test: import research oracle probes

This commit is contained in:
732642856
2026-07-18 00:46:41 +08:00
parent 563b73b1ea
commit 2114129a35
19 changed files with 25055 additions and 0 deletions
@@ -0,0 +1,62 @@
# Jyotishyamitra Pinned Adapter Probe - 2026-07-18
## Status
`jyotishyamitra` is pinned as a fourth independent observation oracle only.
It is not part of the three-engine truth matrix and does not promote any claim.
## Identity
- Package: `jyotishyamitra`
- Version: `1.4.0`
- GitHub commit: `86f7eb610a66b06b3f0817d2c53355bec8b3bf8d`
- License: MIT in wheel license file
- Wheel SHA-256: `4f1a16facfa86ef01cb44de505c6de2e66b2f8d2c8b5e125c090e3f8485c6a9d`
- Python wheel tag: `py3-none-any`
- Requires Python: `>=3.7`
## Execution Boundary
- Adapter: `scripts/jyotishyamitra_adapter_probe.py`
- Tests: `tests/test_jyotishyamitra_adapter_probe.py`
- The wheel is installed with `pip --target` into a temporary directory.
- The package is invoked in an isolated subprocess with temporary `PYTHONPATH`.
- `jyotishyamitra` is not a project runtime dependency and is not imported by commercial runtime code.
- No implementation code is copied from the package.
## Canonical Request
The Steve Jobs probe records:
- birth: `1955-02-24 19:15:00`
- place: `San Francisco, CA`
- longitude: `-122.4194`
- latitude: `37.7749`
- timezone: `-8.0`
- ayanamsa: `package_default`
- node mode: `package_default`
- returnval: `ASTRODATA_DICTIONARY`
## Artifacts
- Metadata-only: `references/oracle/jyotishyamitra_pinned_adapter_probe_2026_07_18.json`
- Steve Jobs execution probe: `references/oracle/jyotishyamitra_steve_jobs_probe_2026_07_18.json`
The input contract issue was resolved: `input_birthdata()` returns string-valued input, while `generate_astrologicalData()` needs the numeric `get_birthdata()` object after `validate_birthdata() == "SUCCESS"`.
The Steve Jobs run now returns structured dictionary raw with top-level keys including `D1`, `D2`, `D4`, `D9`, `D10`, `Balas`, `AshtakaVarga`, and `Dashas`.
Replay stability:
- full raw hash: intentionally unstable because `$.Dashas.Vimshottari.current.date` embeds run time.
- normalized raw hash: stable after replacing `$.Dashas.Vimshottari.current.date` with `<volatile_run_time>`.
- normalized raw SHA-256: `b749d78f42a43aa4e5b647230a7e317602de21e3308a0a650c0e662c630546cf`
- schema SHA-256: `4ed38f03ddc8774b48501969c31c0a9ad87de3e2a4b5bc293cd86280467bc858`
- schema path count: `13524`
## Next
1. Extract D1/D2/D4/D9/D10, Shadbala/Balas, AshtakaVarga, and Vimshottari fields.
2. Compare field-by-field against local, Xalen, PyJHora, and jyotishganit as independent observation only.
3. Keep `promotion_allowed=false` unless external worked examples arbitrate method variants.
@@ -0,0 +1,42 @@
# Oracle Blocker Resolution Attempt - 2026-07-17
## Summary
This pass converts the three non-fakeable blockers into executable evidence gates.
## VedAstro hosted identity / output drift
- Added `scripts/vedastro_identity_archive.py`.
- Archived NuGet self-host candidate identity for `VedAstro.Library` `1.2.0`:
- package hash algorithm: `SHA512`
- NuGet catalog commit: `c707690b-f7bd-4813-a2cb-876e943e9667`
- license: `MIT`
- target framework dependencies include `SwissEphNet [2.8.0.2, )`
- Evidence: `references/oracle/vedastro_reproducible_identity_archive_2026_07_17.json`.
- Boundary: this fixes a reproducible NuGet package identity only. It does not prove `api.vedastro.org` runs that package, DLL, container, or method semantics.
- Status: `hosted_api_status=blocked`.
## Timing independent negative holdout
- Added `scripts/timing_negative_holdout_audit.py`.
- Audited current candidate sources:
- public positive timelines are not negative labels.
- existing 40 control dates are observed and not independently reviewed, so diagnostic only.
- Evidence: `references/real_case_calibration/timing_negative_holdout_source_audit_2026_07_17.json`.
- Output policy remains: return ranked `candidate_day_window` with `claim_status=exploratory_unvalidated`; do not label day/month candidates as verified predictions.
## Xalen / Shadbala / Ashtakavarga arbitration
- Added `scripts/xalen_arbitration_gate.py`.
- Regenerated:
- `references/oracle/xalen_formula_unit_attribution_2026_07_17.json`
- `references/oracle/xalen_public_case_batch_2026_07_17.json`
- `references/oracle/xalen_ephemeris_mode_comparison_2026_07_17.json`
- `references/oracle/xalen_formula_arbitration_gate_2026_07_17.json`
- Five public AA cases replayed.
- Independent ephemeris comparison remains ready:
- max longitude delta: `0.005305966338369217` degrees
- varga differences: `0`
- Gate result: `truth_status=blocked`, `unresolved_variant_count=46`, `promotion_allowed=false`.
- Boundary: multi-case replay and independent ephemeris reduce implementation risk; they do not arbitrate formula variants without external numeric worked examples.