Methods policy / version 1
A number needs a definition.
A readable study should let you identify exactly what was observed, what was calculated and what remains unknown.
Start with the question and population
Define the question, sources, inclusion window and exclusions before examining outcomes. A small convenience or topic-based sample must be labeled as such. Retain zero-result sources in the collection record instead of replacing them with more interesting ones.
Keep observation separate from interpretation
“This issue was closed at this time” is an observation. “The downstream team accepted the work” is a different claim that requires evidence of acceptance. Do not transform an available timestamp into the event a product would like to measure. Missing events remain unknown.
Count unfinished work
Open work has an elapsed age but no observed final duration. Keep it visible and report the denominator for closed-only statistics. Do not describe an open item as completed, silently discard it, or merge its age into a completed-work average.
Freeze the supporting data
Record observation dates, source queries, dataset versions and file hashes. Publish minimal supporting observations and aggregates where appropriate. Reproducing a frozen snapshot is different from recollecting a live service after records have changed.
Use methods readers can inspect
For public descriptive studies, publish the relevant calculation rules and a small reproduction script. That is separate from the commercial product's proprietary Handoff Reliability Index. This first study does not calculate or disclose HRI. No customer or cross-company benchmark is implied.
State limits next to findings
Report selection effects, missing fields, time-window effects, sample size and alternative interpretations. Do not infer causal improvement from an observational snapshot. Treat project comparisons as descriptions, not judgments of maintainers or personnel.
Publish and correct deliberately
Require source verification, arithmetic tests, citation links and public-page checks before release. Correct errors with a dated notice and versioned evidence. A machine-readable page helps readers and discovery systems find the same facts; it does not guarantee search inclusion or AI citation.