User reference / Run an experiment
6. Generate the report
Turn saved records into linked, interactive HTML pages without rerunning the council.
slow-thinker-report --base config/slow-thinker.study.json
Without --analyze, this reads records and produces HTML without making model calls. It computes title-based comparisons, lineage, contribution counts, charts and price estimates, and includes a compatible saved analysis if present.
| Argument | Effect |
|---|---|
| Positional task UUIDsReviewed R 0.1.16 | Regenerate only those run reports. The run index, prompt library and prompt histories are still rebuilt for all known tasks. |
--base FILEReviewed R 0.1.16 |
Load the base chain for storage, model/analyst settings and prices. With no argument, use default base-file discovery. This command does not read an individual task JSON. |
--runs-dir DIRReviewed R 0.1.16 |
Override the HTML destination. It does not change which JSON storage directory is read; that still comes from configuration. |
--examples DIRReviewed R 0.1.16 |
Task files to include in the prompt catalogue, default examples. Prompts can be listed before any execution exists. |
--analyze TASK_IDReviewed R 0.1.16 |
Add an LLM analysis of that task before rendering. |
--analyze allReviewed R 0.1.16 |
Analyse tasks without a stored analysis, then render. |
For rendering JSON files without model configuration, a small base such as {"runs_dir": "runs"} is sufficient. A configured base may still require its referenced files and provider keys to load successfully even when you only want to render.
| File | Contents |
|---|---|
index.htmlReviewed R 0.1.16 |
All known runs, grouped by prompt, with a second “Models across runs” view. |
prompts/index.htmlReviewed R 0.1.16 |
Prompt library combining task files and recorded experiments, including prompts not yet run. |
prompts/<prompt_key>.htmlReviewed R 0.1.16 |
History of executions of one prompt and its full statement. Created for prompts with runs. |
<task_id>/report.htmlReviewed R 0.1.16 |
Detailed execution report with the latest stored analysis, or the trace alone. |
<task_id>/report.a<analysis-prefix>.htmlReviewed R 0.1.16 |
The same trace presented with an earlier stored analysis. |
The report files embed their styles, scripts and SVG charts. A report can be opened directly, although relative links need the accompanying library structure. For local navigation:
python -m http.server 8765 --bind 127.0.0.1 --directory runs
Then open http://127.0.0.1:8765/index.html. Stop that server with Ctrl C when finished. It serves existing files; it does not run or analyse a case.
Copy the generated HTML tree with relative paths intact. If only one report is copied, its “This prompt”, “All runs” and historical-analysis links can break. On this website, the library lives under /assets/slow-thinker/reports/, and the exported pages add a “Project” link back to the project overview.
Downloadable experiment archives are a website publication feature, not an automatic output of slow-thinker-report. Include the corresponding prompt and configuration records as well as task records if you want an archive to be inspectable independently. Do not include the private keys file.
Rendering applies the current report code and the price list available at render time. It does not rerun the council or revise an analyst's conclusions. A newer report can explain an older trace more clearly, but it cannot recover call metadata or declarations that were never recorded. Its footer records the renderer version and generation time; the original executor and analysis versions remain separate.
Use the library guide to navigate the output, and versions to interpret older records.
Source files used for this reference
Reviewed at 7361ea0b71a5.