Harden Shadbala oracle and VedAstro range evidence
This commit is contained in:
@@ -34,6 +34,8 @@ This file is the small index for the current engineering fronts that still drive
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/oracle_benchmark_inventory_latest.json`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/oracle_closure_master_dashboard_latest.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/public_benchmark_dashboard_latest.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/shadbala_absolute_oracle_comparison_audit_2026_06_28.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/scripts/shadbala_oracle_comparison.py`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/references/oracle/`
|
||||
- Before changing oracle-dependent adjudicator or benchmark claims, run:
|
||||
- `python3 scripts/oracle_benchmark_inventory.py --format json`
|
||||
@@ -43,6 +45,7 @@ This file is the small index for the current engineering fronts that still drive
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/vedastro_parity_matrix_latest.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/vedastro_parity_matrix_latest.json`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/life_event_graph_v1_audit_2026_06_28.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/docs/research/vedastro_range_scan_allowlist_audit_2026_06_28.md`
|
||||
- `/Users/wuyongnaren/Documents/印度占星/scripts/vedastro_service_adapter.py`
|
||||
- follow only after the Jaimini marriage bridge v1 regression loop is closed
|
||||
- Use the parity matrix before adding or claiming VedAstro-equivalent capability.
|
||||
|
||||
@@ -0,0 +1,52 @@
|
||||
# Shadbala Absolute Oracle Comparison Audit - 2026-06-28
|
||||
|
||||
## Scope
|
||||
|
||||
This pass promotes the local Shadbala absolute-value comparison helper from an untracked fragment into a formal oracle-closure asset.
|
||||
|
||||
## New Entrypoint
|
||||
|
||||
- `/Users/wuyongnaren/Documents/印度占星/scripts/shadbala_oracle_comparison.py`
|
||||
- Primary function: `compare_case(oracle_file, case_id)`
|
||||
- CLI example:
|
||||
|
||||
```bash
|
||||
python3 scripts/shadbala_oracle_comparison.py \
|
||||
--oracle-file references/oracle/dasha_shadbala_oracle_cases.json \
|
||||
--case-id template_steve_jobs_dasha_lahiri \
|
||||
--format json
|
||||
```
|
||||
|
||||
## Output Contract
|
||||
|
||||
The report exposes:
|
||||
|
||||
- `scope = shadbala_absolute_oracle_comparison`
|
||||
- `case_id`
|
||||
- external oracle `status`
|
||||
- birth and settings metadata
|
||||
- per-planet `oracle_total_rupa`, `local_total_rupa`, and `diff_total_rupa`
|
||||
- per-component Rupa deltas
|
||||
- `global_scaling_check`
|
||||
- tolerance metadata
|
||||
|
||||
## Boundary
|
||||
|
||||
- This is diagnostic oracle evidence.
|
||||
- It does not authorize global Shadbala scaling.
|
||||
- It does not tune constants from a single case.
|
||||
- It preserves component-level deltas so future fixes can target the actual failing sub-strength instead of hiding mismatch inside a total score.
|
||||
|
||||
## Current External-Verified Case Snapshot
|
||||
|
||||
For `template_steve_jobs_dasha_lahiri`:
|
||||
|
||||
- `planet_count = 7`
|
||||
- `planets_within_total_tolerance = 5`
|
||||
- `max_abs_total_delta_rupa = 2.085`
|
||||
|
||||
This is useful enough to keep in the main repo, but not sufficient to claim Shadbala oracle closure.
|
||||
|
||||
## Regression Coverage
|
||||
|
||||
- `/Users/wuyongnaren/Documents/印度占星/tests/test_shadbala_oracle_comparison.py`
|
||||
@@ -0,0 +1,48 @@
|
||||
# VedAstro Range Scan Allowlist Audit - 2026-06-28
|
||||
|
||||
## Scope
|
||||
|
||||
This pass hardens the VedAstro range-scan adapter so external `EventsAtRange`-style responses do not enter the local evidence ledger unfiltered.
|
||||
|
||||
## Behavior
|
||||
|
||||
`scripts/vedastro_service_adapter.py` now filters range-scan events through a per-domain allowlist before writing `evidence_ledger`.
|
||||
|
||||
Supported domains:
|
||||
|
||||
- `marriage`
|
||||
- `wealth`
|
||||
- `career`
|
||||
|
||||
Each domain can accept an event by either:
|
||||
|
||||
- known event id, or
|
||||
- at least one relevant tag.
|
||||
|
||||
## Boundary
|
||||
|
||||
- The allowlist only filters external candidate evidence.
|
||||
- It does not promote VedAstro events into final conclusions.
|
||||
- It does not change local Dasha, Narayana, Jaimini, Varga, Shadbala, Ashtakavarga, or strict workflow scoring.
|
||||
- External range evidence remains secondary/oracle evidence until a domain adjudicator explicitly promotes it through tests.
|
||||
|
||||
## Contract Exposure
|
||||
|
||||
`--print-schema` now exposes:
|
||||
|
||||
```json
|
||||
{
|
||||
"range_scan_event_allowlist": {
|
||||
"marriage": {
|
||||
"event_ids": [],
|
||||
"tags": []
|
||||
}
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
This prevents future agents from treating the filtering layer as hidden behavior.
|
||||
|
||||
## Regression Coverage
|
||||
|
||||
- `/Users/wuyongnaren/Documents/印度占星/tests/test_vedastro_service_adapter_executor.py`
|
||||
Reference in New Issue
Block a user