The HTML results directory does not store results for multiple attempts at the same task

[joellabes]

Two rows in the results table:

But they both have the same path

And there is only one folder in the html directory, which points to the first run's results

Whereas there are two folders in the airbnb004 directory:

To reproduce:

1. Run ade run airbnb004 --agent claude --db duckdb --project-type dbt --n-attempts 2
2. Run ade view
3. Click the yellow Results button beside each run
4. Notice that the number of output tokens is identical, even though the summary table had different data per run

[bstancil]

So, this makes this a slightly bigger problem, but I think I'd argue that we might want to change how we name each run. Right now, it creates a run_id that is:

• usually the task_id
• is the task_id.[key] if there are multiple prompts
• adds 1-of-2 or whatever if there are multiple attempts.

it's pretty messy, and can create collisions of all kinds (including in how it names containers). It makes stuff a little messier, but I think I'd be tempted to name all of this stuff task_id.[key if needed].[1-of-2 if needed].[a very short hash], and then do all of the matching and lookups and stuff on the hash. So basically, create a (short) unique id for the run, and use that to match containers to results to executions and stuff. (The way it's done now also means there's no way to run mutiple variants at once, which is also annoying. eg, if you create a new task, you really want to run ade run foo001 --db duckdb snowflake --project-type dbt dbt-fusion --agent none sage and have it spin up like 10 tasks. You can't do that for other reasons right now - the run has a single db, etc - but we'd also have to update this naming convention to support that.