QHuman

Work the order queue

The office multi-step flagship. Click a row to open its detail, read the recommended decision, pick the matching status, save, and move to the next order. This is read then act across a detail view, with no colour shortcut.

master / detail OCR + branch multi-step Pro · free preview 20 min

Most labs let you cheat a little with colour. This one does not. A macro cannot pixel-match its way to the right answer, so it has to read each order's recommendation and branch on the words. That makes it the best target for learning QHuman's read then act loop end to end: master list, detail view, a real decision, and a save.

Reading a value with OCR and previewing that read is free. Acting on the read in a live run is a Pro feature, so this lab builds and rehearses the whole flow for free with Try it (dry run), then goes live on Pro.

Open the target first. It opens in a new tab. Point QHuman at it and read its counters as you go.

Open Order Desk

Seven orders on the left, the open one in the middle, the score on the right.

Look for a colour that means something and you will not find one. Every chip is the same grey whether it says New, Approved, Held or Rejected, and the three status buttons are identical apart from the word on them. The only thing that tells this order from the next is the sentence Recommended: Approve, and reading a sentence is OCR's job. R resets the counters and reshuffles the queue.

Build the flow

  1. Position the target and start a detection flow Put the Order Desk where QHuman can see it and leave it there: the OCR region you draw is absolute screen coordinates, so moving the window later breaks the read. In a New Detection named Order Sweep, add a step called read. Being first, it is the start.
  2. Add the three route steps The destinations have to exist first: a transition can only offer steps that already exist. Add three more steps named do approve, do hold and do reject. Leave them empty for now; you fill them in step 6. The three status buttons share one fill and one border, so the word is the only thing that tells them apart: that is why you snip each button's word (not a colour) for its click, and why the read has to be OCR.
  3. Read the recommendation with Add Text (OCR) In read's Then list, add a transition with Go to step: do approve. Switch Move on: to Advanced condition (AND / OR / NOT), then click Add Text (OCR), set the comparison to contains, and the text to Recommended: Approve (the whole line, not the bare word). Reading the whole sentence keeps a stray Approve from ever satisfying this branch: the button below says Approve, but it never says Recommended: Approve.
  4. Select the region on screen Tick Limit detection to a screen region, press Select region on screen..., and drag a box around the Recommended: Approve line only: not the eyebrow above, not the buttons below. Press Preview detection... to check the read, then OK.
    The box goes around one sentence. The eyebrow says RECOMMENDED ACTION on every single order, so it carries no information and only gives OCR more to get wrong. And the buttons below say Approve, Hold and Reject all the time, no matter what this order needs, so a region that dips into them reads a stray status word into your line. Matching the whole Recommended: sentence is your backstop against that; keeping the box tight to the line is what stops it happening in the first place.
  5. Branch into three routes Add two more transitions the same way: Go to step: do hold with contains Recommended: Hold, and Go to step: do reject with contains Recommended: Reject. Draw each the same region box over the Recommended: line and OK. Three reads of one line, three destinations. This is the branch, and it is the whole reason this flow cannot be a recording: the same screen position means Approve on one order and Reject on the next.
  6. Click the matching status, then Save Fill the three route steps. In do approve, add a click, leave Find: on Image, press Capture snippet..., and snip the Approve button's word (nothing but the word, since the buttons are otherwise identical). Add a second click for Save decision. Repeat for do hold (its button says Hold) and do reject (Reject). The pairing is the whole risk: wire do hold to the Approve snippet and every counter still moves, just into the wrong column.
  7. Loop back for the next order Each route needs to come back for the next order. In each of do approve, do hold and do reject, add a transition with Go to step: read and Move on: Always (every time). Saving opens the next order by itself, so by the time the flow returns there is already a new recommendation waiting in the same box.
    One read, three ways out, all three coming back to the same place. Four steps total. Note that the word in the condition and the word on the snippet have to agree, and nothing in QHuman checks that for you: contains Recommended: Hold wired to a step that clicks Approve is a flow that runs perfectly and is wrong seven times out of seven.
  8. Rehearse with Try it (dry run) Press Try it (dry run), then Start. Event timeline shows each read and the transition it took, with nothing clicked and nothing saved. This is the free half of the lab, and the half that matters: it proves the reads and the routes before a single decision is committed to the queue.
  9. Go live (optional, Pro) Not needed to finish the lab. Acting on an OCR read in a live run is a Pro feature. On Pro, reset the queue with R and run for real: the flow reads, picks the status, and saves down the whole queue. To bail out, slam your mouse into any screen corner: the fail-safe is on by default, under Settings → Playback.
Checkpoint

The free bar is the dry run: Event timeline takes the route that matches each order's Recommended: line, seven times, with no mismatch. That proves the reads and the branches, and it needs no Pro. A live run on Pro then drains the queue for real and ends like this.

Orders processed is always Correct plus Wrong-status saves, so a 7 on the top tile only means it saved seven times, not that it was right seven times. Correct at 7 is the real result.
Make it yours

Press R and run the dry run again. The queue reshuffles on every reset, so the seven orders arrive in a new order with their recommendations in new places, and a flow that truly reads will not notice. Watch Event timeline take a different route than last time and still be right every time. That is the difference between a macro that knows the answers and one that reads the question.

If it went wrong

  • Wrong-status saves climbing A route saved a status that did not match the recommendation. Rehearse with Try it (dry run) and read back what OCR returned. If a Hold or Reject order shows a stray status word tacked onto the line (Recommended: Reject … Approve), your region is dipping into the buttons below, so pull its bottom edge up to the Recommended: line. Matching the whole Recommended: phrase already stops that stray word from flipping the branch; a tight region keeps the read clean on top of it.
  • Every order takes the same route, or none fires The read is returning the same thing each time (usually a region that misses the line), or nothing your conditions match. Reselect a tight region on the Recommended: line and Preview detection...: it must show the whole line, Recommended: Approve and its colon, with the word changing order to order.
  • Reads are flaky or blank Tighten the region first, and check it with Preview detection...: on flat text like this, a box that is too big is almost always the cause. OCR preprocessing under Settings → Detection is the other knob, but it is greyed out on Free, so it is not the one to reach for here.
  • Clicks land on the wrong spot Moving the window is not the problem: a click finds its picture again every run, wherever it moved to. The read is what got pinned, because the OCR region is fixed screen coordinates. Move the Order Desk back, or select the region again where it now sits.

Stuck?

Grab the finished flow. Download it and open QHuman's Import shared file... to load it, then compare it to what you built. It is a reference to read, not a macro to run: the detection snippets inside were captured on another screen, so they will not match your display. Run with the ones you took in the lab instead.

Download the solution (.qhz)

Keep going

Want the full detail behind reading values off the screen? Read the read text & numbers guide. Ready for the next lab? Next up is Morning routine, where you chain whole macros into one start-of-day run.