REFACTOR: Count execution units in pyrit_scan progress bar

[Richard Lundeen (rlundeen2)]

Stacked on #2374. Base branch is copilot/romanlutz-scenario-pr-03-live-upstream, not main. Review only the diff on top of that branch. This change was extracted from #2374 (commit 5b3f8f2d7 removed it there) so it can be reviewed on its own.

What changed

pyrit_scan run prints a live progress bar while it polls a scenario run. The bar counted techniques completed. It now counts execution units — completed_attacks / total_attacks on ScenarioRunSummary — and the label changes from techniques: to units:.

An execution unit is (atomic_group_id, seed_group_id): one objective run under one atomic attack group.

Before:

  [███████████████░░░░░░░░░░░░░░░] techniques: 2/4 (50%) | success rate: 30% | IN_PROGRESS

After:

  [███████████████░░░░░░░░░░░░░░░] units: 5/10 (50%) | success rate: 30% | IN_PROGRESS

Why the technique-based bar existed, and why that constraint is gone

The removed code carried this comment:

# The bar tracks techniques completed / total, which is the only ratio we can
# honestly compute mid-run: the server only knows about attacks already persisted,
# so an attacks-based bar would always read 100%.

Techniques were a proxy metric. The server could not report a planned attack total, so total_attacks only ever held the number of attacks already written to memory, and completed / total was always 1.

The backend now persists a normalized run plan and computes canonical planned-vs-completed execution-unit counts. total_attacks is now a real planned total, so the bar can measure the work the user actually cares about instead of a proxy.

Deleted plumbing

total_techniques and scenario_meta were threaded through _run_scenario_async and _poll_until_terminal_async only to compute the old ratio. Both are gone, along with the max(total_techniques, techniques_done) reconciliation that existed because a user-supplied technique could be an aggregate (single_turn) that expands into N concrete techniques on the server.

This also removes a real inconsistency: pyrit_shell.py computed total_techniques from request.techniques only, with no all_techniques fallback, so the shell's bar could disagree with pyrit_scan's for aggregate techniques. Both now read the same server-computed counts.

Files

• pyrit/cli/_output.py — print_scenario_run_progress drops the total_techniques parameter and computes the bar from the run's unit counts.
• pyrit/cli/pyrit_scan.py — removes the total_techniques / scenario_meta threading.
• pyrit/cli/pyrit_shell.py — same call-site update in the interactive shell.
• tests/unit/cli/test_output.py, tests/unit/cli/test_pyrit_scan.py — updated assertions and call sites.

Net: 20 insertions, 40 deletions. No frontend or server change; this reads fields that already exist on ScenarioRunSummary.

Tests

uv run pytest tests/unit/cli -q → 516 passed.

pre-commit run --all-files: the two failing hooks (Check Memory Migrations, ty) fail identically on the base branch without this change.