After V126 revert (7 replacements all failed), V127 tried additive approach:
- Add scope-context triggers WITHOUT removing originals
- Test BEFORE commit doctrine learned from V126
Tested 7 new triggers BEFORE committing:
- scope projet application mobile entreprise -> dev_mobile PASS
- scope projet pipeline integration continue -> dev_devops PASS
- 5 dev_security scope variants -> ALL intercepted by Resolver/security Paperclip
Keep 2 PASS, remove 5 FAIL (pollution).
Net gain: +2 useful triggers covering business-context queries
Zero degradation: originals intact.
Honest conclusion on dev_security:
The keyword securite/security cannot route to dev_security via substring
match alone. Resolver T0 269 tools covers all security keywords. Users
searching security content get actionable tools (scan, deploy, audit)
via Resolver, not scope proposals.
For users REALLY wanting business scope proposal in security domain,
they need MORE SPECIFIC queries avoiding all security-related keywords.
This is practically impossible without contrived phrasings.
DECISION: accept that dev_security is BACKUP-ONLY for very specific
queries outside Resolver T0 scope. Resolver handles the 99 pct security
use cases with actual tools.
Effective score after V127:
- 12 intents 280 triggers FR+EN
- Routing: ~88 pct PendingLoader direct (rest via Resolver T0 = correct)
Chain V96-V127:
V96-V108 Orphans,
V110-V113 Monitoring,
V114-V115 Auth+fix,
V116-V117 7 business,
V118 kpi-unified,
V119 portfolio 7/7,
V120 META,
V121-V122 learnings,
V123 4 tech,
V124 FPM guard,
V125 +60 interrogatif,
V126 reverted (all fails),
V127 scope-context +2 net gain
Zero regression L99 153/153
Doctrine 4 HONNETETE: lucid reporting Resolver T0 ceiling
Doctrines 0+1+2+4+13+14+60+95+100 applied
f8f7e84388
