The Comm LayerI write about AI communications and fraud at the platform layer. A decade on both sides of the comms API: American Express, Twilio, now Microsoft. The hard problems are about accountability and architecture, not better models. Views my own.https://www.archanakumari.fyi/Why a 94.7%-accurate smish classifier is a false-positive machine in productionhttps://www.archanakumari.fyi/blog/article_precision_vs_prevalence/https://www.archanakumari.fyi/blog/article_precision_vs_prevalence/Same model. Two prevalence assumptions. ~75 percentage points of precision difference. This is why public-benchmark accuracy does not transfer to fraud productionSat, 30 May 2026 00:00:00 GMTFriction Debthttps://www.archanakumari.fyi/blog/friction-debt/https://www.archanakumari.fyi/blog/friction-debt/Friction debt is the cost a platform accumulates every time it makes it easier for users to do actions at scaleSun, 17 May 2026 00:00:00 GMTThe Eval You Run When the Clock Is Runninghttps://www.archanakumari.fyi/blog/post1_lightweight_eval/https://www.archanakumari.fyi/blog/post1_lightweight_eval/Most writing about evaluation assumes you have time. It assumes a held-out dataset, a labeling panel, a power analysis, weeks of iteration. That assumption is fine for shipping decisions. It is useless during an active attackMon, 25 May 2026 00:00:00 GMTThe Eval That Decides Whether You Shiphttps://www.archanakumari.fyi/blog/post2_heavy_eval/https://www.archanakumari.fyi/blog/post2_heavy_eval/The companion to this post described the lightweight evaluation: the one you run under a clock, when a campaign is active and a response ships in hours. This post describes the other instrument. The heavy evaluation is what gates a model release. It is slow, deliberate, and expensive, and those properties are features rather than defects.Wed, 27 May 2026 00:00:00 GMTRules Are the Bones, Models Are the Musclehttps://www.archanakumari.fyi/blog/post3_rules_vs_ml/https://www.archanakumari.fyi/blog/post3_rules_vs_ml/This is the third post in a short series on evaluating fraud detection. The first covered the lightweight evaluation you run during an active campaign. The second covered the heavy evaluation that gates a model release. This one steps back to a question that comes up every time someone reviews a detection system for the first time.Sat, 30 May 2026 00:00:00 GMT