progress

progress — Trend Review Workflow

Minimum Data Thresholds

The value of progress scales with the data available. Before running the full protocol, assess what's in coaching_state.md and adapt:

Data Available	What You Can Do	What You Can't Do
1 scored session	Show baseline scores, identify initial patterns, set priorities. Say: "This is your starting point. I need 2-3 more data points before I can show you trends."	Trend narration, outcome correlation, graduation check (not enough data)
2-3 scored sessions	Show direction (improving/flat/declining), early pattern detection, preliminary self-assessment calibration	Reliable trend narration (inflection points need more data), outcome correlation (need 3+ real interviews)
4+ scored sessions	Full trend narration with inflection points and plateau diagnosis	Outcome correlation still requires 3+ real interviews
3+ real interview outcomes	Full outcome-score correlation analysis	Nothing — full protocol available

When data is thin (1-2 sessions): Don't run a hollow version of the full protocol. Instead, focus on: (1) what the available scores tell you right now, (2) what the most important next step is, and (3) what data you need before the next progress review will be useful. Say: "We don't have enough data for a full trend review yet. Here's what I can see from your [N] sessions, and here's what I need to give you a more useful picture next time."

When the candidate runs progress with no scored sessions: Don't output an empty schema. Say: "Progress tracks your improvement over time — but we need scores first. Run practice or analyze to get your first data point, then come back here."

Sequence

1. Check Score History and Intelligence size. If Score History exceeds 15 rows, run the archival protocol from SKILL.md: summarize the oldest entries into a Historical Summary narrative (preserving trend direction, inflection points, and what caused shifts per dimension), then keep only the most recent 10 rows as individual entries. Do the same for Session Log if it exceeds 15 rows. Also check Interview Intelligence archival thresholds (defined in SKILL.md): Question Bank at 30 rows, Effective/Ineffective Patterns at 10 entries, Recruiter/Interviewer Feedback at 15 rows, Company Patterns for closed loops. This keeps the coaching state file lean for long-running engagements.
2. Check data availability (see minimum data thresholds above). Adapt the protocol to what's actually possible.
3. Ask self-reflection first: "How do you think you're progressing? Rate yourself 1-5 on each dimension."
4. Compare self-assessment to actual coach scores over time (this is the most valuable part).
5. Narrate the trend trajectory (see Trend Narration below — don't just show numbers). Skip if < 3 sessions. 4a. Hard Truth (Level 5 only). Based on all accumulated data (Score History trends, storybank gaps, avoidance patterns from Coaching Notes, self-assessment deltas, outcome patterns), identify the single most important uncomfortable truth. One paragraph. No softening. No "but here's the good news." Just the truth the candidate needs to hear. See references/challenge-protocol.md for the Hard Truth lens. At Levels 1-4: omit entirely.
6. Check for outcome data and correlate with practice scores (see outcome tracking below). Skip if < 3 real interviews. 5a. Scoring Drift Detection (requires 3+ outcomes). Run the Scoring Drift Detection Protocol from references/calibration-engine.md: build the outcome-score matrix, check for systematic drift per dimension, check for feedback contradictions, generate drift report, present adjustments to candidate. Update coaching_state.md → Calibration State. Skip if < 3 outcomes. 5b. Cross-Dimension Root Cause Review. Check Calibration State → Cross-Dimension Root Causes (active). For each active root cause: assess treatment effectiveness (are affected dimensions improving in tandem?), check if resolution criteria are met (1+ point improvement sustained over 3+ sessions), update status. If a root cause isn't responding to treatment, recommend a pivot: "We've been treating [root cause] with [treatment] for [N] sessions. Affected dimensions aren't improving together. Let's try a different approach." 5c. Success Pattern Analysis (requires 1+ advancement or offer). Run the Learning from Successes protocol from references/calibration-engine.md: validate fit assessments, track positive dimension-outcome correlation, update storybank with success annotations, extract success patterns from 3+ successes. This ensures the system learns from what it got right, not just what it got wrong. 5.5. Outcome-Based Targeting Insights — When 3+ real interview outcomes exist, analyze rejection patterns for targeting signals. See Step 5.5 below. Also validate fit assessment accuracy: if fit assessments were recorded, check whether they predicted outcomes — learn from correct verdicts as well as incorrect ones. Skip if < 3 outcomes.
7. Check graduation criteria — are they interview-ready? (see Graduation Criteria below). Skip if < 3 sessions.
8. Identify top priorities based on triage, not just lowest scores.
9. Recommend drills and story updates.
10. Review and update Active Coaching Strategy. Check whether the current approach is producing results. If scores are flat for 3+ sessions on the target dimension, recommend a pivot: "We've been focused on [X] for [N] sessions and it's not moving. That usually means we need a different approach." Update the strategy in coaching_state.md — record the old approach in Previous approaches with the reason it was abandoned, and write the new approach with rationale and pivot conditions.
11. Run coaching meta-check (every 3rd session or when triggered): "Is this feedback useful? Are we working on the right things? What's not clicking?" Record the response in the Meta-Check Log.