What did the survey
actually establish?
Review detection, quantitative error and the observations that disappeared behind a summary score. An executable audit of published TADI methane rate reports.
HISTORICAL EVIDENCE · SAVED RESULTS · NO LIVE GAS INFERENCE152 controlled releases
13 source systems · 804 saved rows
429 rows pass both recorded QC flags.
One facility, unequal schedules and rate ranges. This is a supplier-evidence worksheet; it does not commission a sensor or rank vendors for your site.
Primary study · February 2026Factor two means 0.5–2× the reference rate. Positive-truth errors include explicit zero reports. Counts are conditional on recorded QC-pass rows, not field uptime or a universal detection probability.
Zero-release controls and missing evidence
| QC-pass classification | Count | Meaning |
|---|
Stored rate / QC row evidence
| Release key | Reference kg/h | Report kg/h | Source QC | Rate classification |
|---|
Excluded numeric placeholders are not proof of a measurement. Unknown rate reports remain unknown. All rows, source flags and corrected ID namespaces survive in the download.
Current system's complete saved audit
Keep the denominator
146 missing or invalid rates were saved with “not detected” flags. This audit keeps them unknown. Many source rows are marked “not measured”; scheduling and QC exclusions must stay visible.
A source Boolean inconsistency affects 777 fail-all flags and 348 QC labels. That does not establish those rows were measured or that the paper results change.
Retain the failed analysis
The first join left 380 rows unmatched; its repair missed 74 and mismatched six rates. The corrected source-defined week-set join covers all 804 rows with zero rate mismatches.
18 authored checks, 182 separate Decimal metric comparisons and exact offline replay. No fitting, parameter search or fresh prediction confirmation.