Audit dignity guardrail v1 boundary
This commit is contained in:
@@ -36,3 +36,9 @@ This file is the small index for the current engineering fronts that still drive
|
||||
|
||||
- `/Users/wuyongnaren/Documents/印度占星/scripts/vedastro_service_adapter.py`
|
||||
- follow only after the Jaimini marriage bridge v1 regression loop is closed
|
||||
|
||||
## Dignity / Role Guardrails
|
||||
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/superpowers/specs/2026-06-28-dignity-guardrail-v1-design.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/dignity_guardrail_v1_boundary_audit_2026_06_28.md`
|
||||
- D1-only dignity guardrail is now landed; divisional dignity repair and a separate functional-role guardrail remain open fronts.
|
||||
|
||||
@@ -0,0 +1,61 @@
|
||||
# Dignity Guardrail v1 Boundary Audit - 2026-06-28
|
||||
|
||||
## What Landed
|
||||
|
||||
- `dignity_guardrail` has been added to `relationship` and `finance` strict workflows.
|
||||
- The guardrail reads only D1 `chart.planets.status`.
|
||||
- Only route-relevant planets may affect the guardrail result.
|
||||
- Score impact is bounded to a single `-5 | 0 | +5`.
|
||||
- The previous whole-chart dignity broadcast has been removed from `_derive_event_judgement`.
|
||||
|
||||
## What Did Not Land
|
||||
|
||||
- No D9 dignity scoring
|
||||
- No D10 dignity scoring
|
||||
- No Vimsopaka dignity scoring
|
||||
- No `dominant_label` lifting from dignity
|
||||
- No merge with Functional Benefic/Malefic logic
|
||||
- No multi-planet cumulative dignity scoring
|
||||
|
||||
## Verified Boundary
|
||||
|
||||
The current implementation now enforces these behaviors:
|
||||
|
||||
1. non-relevant planets are ignored for route scoring
|
||||
2. mixed relevant `NEECHA_BHANGA` and `GREAT_ENEMY` signals collapse to `conflict`
|
||||
3. `conflict` does not directly change event labels
|
||||
4. `relationship` and `finance` retain their pre-existing label logic
|
||||
5. dignity only modifies:
|
||||
- `event_judgement.score`
|
||||
- `event_judgement.secondary_context`
|
||||
- `confidence_cap` when a conflict is present
|
||||
|
||||
## Why This Is Safer Than The Previous State
|
||||
|
||||
The previous implementation scanned all natal planets and added or subtracted score directly from chart status strings. That allowed unrelated dignity states to alter relationship or finance verdicts.
|
||||
|
||||
The new guardrail restores strict-workflow discipline by making dignity:
|
||||
|
||||
- domain-aware
|
||||
- bounded
|
||||
- auditable
|
||||
- non-authoritative
|
||||
|
||||
## Known Follow-Ups
|
||||
|
||||
1. Repair divisional dignity data-flow separately in `scripts/jyotish_engine.py`
|
||||
- current divisional dignity calculations still need an isolated follow-up review
|
||||
2. Build `functional_role_guardrail`
|
||||
- this should consume the already existing Functional Benefic/Malefic layer
|
||||
3. Decide later whether `GREAT_FRIEND` should receive context-only semantics
|
||||
4. Consider whether a route-specific audit table row should be emitted for dignity guardrail status
|
||||
|
||||
## Verification
|
||||
|
||||
Executed:
|
||||
|
||||
- `python3 -m pytest tests/test_mcp_strict_workflow_relationship.py tests/test_mcp_strict_workflow_finance.py -q`
|
||||
|
||||
Result:
|
||||
|
||||
- `25 passed`
|
||||
Reference in New Issue
Block a user