Read the status before pressing again, and report what EC actually said

[amrali-eg]

Three findings from reviewing the cancellation state machine merged in #97, in
the order they matter. No production code changes; MainForm was read and
not edited. Rebased onto the #98 merge and revalidated as a combination.

1. A refused press could starve the status check

The press sat above the status read, and a refusal throws out of the press - so
on a refusing iteration the status was never examined. Progress depended on the
button disappearing rather than on the run reporting.

The status is now read first. With a button that stays visible and refuses, the
previous order runs the full thirty seconds and then claims no Cancel button
appeared, while the button was present throughout and the run had already
reported; reading first, the same conditions classify correctly.

2. A failed conversion was reported as a cancellation problem

Any final status other than Conversion stopped produced "Cancellation was not
exercised", including Conversion failed., which MainForm writes when a
genuine error occurs mid-run. That reports a real EC defect as an instrument
timing problem, in the phase whose purpose is checking EC's reporting.

Three outcomes now get three answers: Conversion stopped passes, Conversion complete says the run finished before it could be stopped, and anything else
says the run neither stopped nor completed and this phase proved nothing about
cancellation.

3. Two messages claimed more than they knew

pressed was set for a press whose delivery is unknown as well as one that
succeeded, and the timeout said "Cancel was pressed" - which the next clause
retracted. It is now pressAttempted, and says a press was attempted.

The other branch said "No Cancel button appeared"; it now says the driver could
not find one. Automation failing to see a control is not evidence about the
control.

Evidence

Control	Corrected	Previous
Button always present, always refusing	Classifies from the status	30s timeout, then claims no button appeared
EC reports Conversion failed.	"neither stopped nor completed... proved nothing about cancellation"	"Cancellation was not exercised"
Press lands, no final status	"A Cancel press was attempted..."	"Cancel was pressed..."
Full suite, after rebase	8/8
Unit tests	756 passed

Mutations required a successful build before running, with the source restored
byte-for-byte afterwards.

EC-28 remains open: none of this explains the phase A
failure that opened it.

🤖 Generated with Claude Code