# Feedback intelligence review

Rule-based triage, ready for analyst review. These counts describe this CSV, not unique customers or population sentiment.

Source: feedback.csv
Source SHA-256: `6912e6e591fbf31b65bb3b20c68abc0df86e4c943c18afca93f32b61ff53d6e9`
Rules: Synthetic subscription feedback - starter rules
Rules SHA-256: `bf1c8f72d160d0a040fc5da111e5f1312addd60d9f373288178fd30134b5043f`

## Inclusion and data quality

- Accepted feedback: 11
- Blank records ignored: 1
- Exact duplicate records removed: 1
- Conflicting IDs: 1 (2 records excluded)
- Accepted feedback with review flags: 5

Repeated IDs with different normalized fields are excluded entirely until the source owner resolves them. Exact duplicates are counted once. Physical source row references include the header and quoted multiline text; see evidence.csv.

## Theme coverage

Each theme coverage percentage uses accepted_feedback. A feedback can match multiple themes; percentages can total more than 100. Sentiment totals count accepted feedback once.

| Theme | Feedback | Coverage | Positive | Negative | Mixed | Unscored |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |
| Onboarding | 3 | 27.27% | 2 | 0 | 1 | 0 |
| Reporting and export | 4 | 36.36% | 2 | 1 | 1 | 0 |
| Billing | 3 | 27.27% | 0 | 0 | 2 | 1 |
| Support | 4 | 36.36% | 1 | 1 | 2 | 0 |

Sentiment belongs to the whole feedback record. It is not evidence of the sentiment toward each individual theme.

## Source evidence examples

Up to three accepted examples per theme, in input order. They illustrate the rules and are not a representative sample or ranked priorities.

### Onboarding

- FB-001, source row 2 (positive): Setup was easy, and the onboarding checklist was clear.
- FB-005, source row 6 (mixed): Getting started was confusing. A support agent was helpful.
- FB-010, source row 11 (positive): The setup page has an excellent checklist.

### Reporting and export

- FB-002, source row 3 (negative): The dashboard is slow; export failed twice.
- FB-004, source row 5 (positive): I love the new reports. The export is fast.
- FB-009, source row 10 (mixed): The report labels are clear, but billing is confusing.

### Billing

- FB-003, source row 4 (mixed): Support was helpful, but the invoice shows the wrong charge.
- FB-006, source row 7 (unscored): The refund arrived yesterday.
- FB-009, source row 10 (mixed): The report labels are clear, but billing is confusing.

### Support

- FB-003, source row 4 (mixed): Support was helpful, but the invoice shows the wrong charge.
- FB-005, source row 6 (mixed): Getting started was confusing. A support agent was helpful.
- FB-008, source row 9 (negative): The ticket form is broken and frustrating.

## Review queue

Resolve all ID conflicts first. Then inspect flagged accepted records and audit unflagged records for rule errors. Full flags and exact source fields are in evidence.csv.

- ID conflict: FB-011; source rows 12, 13
- FB-003, source row 4: mixed_sentiment
- FB-005, source row 6: mixed_sentiment
- FB-006, source row 7: no_sentiment_match
- FB-007, source row 8: no_sentiment_match, no_theme_match
- FB-009, source row 10: mixed_sentiment

## Analyst next steps

1. Correct or replace conflicting source IDs; rerun into a new output folder.
2. Review mixed, unscored and untagged records; document rule changes before comparing runs.
3. Read source evidence before proposing actions. No customer totals, financial impact, root cause or statistical significance is inferred.
4. Validate against human-labeled feedback from your domain. This tool has no claimed NLP accuracy.