“Week five of Consent Architecture made us delete advertising_id from three ‘anonymous’ events. Growth was unhappy for a month. Counsel was not.”
Reviews
What changed after the critique.
These notes reference specific modules. We keep a mild reservation in the mix on purpose — evening density and Android consent flows are not glamorous, and they matter.
I took Event Schemas Without Device Fingerprints after our probabilistic graph failed a PDPA questionnaire. The coarse-cohort worksheet is still taped above my monitor. The recorded Field Notes version would have been enough if I did not need live markup.
★★★☆☆ · desk copy
CMP Bridges is short and sharp. I wanted more time on Huawei HMS consent mirrors; the lab stayed on Play + App Store. Still used the fan-out diagram in a vendor rejection letter.
A client in grocery delivery sent two engineers to PDPA Measurement Studio. They came back arguing about retention clocks instead of dashboard colours, which is the first time legal and product shared a spreadsheet without a fight.
“The revoke drill in week ten exposed a BigQuery extract that kept opted-out rows for ‘completeness.’ We scheduled the purge the same night.”
Evening Instrument Bench after a full office day is a lot. I missed one critique window and felt the gap. The deny-by-default wrapper sample still saved us a sprint of speculative SDK flags.
Case studies
Two longer readouts
Ride-hailing: sleeping the SDK until the CMP returned
A Bangkok mobility app initialised its analytics library in Application.onCreate. Consent Architecture fellows mapped every disk write before the first activity. They introduced a gate that queued only crash breadcrumbs locally, still without identifiers, and flushed product events only after a documented basis.
Opt-in held near 61% on Android after the change. Funnel completeness dropped. The team now publishes an “instrumented sessions” metric beside MAU so leadership does not treat silence as churn.
Regional marketplace: PDPA retention vs. vendor default of 26 months
PDPA Measurement Studio participants discovered their CDP retained advertising identifiers for the vendor’s default window, far past the controller’s stated purpose. The studio critique produced a 13-month clock for marketing properties and a 90-day clock for diagnostics, with a documented exception for fraud holds.
Cross-border processing stayed; the contract addendum now names the Thai controller and the deletion SLA. Attribution lookback shortened. Finance accepted a messier LTV chart rather than an indefensible one.