root/Jyotisha
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

Calibrate Jyotish benchmarks and downgrade partial modules

Browse Source
This commit is contained in:
732642856
2026-06-04 03:06:43 +08:00
parent 9fd26dbea5
commit 3d2faf3380
15 changed files with 287 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files
references/technique-capability-matrix.md
+8 -6
View File
@@ -1,4 +1,4 @@
# Technique Capability Matrix — Jyotish Skill v6.0.2
# Technique Capability Matrix — Jyotish Skill v6.0.11
> Purpose: prevent vague statements like “covered” or “missing”. Each technique is classified by four layers: knowledge, computation, orchestration, and output.
@@ -21,10 +21,10 @@
| D9 / Navamsa | Yes | Yes | Yes | Yes | `covered` | `varga-full`, D9 expanded dignity. |
| D10 / Dasamsa | Yes | Yes | Yes | Yes | `covered` | `varga-full`, strict career route requires it. |
| Vimshottari Dasha | Yes | Yes | Yes | Yes | `covered` | `dasha` and `full-reading`. |
| Chara Dasha / Jaimini | Yes | Yes | Yes | Yes | `covered` | `jaimini` module with antardasha. |
| Chara Dasha / Jaimini | Yes | Partial | Yes | Partial | `partial` | Chara Karaka/Karakamsha are usable, but Chara Dasha timing is a simplified implementation. Round 7 benchmark vs PyJHora KN Rao method matched only 58/240 fields (24.17%); do not use as high-confidence timing until KN Rao/PVN Rao/Iranganti method is implemented and revalidated. |
| Karakamsha / AK | Yes | Yes | Yes | Yes | `covered` | `full-reading` uses AK, not DK. |
| Argala | Yes | Yes | Yes | Yes | `covered` | `argala` module and strict routes. |
| Shadbala | Yes | Yes | Yes | Yes | `covered` | `shadbala` module and `full-reading`. |
| Shadbala | Yes | Partial | Yes | Yes | `partial` | Internal invariants passed 1200/1200 in Round 9, and `shadbala` matches `full-reading`; however current implementation still contains simplified Nathonnata Bala, Saptavargaja approximations, Chesta speed buckets and simplified Drik weights. Use as internally consistent strength reference, not externally calibrated Parashara absolute Shadbala. |
| Ashtakavarga | Yes | Yes | Yes | Yes | `covered` | `ashtakavarga` module and validation. |
| Avastha | Yes | Yes | Yes | Yes | `covered` | `scripts/avastha_calculator.py` integrated in `full-reading`. |
| Vargottama | Yes | Yes | Yes | Yes | `covered` | App existed before; v6.0.2 adds `full-reading.modules.vargottama`. |
@@ -44,6 +44,8 @@
When producing a Technique Audit Table:
1. Do not mark a technique as missing if any layer exists.
2. Use `partial` for Bhava Chalit and Sudarshana until full traditional modules are added.
3. Use `covered` for A10, Vargottama, Pushkara, Avastha, and Dasha Sandhi from v6.0.2 onward, but still state whether the current run actually called `full-reading`.
4. If a user provides only PDF text without exact degrees, Pushkara/Vargottama/A10 may degrade to `manual` or `unavailable` because the data layer is insufficient.
2. Use `partial` for Chara Dasha/Jaimini timing, Shadbala, Bhava Chalit, and Sudarshana until full traditional modules are added or externally benchmark-validated.
3. For Jaimini output, separate reliable Karaka/Karakamsha indicators from partial Chara Dasha timing; cap timing confidence if Chara Dasha is used.
4. For Shadbala output, state that current scores are internally consistent but not yet externally calibrated to a full Parashara/JHora-style absolute Shadbala table.
5. Use `covered` for A10, Vargottama, Pushkara, Avastha, and Dasha Sandhi from v6.0.2 onward, but still state whether the current run actually called `full-reading`.
6. If a user provides only PDF text without exact degrees, Pushkara/Vargottama/A10 may degrade to `manual` or `unavailable` because the data layer is insufficient.