🐛 Bug - Inconsistent stats between sprint review and summary meetings

[yaqi-lyu]

Cc: @suiyangqiu @yaqi-lyu @calumjs @joshbermanssw
Hi Team!

🟥 Watch the video (01:21)

Pain

For sprint review meetings, there are two recordings: the actual review/planning recording and its summary. However, these two recordings are showing different stats for the same sprint:

• Actual review recording: 48 points completed
• Summary recording: 50 points completed

These numbers should match since they refer to the same sprint's completed work. The inconsistency suggests that Claude is extracting different information from the transcripts, likely picking up different mentions of points from the conversation.

Acceptance Criteria

1. Sprint review stats (points completed) should be consistent across the actual meeting recording and its summary
2. The system should correctly identify and extract the canonical sprint metrics from transcripts
3. If there are multiple mentions of points in a transcript, the system should prioritize the official/completed sprint metrics over estimates or projections
4. Add validation to detect when two related meetings (e.g., sprint review and summary) have conflicting stats

Reproduce Steps

1. Process a sprint review meeting recording through SSW.Tiger
2. Process the corresponding sprint review summary recording
3. Compare the "points completed" stats between the two generated dashboards
4. Observe that the numbers differ (e.g., 48 vs 50 points)

Related Issues

• ✨ Manual upload .vtt transcript - Generate dashboard for Teams ad-hoc calls #58 - Manual upload .vtt transcript - Generate dashboard for Teams ad-hoc calls
• 🐛 Bug - Claude Stats workflow not triggered for recorded meetings #18 - Bug - Claude Stats workflow not triggered for recorded meetings

Thanks!

Prototype by @calumjs

[yaqi-lyu]

As per @18-th , the efficiency format is inconsistent too. It’s shown as “B+” in the sprint review but as a number in the summary. Please make sure this is consistent.

[yaqi-lyu]

After checking the transcript:

• Part 1 around 00:30:14: “which is 8 points”
• Part 1 around 00:43:36: “that brings us to 36 points”
• Part 2 around 00:08:29: “thirty-two completed and twenty-one remaining”
• immediately after: “our sprint tracker is not exactly correct”

No “48 completed” statement appears in those two original files. Likely reason Claude produced 48: it inferred or synthesized from point discussions in Part 1/2 instead of extracting an authoritative metric. The original transcript has multiple nearby numbers. And because there's a discussion in the summary, so it get 50 points.

There is no deterministic rule telling it the point metrics.

• The analytics/consolidation schema supports both:
  - score: numeric, e.g. 58
  - grade: letter, e.g. D+
  • The analytics prompt also includes keyNumbers with "Effectiveness": "58/100" and context "D+ grade".
  • So different runs can choose to display the letter grade in one place and the numeric score in another.
    We should make this consistent.