> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tryrekkal.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Reading a change

> Why a number moving is not always a change, what each verdict on the card means, and why a gap in a chart stays a gap.

Every card compares a window — 1, 4 or 12 weeks of daily slots — against the equal-length window
immediately before it. What it says about that comparison is one of four things, and only
one of them comes with an arrow.

## A number moving is not a change

Suppose you measured 18% one month and 24% the next. Six points up. Is that a result?

It depends entirely on how many answers each figure came from. From 25 answers a side, a
six-point gap is well inside what re-running the same prompts on the same brand would
produce by chance. From 2,000 answers a side, the same six points are a real movement.

The number alone cannot tell you which. rekkal decides it for you, with an explicit rule:

<Info>
  **The rule:** rekkal shows a direction only when the two periods' confidence intervals do
  **not** overlap. If they overlap at all, the movement is reported without a direction, an
  arrow, or a colour.
</Info>

That is intentionally conservative. It means rekkal will sometimes stay quiet about a change
that was real — and it means an arrow on your dashboard is worth something, because it was
never given away to noise.

## The four verdicts

<AccordionGroup>
  <Accordion title="+3.2 pts / −1.8 pts — a measurable movement" icon="arrow-trend-up">
    The two periods' confidence intervals do not overlap, so this movement is measurable.
    This is the only verdict that carries a direction and a colour.
  </Accordion>

  <Accordion title="No measurable change" icon="equals">
    The point estimate moved — the card tells you by how much — but the two periods'
    intervals overlap, so this sample cannot tell the difference from noise.

    Read this as *"we do not have enough evidence yet"*, not as *"nothing happened"*. More
    weeks, or more prompts, narrow the intervals and may resolve it either way.
  </Accordion>

  <Accordion title="No comparison" icon="circle-half-stroke">
    One half of the comparison is missing, and the card names which half:

    * **No prior period to compare against** — the previous window recorded nothing to
      measure over.
    * **Nothing measured in this period yet** — so there is no change to report.
    * **The prior period holds only *n* runs, below this measure's threshold** — too small
      to compare against.

    Three different situations with three different next steps, which is why they get three
    different sentences rather than one shrug.
  </Accordion>

  <Accordion title="Change not testable" icon="ban">
    Only average position gets this. It is a mean, and rekkal's stored data carries no
    interval for a mean, so there is nothing to test the two periods against each other
    with. Both periods are shown side by side instead. See
    [What rekkal measures](/metrics/what-we-measure#average-position).
  </Accordion>
</AccordionGroup>

## Gaps stay gaps

If nothing ran in a given week — an outage, an engine failure, a workspace paused — that
week has **no point on the chart**. The line stops and restarts on the other side, the week
keeps its place on the axis with a dashed marker, and the missing week is named underneath.

Nothing is interpolated across it, because a smooth line through a week that was never
measured is a claim rekkal cannot support. Most charting libraries join across a missing
point by default; rekkal's charts are drawn by hand specifically so they cannot.

<Note>
  **A measured zero is different from a gap, and is drawn differently.** If a week *did* run
  and no answer named your brand, that is a real 0% against a real denominator, and it is
  plotted as zero. Absent data is absent; observed absence is zero. The dashboard never
  confuses the two.
</Note>

## Empty states say why they are empty

"No data" is not an answer. Wherever a surface has nothing to show, it names the cause,
because the causes have different remedies:

* **Nothing has been measured yet** — set up your brand, or wait for the first slot.
* **Answers were collected, but their sources were never captured** — the numbers on the
  dashboard for that window are real; only the sources are missing.
* **Sourced answers were read and cited nothing** — a real result about your category.
* **No competitors are tracked yet** — so there is nothing to be missing from.
* **Your plan does not include this view** — a plan boundary, which is shown as its own
  screen naming the plan and the remedy, never as an empty chart. See [Plans](/plans).

## What to do with a "No measurable change"

<Steps>
  <Step title="Check the sample size">
    If the `n` under the figure is below the reporting threshold, the honest answer is that
    you do not have enough evidence yet.
  </Step>

  <Step title="Widen the window">
    Switch from 1 week to 4 or 12. More slots means more runs on both sides of the
    comparison and narrower intervals.
  </Step>

  <Step title="Upgrade for a larger prompt allowance">
    Prompts are the fastest lever on sample size: doubling your prompts doubles the runs per
    week, and Pro also covers three engines rather than one. Your prompt set is chosen during
    setup and cannot be edited afterwards yet, so a higher allowance is how you widen it
    today. See [Plans and limits](/plans).
  </Step>

  <Step title="Wait">
    Daily cadence does not lower the bar, it shortens the wait. A single day on this kind of
    data is mostly a report on sampling noise; a week of daily slots is seven times the
    sample a weekly cadence would have given you by then.
  </Step>
</Steps>