hey jolanta,

sorted out the query consistency issue..

**the problem**:
- cityfibre exists as 6 different operator names in the db (wholesale relationships)
- chat agent was counting 'CityFibre Vodafone', 'CityFibre TalkTalk', etc as separate operators
- this inflated the overbuild counts massively

**bonus bug found**:
- nexfibre virgin media (80k postcodes) should be excluded from altnet queries
- your non-chat version was also including it incorrectly
- new version fixes both issues

**the fix (v0.1.57)**:
- added explicit operator canonicalization rules to context
- clarified when to use time-series vs static tables
- added correct sql pattern example for overbuild queries
- added nexfibre exclusion to all altnet queries

**verification**:
```
wrong way (chat):       1=9.6M, 2=1.6M, 3=3.1M, 4=881k, 5-7=more nonsense
better (non-chat):      1=13.3M, 2=1.9M, 3=180k, 4=17k (but included nexfibre bug)
correct (fixed):        1=13.0M, 2=1.2M, 3=100k, 4=8k
```

**to test**:
```bash
pip install --upgrade point-topic-mcp==0.1.57
```

then rerun your original prompts.. should be consistent and more accurate now.

see `.dev/progress_reports/mcp-query-consistency/summary_for_jolanta.md` for full technical breakdown if you need it..

~ peter

