Results
How Trakkr measures shipped work against a frozen before-and-after window, and reports wins, misses and gaps without claiming causation.
Finishing work is not the same as earning a result. Results is the last step of one loop: Pages finds the URL worth working on, Actions is where the work is chosen and shipped, and Results says what happened afterwards.
It opens from the foot of the Actions queue, or directly at /actions?view=results. Every row is a piece of shipped work with a frozen measurement plan attached, either still running or already judged.
Measurement happens in the data pipeline, using connected crawler, citation and Search Console evidence. The Agent never grades its own work, and the language stays observational: a change may have Coincided with a drop, never caused one. The same restraint applies to a rise.
The plan is frozen when you start
The measurement plan is frozen the moment you start the work. Nobody moves the goalposts afterwards, including Trakkr.
The plan names the subject, the primary measure, any supporting measures, the window in days, and the bar for an earned result. Results reads that saved plan when the window closes. It does not pick the most flattering measure after seeing the data.
This matters most when numbers move in different directions. A page can gain impressions while citations stay flat. The primary measure still decides, because it was chosen before the work began. Supporting numbers stay visible as context.
Setup is the exception. It unlocks measurement or delivery rather than earning a page-level outcome, so it never opens a window.
The five states
| State | What it means |
|---|---|
| Measuring | The work shipped and its frozen window is still open. The row shows day N of the run. |
| Earned | The measure cleared the fixed bar when the window closed. |
| No movement | There was enough evidence to judge, and the change did not clear the bar. |
| Coincided with a drop | The measure fell far enough, on a large enough baseline, to be worth a decision. |
| Couldn't measure | The window closed without enough valid evidence to judge. |
When a drop is put back through its original publishing path, the row reads Reverted instead.
No movement is shown, not hidden. If Trakkr displayed only wins, every other number here would be worthless. A miss says the work shipped, measurement worked, and the measure did not move enough.
Couldn't measure is equally important. It never becomes No movement, never counts against your earn rate, and never triggers a rollback.
Measurement windows by kind of work
Different work needs different time to settle. These are the defaults; the frozen plan shows the exact window before you start, and each kind has a permitted range around its default.
| Kind of work | Window | Primary measure |
|---|---|---|
| Technical fix | 14 days | AI bot fetches of the page |
| Page improvement | 28 days | Search Console clicks to the page |
| Content refresh | 42 days | New citations of the page |
| New content | 42 days | New citations of the page, with bot fetches as support |
| Outreach draft | 42 days | Citations from the target source |
| Reply draft | 42 days | Whether an AI platform cites the thread |
| Campaign | 70 days | A combined view of citations, fetches and clicks |
| Setup | Not measured | It unlocks a source or delivery path |
| Watch | Not measured | It keeps checking for a change that needs attention |
Supporting measures never extend the window after the work starts.
The earned bar
Trakkr publishes the rule instead of hiding it behind a score. The after value must be at least max(before × 1.25, before + 3).
Both halves matter. A rise from 2 to 3 is 50%, but it is one extra event and too small to call earned. A rise from 100 to 103 adds three, but it is 3% and also misses. From a baseline of 20 the after value must reach 25, because the 25% rule is the higher of the two.
Impressions use an absolute floor of 50 instead of 3, because they are larger and noisier. A softer bar would produce more green rows and make the word Earned worthless.
Reading the table
The table lists both open windows and judged outcomes. Its columns are Work, Page, Before → after, Window, Result and Date, with a Who column available where Agent attribution is on. Filter pills across the top carry live counts for the five states, and a pill only appears once it has rows.
Search sweeps every result, not just the loaded page. Opening a row shows the metric evidence, the work history behind it, and links back to the action and the page record. Where CSV export is allowed, the download carries the metric, before and after values, the source, the window in days and the shipped date.
When Trakkr couldn't measure
Open the row to see the reason. The result stays in the list, and the next step depends on what was missing.
| Reason shown | What to do |
|---|---|
| Crawler tracking is not connected | Connect crawler tracking so Trakkr can see AI bot visits to the page. |
| Search Console is not connected | Connect Search Console to measure page impressions and clicks. |
| The work is not linked to a page | Attach the work to the page it changed. Brand-wide movement is not a substitute. |
| The page had no activity in the window, before or after | Check the right page was attached. If it was, wait for real activity rather than treating zero as a judged miss. |
| No connected source can see this page yet | Connect crawler tracking or Search Console, depending on the measure in the plan. |
| Citation history was not available for this window | Let the next citation snapshot land. Trakkr does not invent a zero while the source is unavailable. |
| The work never recorded a completion date | Correct the completion record so the before and after windows have a fixed boundary. |
| There was no before reading to compare against | Keep the result unjudged. A later reading cannot recreate a baseline that was never saved. |
Where the cause could clear on its own, the row's menu offers Retry measurement. Where the frozen plan itself is broken or incomplete, there is no retry, because rerunning it would produce the same answer. Either way the result stays Couldn't measure. Trakkr does not rewrite it later as a loss, and it does not undo the work because measurement was missing.
When a change coincides with a drop
The state is worded Coincided with a drop because a before-and-after window does not prove causation. Search demand, a platform update, a site incident or another change may have moved at the same time.
Even this cautious state has a high bar. The fall must be at least 25% below the baseline, the fall must clear an absolute floor (5 fetches or clicks, 2 citations, 50 impressions), and the baseline must be at least four times that floor before a drop is judged at all. Trakkr also checks fleet-wide data quality: when bot fetches fall more than 40% across unrelated brands on the same day, that day is treated as suspect rather than as harm.
Whether the change can be put back depends on the exact change and the platform:
| Change | WordPress | Webflow | Shopify | GitHub |
|---|---|---|---|---|
| Meta title or description | Yes | Yes | Yes | Yes, through a pull request |
Open Graph, canonical, llms.txt or robots.txt | Yes | Yes | No | Yes, through a pull request |
| Structured data | Yes | No | No | Yes, through a pull request |
| Alt text | Yes | Yes | No | No |
| Heading structure | No | No | No | Yes only |
A revert also needs a saved original value and must still be inside the 30-day rollback window. When every check passes and the automation is on Handle it with auto-rollback armed, Trakkr reverts it and tells you. On any lower rung the revert is staged in Needs your call as Revert or Keep. Where the platform cannot restore it safely, Results shows the evidence rather than a button that cannot keep its promise.
Earn rate
Earn rate is Earned ÷ (Earned + No movement).
Couldn't measure is excluded, because it says nothing about whether the work helped. Coincided with a drop and Reverted are also kept outside this win-versus-no-movement rate and stay visible as their own outcomes. The count of unmeasured results is always shown beside the rate: 75% with eight unmeasured items is not the same as 75% with none. Metrics has the worked example.
Your quarter
The quarter recap lives at /results/quarter, reachable from the link above the Results table. Older /proof/quarter links redirect there, so reports and emails already sent keep working.
Not everyone sees it. The recap follows the same per-brand rollout as Pages, so a brand outside that rollout is sent back to Results, and white-label client portal users never have it at all.
The recap opens with one claim sentence, the best measured change as a before-and-after chart, and the four judged counts beside the earn rate. Where goals are set for the period, a Goals band compares each one against where it stands now. Misses carry the same weight as wins, because the recap is meant to survive a budget review. It prints to PDF from the header.
Share a result as an image
An Earned result that was not reverted and whose primary measure went up can be copied as a 1200 by 630 image. The card carries the brand name, the claim sentence, the before and after values (or a chart with the shipped date marked), the measured date range and the source.
The ship marker stays visible because two numbers without their time boundary imply more than the data supports. Under a white-label portal the card is measured by the agency's name and never shows the Trakkr mark.
Common questions
Why does this say Coincided with a drop instead of naming a cause?
The measurement compares two windows. It does not isolate every other event that changed at the same time, so a causal claim would be stronger than the evidence.
Why is my window still open?
Results waits for the full frozen window. It does not turn a good first week into Earned or a quiet first week into No movement. The row shows the current day and the total.
What if I did the work outside Trakkr?
You can still track it if the action is linked to the page and you record when it shipped. Trakkr needs that date to set the before and after boundary, and it will not use a brand-wide score as a substitute for page-level evidence.
Can I change what gets measured after starting?
No. That would move the goalposts. If the wrong page or measure was chosen, dismiss that item and start a new one with the right frozen plan.
Why did setup skip Results?
Setup connects a source or unlocks delivery. It does not claim an earned page outcome, so there is no window to judge.
Does Couldn't measure trigger a revert?
No. Missing evidence is not evidence of harm. The work stays in place while the row explains what was missing.