**

PDF reports have a new layout built around a wide margin that carries supporting detail beside each section, rather than interrupting the main column.

Each section can carry its own short aside: a heading and a few bullets sitting alongside the content it relates to.

Reports open on a branded title area, and pages after the first carry a running header.

Written sections can now use formatting such as headings, lists, and emphasis. Tables can display money and percentage columns in the same formats used elsewhere in the product, instead of raw numbers.

**Clearer figure titles and spacing**

Figure titles across dashboards now use one consistent style, and charts and tables from an analysis sit on the same rounded card as everything else.

#### Fixed

**Questions about large dashboards could fail**

Previously, asking the assistant about a dashboard with a large number of figures could fail outright. Reading the dashboard pulled in the full definition of every card on it, which on a large dashboard was enough to exceed the limit on how much the assistant can consider at once, particularly if it needed to look twice while working.

The assistant now reads the overall structure first and pulls in full detail only for the figures it actually needs.

**Unhelpful message when a conversation grew too long**

Previously, when a conversation grew past what the model could consider at once, the reply failed with a technical message reading "status 400" and an invitation to try again. Retrying sent the same oversized conversation and failed the same way, so the advice could not work.

The message now explains that the conversation has grown too long and suggests starting a new one. It appears in all supported languages.

**Reasons for a failed response were hard to find**

Previously, when a response failed partway through, the reason was only visible by hovering the small model label beneath it, and a response that failed before producing anything showed only a bare red line.

A failed response now ends with a card stating that the response could not be generated, along with the reason.