Conducty Checkpoint

Conducty Checkpoint — Health-Aware Group Gate

Gate between parallelization groups. Verifies every prompt using calibrated review, computes health metrics, detects systemic failures, updates hill charts, and generates leverage-appropriate fix prompts before the next group begins.

The Iron Law

NO GROUP ADVANCEMENT WITHOUT INDEPENDENT VERIFICATION OF EVERY PROMPT

Do not trust self-reports. Verify independently using [[conducty-verify]] at the prompt's assigned review level.

Workflow

Step 1: Load the Plan

Read the active plan note from the vault (Plans/Plan YYYY-MM-DD HHmm [Topic].md — see [[conducty-obsidian]]). Identify the just-completed group.

Step 2: Systemic Failure Check

Before reviewing individual prompts, scan for systemic issues:

• How many prompts in the group completed vs. failed/blocked?
• Do failures share common symptoms (same error, same module, same missing context)?
• Did the tracer succeed but other prompts fail? (Suggests the tracer wasn't representative)

If 2+ prompts failed with related symptoms: Flag as potential systemic issue. Investigate at the plan/design level (via [[conducty-debug]] leverage analysis) BEFORE generating individual fix prompts. Individual fixes for a systemic problem waste retry slots.

Step 3: Verify Each Prompt (Calibrated)

For each completed prompt, apply its assigned review level:

verify-only prompts:

1. Run the verification command via [[conducty-verify]]
2. Pass? → mark completed
3. Fail? → mark needs-fix

spec-review prompts:

1. Run the verification command
2. If verification passes, dispatch spec compliance reviewer
3. Both pass? → mark completed
4. Either fails? → mark needs-fix, note which stage failed

full-review prompts:

1. Run the verification command
2. If verification passes, dispatch spec reviewer
3. If spec passes, dispatch quality reviewer
4. All three pass? → mark completed
5. Any fails? → mark needs-fix, note which stage failed

For every prompt, record the result in the evidence format from [[conducty-verify]].

Step 4: Compute Health Metrics

After reviewing all prompts, calculate:

### Group {X} Health — {timestamp}

| Metric | Value |
|--------|-------|
| Total prompts | N |
| First-attempt passes | N |
| Needs fix | N |
| Blocked | N |
| First-attempt pass rate | N% |
| Tracer result | pass/fail |

Interpret the metrics:

• Pass rate > 80%: Healthy. Plan and prompts are well-calibrated.
• Pass rate 50-80%: Concerning. Prompt quality or context may need improvement. Note patterns.
• Pass rate < 50%: Unhealthy. Likely a plan-level or design-level issue. Investigate systemically before generating fixes.
• Tracer failed but others passed: The tracer wasn't representative — note this for future planning.
• Tracer passed but others failed: The group had hidden dependencies or context gaps.

Update the running health metrics in the plan's ## Health Metrics (Running) section.

Step 5: Update Hill Chart

For each goal that had prompts in this group:

• Did all its prompts pass? → Move toward downhill (making it happen)
• Did prompts fail with design issues? → Move toward uphill (still figuring it out)
• Was progress made but work remains? → Note position

Update the plan's ## Hill Chart Update section.