Replay the Trace Before Changing Hardware

Replay the Trace Before Changing Hardware

animation

Replay the Trace Before Changing Hardware

Start assessment by replaying the trace against the intended rule. In the shelf example, a defensible replay table has one row per expected item and one row per boundary item. For each identity, record first-seen time, last-seen time, antenna IDs, signal indicator range if available, raw read count, middleware state, and operator note. The important comparison is not highest signal; it is whether the middleware decision matches the scenario label.

Use simple arithmetic so the result cannot hide in prose. If 12 target tags were expected and 11 became present, the target acceptance rate is 11 / 12 = 91.7%. If 4 boundary tags were present outside the intended zone and 1 became accepted, the boundary false-accept rate is 1 / 4 = 25%. If those 11 accepted tags produced 42 raw reads inside the dwell window but only 11 present events, then 31 raw observations were duplicate evidence that the suppression rule handled. None of those numbers is a universal pass threshold; the pass threshold is the one the lab wrote before testing.

Evidence field Why it matters Release implication
Expected target count Defines the denominator for missed-read analysis A missing target is a failed inventory event until retested or accepted as a known limit.
Boundary tag count Proves whether the zone excludes nearby objects An accepted boundary tag means antenna aim, power, shielding, or workflow geometry needs review.
Dwell window Prevents repeated raw reads from becoming repeated business events The release record must document the window because changing it changes behavior.

Change one variable after replay, then repeat the same scenarios. If moving the antenna down 10 cm produces 12 of 12 accepted target tags and 0 of 4 accepted boundary tags, the improvement is credible only because the tag set, dwell rule, and scenario sequence stayed fixed. If the same change also increases duplicate raw observations from 42 to 93, that is not automatically bad, but the release record must show that duplicates still collapse to one application event and that gateway load remains acceptable for the exercise or pilot.

Back to the chapter