The experiment Ids change the numerical effort numbers. I have tested it.
retained text
Following the recent reports about experiment Ids used for our chats, i investigated myself. My results show a clear link between the numeric effort Level and the experiment Ids. | Setup | medium | high | xhigh | |:--|--:|--:|--:| | Account A, has ID \`eab\` | 20 | 30 | 100 | | Account B, no ID | 10 | 15 | 40 | Account A was a new account with a new email address (Max 5). Account B was my old one, also Max 5. Removing the ID like described in the related post worked too. It lowered account A's effort numbers back to 10/15/40. Also on Nov 5, when the numeric effort post came up, i did some test chats, which allowed me now to link the ids \`763\` and \`add\` to the numeric effort levels of 15/30/100. Analysing my chats, these numbers did result in clear changes in real effort: Automated jobs, per reply (Opus 5.5, mixed effort settings, so all three effort numbers are shown) | | No ID (414 jobs) | \`763\` (14) | \`add\` (125) | \`0c7\` (25) | \`eab\` (35) | |:--|--:|--:|--:|--:|--:| | Reported effort number: medium / high / xhigh | 10 / 15 / 40 | 15 / 30 / 100 | 15 / 30 / 100 | not measured | 20 / 30 / 100 | | Tool calls per reply | 1.06 | 2.42 | 1.84 | 1.83 | 1.80 | | Output tokens per reply | 457 | 1,354 | 972 | 983 | 1,020 | | Thinking per reply, adjusted | baseline | ×2.9 | ×3.4 | ×3.9 | ×4.2 | | Replies per job | 52 | 19 | 20 | 19 | 16 | The effort numbers are the model's own statements, measured on 5 Nov for \`763\` and \`add\`, and on 9 Nov for \`eab\` and no ID.
document fetched 2026-11-10 00:39 · canonical https://reddit.com/r/ClaudeCode/comments/1xrmhiu/the_experiment_ids_change_the_numerical_effort