Quality Ranger · Learning batch · QR_Fix_Yes_Tasks.csv

Why QA tickets get sent back to QR Fix

Every one of the 74 QA-OPT tickets flagged “QR Fix = Yes” was traced to the exact reviewer bounce that caused it — the Waiting For Fix (QR) transition and the comment behind it — then grouped into cause families. This is a living tally: as the team fixes behaviors upstream, the ranking will shift, and that shift is the signal worth tracking.

74Tickets analyzed

84Cause occurrences

~52%Are data-privacy

1Indeterminate

The dominant families

01 / SHARE OF CAUSES

Share of all 84 cause occurrences. One masking family accounts for more than the other four combined — and it is almost always a leak in an attached screenshot or sheet, not in the written note.

Data Privacy / Masking 44 · 52%

PII in QA’s own evidence (P-001) + DB table/column names, tokens, endpoints (P-005)

Investigation / Validity 19 · 23%

Hand-waved or unverified conclusions; too narrow a sample (P-003/06/07/08/09)

Communication hygiene 11 · 13%

Missing QR note (P-004) + untranslated Turkish (P-002) — cheap, mechanical catches

QA process 8 · 10%

Leftover test artifacts (P-010) + recommending/temporary-fixing instead of doing (P-011)

Bug escalation 1 · 1%

Escalation template incomplete — missing Estimated Due Date (Criterion 5)

Every cause, ranked

02 / BY PATTERN

The twelve distilled patterns, most frequent first. Bars are relative to the top cause (P-001, 27 hits). Each links back to its rule in QR-FAIL-PATTERNS.md.

#	Cause	Pattern	Occurrences
1	Data privacy — PII exposure	P-001	27
2	Data privacy — internals & secrets	P-005	17
3	Validity — hand-waved / unproven conclusion	P-009	8
4	Test coverage — thin reproduction / verification	P-006	6
4	QR note — missing / incomplete at submission	P-004	6
4	Test cleanup — leftover campaigns / segments	P-010	6
7	Language — untranslated Turkish	P-002	5
8	Verification — unverified-fix closure	P-008	2
8	Coverage — requested deliverable skipped	P-003	2
8	Action not taken — workaround instead of the real fix	P-011	2
11	Verification — viewer-context not confirmed	P-007	1
11	Bug escalation — template incomplete	Crit 5	1

What the reviewer is really guarding

03 / TAKEAWAYS

Three habits catch almost everything

Read every attached image and sheet for leaks. Data privacy is ~52% of all bounces and the leak is nearly always in a screenshot or export — an unmasked email, a UUID in a URL bar, a DB column header, a token. The prose is usually clean; the evidence is not.
Refuse a conclusion that is too thin. Roughly a quarter of bounces are validity: an answer that is hand-waved, contradicts its own evidence, or rests on a single data point (one UTM parameter, one platform) the reviewer wanted broadened.
Run the mechanical gate. A missing QR note, untranslated Turkish, or a leftover test campaign are cheap, unambiguous catches — together ~23% of bounces.