{"authors":[{"name":"Marcin"}],"description":"Science-based endurance training, augmented by AI.","feed_url":"https://aerobicmodel.run/feed_json_updated.json","home_page_url":"https://aerobicmodel.run/","icon":null,"items":[{"authors":[],"content_html":"A step-by-step walk through one conversation: last week's review, today's readiness, next week's plan, and a 6×400 m session that waits for your yes before it goes anywhere near your calendar.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-18T00:03:00+00:00","id":"https://aerobicmodel.run/blog/chat-to-watch/","image":null,"tags":["AI for training","Running"],"title":"From chat to watch: planning a training week with an AI assistant and Garmin","url":"https://aerobicmodel.run/blog/chat-to-watch/"},{"authors":[],"content_html":"I measured what my MCP server cuts from Garmin's raw JSON, then asked three Claude models 37 questions with each version of the data. The answers were the same. The bill was not.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:08:00+00:00","id":"https://aerobicmodel.run/blog/compact-output/","image":null,"tags":["Deep dives","Running"],"title":"I fed an AI raw Garmin JSON. Same answers, over 20 times the bill.","url":"https://aerobicmodel.run/blog/compact-output/"},{"authors":[],"content_html":"I asked three Claude models the same 20 questions about my Garmin data, three times each, and itemised every bill. Haiku answered for under a cent on average, Opus for 11 to 23 cents, and all three got the facts right.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:10:00+00:00","id":"https://aerobicmodel.run/blog/cost-per-question/","image":null,"tags":["Deep dives","Running"],"title":"Opus costs 30 times more than Haiku to answer my training questions. It wasn't better.","url":"https://aerobicmodel.run/blog/cost-per-question/"},{"authors":[],"content_html":"How I shaped an MCP server around the questions people ask, kept its tool list small with a test that counts bytes, and what those bytes turned out to be in tokens.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-21T00:04:00+00:00","id":"https://aerobicmodel.run/blog/designed-tools/","image":null,"tags":["Deep dives"],"title":"Designed tools, not an API mirror: 21 MCP tools on a byte budget","url":"https://aerobicmodel.run/blog/designed-tools/"},{"authors":[],"content_html":"I scored 367 sessions of my 2026 season with both. They ranked the sessions almost alike, but one TSS point was worth 0.8 Garmin points on a hike and 3.3 in an interval run. In the weeks when fitness clearly moved, both curves said the same thing 15 times out of 16.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:12:00+00:00","id":"https://aerobicmodel.run/blog/garmin-load-vs-tss/","image":null,"tags":["Running","Science, explained","Triathlon"],"title":"My AI coach has two training load scores. They disagree on most of my sessions.","url":"https://aerobicmodel.run/blog/garmin-load-vs-tss/"},{"authors":[],"content_html":"I scored 186 runs and rides of my 2026 season from heart rate and from pace or power. Inside a 20-second stride, heart rate gave half the stress that pace did. Over whole interval sessions it came out about where it does on my easy runs. The big gaps were mountain races and easy rides, and the heat did not show up at all.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:13:00+00:00","id":"https://aerobicmodel.run/blog/hrtss-intervals/","image":null,"tags":["Cycling","Running","Science, explained"],"title":"Heart rate lags on intervals. Over a season, my AI coach's training load barely noticed.","url":"https://aerobicmodel.run/blog/hrtss-intervals/"},{"authors":[],"content_html":"What fitness, fatigue, form and the workload ratio mean, where they come from, and where they stopped making sense in nine months of my training and racing.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-18T00:02:00+00:00","id":"https://aerobicmodel.run/blog/load-model-explained/","image":null,"tags":["Running","Science, explained","Triathlon"],"title":"CTL, ATL, TSB and ACWR, explained on my own season","url":"https://aerobicmodel.run/blog/load-model-explained/"},{"authors":[],"content_html":"How my Garmin MCP server downloads only what a question needs, paces itself against an API with no published limits, and signs in through Garmin's own page.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-18T00:06:00+00:00","id":"https://aerobicmodel.run/blog/local-first-cache/","image":null,"tags":["Deep dives"],"title":"A local-first cache in front of a rate-limited API (and a sign-in that never sees your password)","url":"https://aerobicmodel.run/blog/local-first-cache/"},{"authors":[],"content_html":"Three Claude models answered 50 training questions from my Garmin data, twice each. Of the numbers they copied from the tools, 2 in 10,009 were wrong. Of the ones they calculated themselves, 10 in 573.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:09:00+00:00","id":"https://aerobicmodel.run/blog/number-faithfulness/","image":null,"tags":["AI for training","Running"],"title":"I checked 10,588 numbers my AI coach gave me. The copying was fine. The maths wasn't.","url":"https://aerobicmodel.run/blog/number-faithfulness/"},{"authors":[],"content_html":"I matched 128 planned sessions from my 2026 Garmin calendar with what I recorded. Seven in ten of my own plans happened on the day, but only 38% of all of them ran as the workout on my watch, and inside those, 65% of the steps with a target landed in it.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:11:00+00:00","id":"https://aerobicmodel.run/blog/planned-vs-done/","image":null,"tags":["AI for training","Running","Triathlon"],"title":"AI can put a workout on my watch. Did I actually run it?","url":"https://aerobicmodel.run/blog/planned-vs-done/"},{"authors":[],"content_html":"A small eval for an MCP server: one prompt, one expected first call, the real server with a fake data layer, and what its misses taught me about tool descriptions.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-21T00:05:00+00:00","id":"https://aerobicmodel.run/blog/tool-selection-eval/","image":null,"tags":["Deep dives"],"title":"How I test that the model picks the right tool: 96 scenarios, 96 first calls right","url":"https://aerobicmodel.run/blog/tool-selection-eval/"},{"authors":[],"content_html":"I built an open-source tool that lets an AI assistant read my Garmin data and answer training questions with numbers I can check: the values, the rules and the papers behind them.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-18T00:01:00+00:00","id":"https://aerobicmodel.run/blog/train-hard-today/","image":null,"tags":["AI for training","Running","Triathlon"],"title":"I asked my AI assistant if I should train hard today. It answered with my HRV.","url":"https://aerobicmodel.run/blog/train-hard-today/"},{"authors":[],"content_html":"How I designed a structured-output schema for an LLM tool with zod and a small compiler, and what 643 calls to Claude Haiku, Sonnet and Opus 5.5 showed against Garmin's own workout JSON: a tie on accuracy, not on the bill.","date_modified":"2026-11-22T17:58:49+00:00","date_published":"2026-11-22T00:07:00+00:00","id":"https://aerobicmodel.run/blog/workout-schema/","image":null,"tags":["Deep dives"],"title":"Stop making the LLM write your API's JSON","url":"https://aerobicmodel.run/blog/workout-schema/"}],"language":"en","title":"Aerobic Model","version":"https://jsonfeed.org/version/1.1"}