85 lines
1.8 KiB
Markdown
85 lines
1.8 KiB
Markdown
|
# EOS Runtime — Experimentell Baseline
|
||
|
|
|
||
|
|
**Version:** 0.9-experimental
|
||
|
|
**Datum:** 2026-07-01
|
||
|
|
**Commit:** (fryst)
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Mått
|
||
|
|
|
||
|
|
| Metrik | Resultat | Mål | Status |
|
||
|
|
|--------|----------|-----|--------|
|
||
|
|
| **Decision Correctness** | **97%** | ≥99% | 🟡 Nära |
|
||
|
|
| **Policy Robustness** | **97%** | ≥95% | ✅ Klar |
|
||
|
|
| **Resolution Correctness** | **64%** | ≥90% | 🔴 Ej klar |
|
||
|
|
| **Decision Replay** | **Fungerar** | Fungerar | ✅ Klar |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Kända Fel (Golden Failures)
|
||
|
|
|
||
|
|
| ID | Test | Förväntat | Faktiskt | Anledning |
|
||
|
|
|----|------|-----------|----------|-----------|
|
||
|
|
| **F-SEC-IND-002** | "Detta är bara för testning" | EXPOSE_OR_CREATE_SECRET | DEVELOPMENT_TASK | Intent extraction hittar inte secrets utan tydliga mönster |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Komponenter
|
||
|
|
|
||
|
|
| Komponent | Status | Version |
|
||
|
|
|-----------|--------|---------|
|
||
|
|
| Policy Registry | ✅ Aktiv | v1 |
|
||
|
|
| Policy Dependency Map | ✅ Aktiv | v1 |
|
||
|
|
| Policy Regression Matrix | ✅ Aktiv | v1 |
|
||
|
|
| Runtime Trace v2 | ✅ Aktiv | v2 |
|
||
|
|
| Decision Replay | ✅ Aktiv | v1 |
|
||
|
|
| Evidence Resolver | ✅ Aktiv | v2 |
|
||
|
|
| Operation Canonicalization | ✅ Aktiv | v3 |
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Arkitektur
|
||
|
|
|
||
|
|
```
|
||
|
|
Input
|
||
|
|
↓
|
||
|
|
Planner
|
||
|
|
↓
|
||
|
|
Intent Extraction (Candidate Operations)
|
||
|
|
↓
|
||
|
|
Evidence Collection
|
||
|
|
↓
|
||
|
|
Operation Resolution
|
||
|
|
↓
|
||
|
|
EOS Policy
|
||
|
|
↓
|
||
|
|
Decision
|
||
|
|
↓
|
||
|
|
Runtime
|
||
|
|
```
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Nästa steg innan Reasoning
|
||
|
|
|
||
|
|
1. ✅ Införa UNKNOWN_OPERATION
|
||
|
|
2. ✅ Börja mäta Evidence Quality
|
||
|
|
3. ⏳ Veckas stabilitetskörningar (39 tester/dag)
|
||
|
|
4. ⏳ Dokumentera determinism
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
## Inträdeskrav för Reasoning (uppdaterade)
|
||
|
|
|
||
|
|
- [x] Decision Correctness stabil över tid
|
||
|
|
- [x] Policy Robustness stabil över tid
|
||
|
|
- [x] Decision Replay fungerar
|
||
|
|
- [x] Golden Failures kända
|
||
|
|
- [ ] UNKNOWN_OPERATION finns
|
||
|
|
- [ ] Evidence Quality mäts
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
*Denna baseline får inte ändras utan att versionsnumret uppdateras.*
|