# A Weekly AI Coding Report Is an Activity Ledger, Not a Productivity Score > Source: > Published: 2026-07-25 23:11:23+00:00 A week of Claude Code and Codex work leaves enough local evidence to answer concrete questions. When were sessions active? Which provider produced the activity? Which models appeared? How did the token mix change by day? It does **not** prove that the code was correct, the week was productive, or an API-value estimate was money spent. That boundary is what makes a weekly report useful. A readable report should cover today plus the previous six local calendar days. The headline, daily bars, provider totals, and model rows must use the same boundary. If the chart uses calendar days but model rows use a rolling 168-hour window, the parts can disagree. A row may exceed the headline even though every individual calculation looks plausible. Timezone and midnight behavior are accounting rules. Input, output, cache creation, and cache reads describe different work. I find it useful to keep two totals: Neither is a productivity score. A cache-heavy refactor can move more tokens than a small, high-impact fix. A failed run can be expensive. A quiet session can be waiting for a person instead of doing nothing. A provider split can show how much recorded activity came from Claude Code and how much came from Codex. It cannot show that one token from each provider represents equal work, latency, quality, or quota pressure. Provider quota is reported state. Local token activity is a reconstructed ledger. Displaying both is fine. Using one to fill missing data in the other is not. A session can switch models. A later token event needs the model context active at that time, not whichever model appears at the top of the file. Unknown named models should stay visible and unpriced. Guessing a nearby model family makes historical reports impossible to audit. Token rank and API-value rank are also different. A model with many inexpensive cache reads can lead by volume while another leads by estimated value. I use the report as a prompt for one operational decision: That change might be reducing unattended runs, fixing a recurring authentication failure, or changing when I start a long session. The number alone does not improve the workflow. The decision does. A local report should stay local by default. Agent Island builds its report from records already stored on the machine; transcript content is not uploaded to an Agent Island service. A shareable card is a rendered summary, not an automatic broadcast. Copying or exporting it requires an explicit action. Before sharing, review visible dates, totals, model names, and project context for anything you do not want to disclose. A weekly report earns trust by saying what each number means and what it cannot prove. The complete workflow and current verified scope are in the canonical guide: [AI coding weekly report](https://agent-island.dev/ai-coding-weekly-report/?utm_source=devto&utm_medium=owned_social&utm_campaign=ai_coding_weekly_report_20260726).