Count every hour behind a useful feedback case
By DX Research Group · · Incentives and participation
A capacity comparison includes participant support, duplicate triage and curation before ranking evidence yield.
A review channel can appear efficient when its cost ledger starts after someone else has prepared the case. We would compare evidence yield against all support and curation time required to produce it. The denominator changes which participation format deserves scarce staff capacity.
Our operating-layer controls paper companion describes historical failure diagnosis followed by harness revisions and replay tests. That sequence motivates a concrete output for new participation work: a distinct case with established expected behavior and enough evidence to reproduce the relevant stage. A submission count stops earlier in the process.
Compare two queues at equal capacity
Run a proposed four-week comparison of structured review sessions and asynchronous reports. Assign consenting, eligible owners randomly to one format before their first task. Cap each arm at the same 40 staff hours and offer equivalent opportunities to inspect recorded agent activity. Keep the case rubric fixed and have an independent reviewer classify issue families without seeing format.
The ledger includes onboarding help, live support, unsuccessful reproduction attempts, duplicate classification and final curation. Shared overhead is allocated by a declared rule, such as actual staff time. Participant effort is recorded separately because shifting work onto users can reduce staff cost while making participation harder.
Consider an illustrative ledger. Structured sessions yield 12 distinct reproducible cases. They use six hours of onboarding, ten of live support, eight of triage and six of curation: 30 total hours. Their yield is 12/30, or 0.40 cases per staff hour. Asynchronous reports yield nine cases with two hours of onboarding, three of support, five of triage and five of curation: 15 hours, giving 0.60 cases per hour.
Counting only curation reverses the ranking: structured sessions yield 12/6, or 2.0 cases per curation hour, versus 9/5, or 1.8. That narrower ratio hides the work needed to get cases into the curator's queue. Both calculations are valid descriptions of different stages; the total-hour ratio better answers the capacity allocation question.
Check what the extra cases cover
Cases vary in engineering consequence. Twelve variants of one explanation problem differ from nine cases covering confirmation recovery, stale state and policy rejection. Report distinct families and severity assessments beside yield. Freeze severity definitions before review and avoid turning an invented weighted score into the sole decision criterion.
DXAP's activity guide gives owners a route to inspect decisions and compare execution records. A proposed session can use that route to establish expected and observed states. Sensitive evidence stays within authorized channels; public research can report mechanisms and aggregate counts.
At the four-week cutoff, label cases still awaiting review and show remaining queue hours. Otherwise an arm can appear cheap by leaving difficult submissions unfinished. For the next cohort, we would allocate capacity to the higher-yield format while preserving a smaller route for issue families it misses. Repairs emerging from either queue still need regression checks and independent evaluation; the staffing result establishes evidence-production efficiency.