Model-generated daily Rasi Phalalu

interpretation.daily-rasi-phalalu-generated

An external model produces prose that is checked against deterministic Gochara facts before acceptance.

Assurance

Level	Current statement
Documented and traceable	Yes — the canonical record includes the process and a worked example.
Regression or reproduction checked	Linked to 1 repository test file(s).
Independently source-supported	Not recorded as independently verified. The disclosed provenance state remains authoritative.

Computation method

Method type: algorithm

Generate structured prose for all twelve Rashis, then reject the batch unless every cited transit matches computed facts and the text passes narrow safety and consistency checks.

Process

1. Compute the sky, Gochara verdicts, Moon house, and named conditions for all twelve Rashis.
2. Request a schema-constrained JSON array with text, advice, and explicit transit citations.
3. Require exactly twelve unique known Rashis, non-empty text/advice, and at least one citation per Rashi.
4. Normalize known English/Sanskrit Graha aliases, then require exact computed position and verdict and require the cited Graha to be named in the prose.
5. Reject intraday precision, waiting-period claims, and medical/legal/investment/contract guidance.
6. Regenerate up to three times on verification drift; retry transient HTTP errors and use the fallback model only after primary 429 exhaustion.

Worked example

Verifier treatment of an English Graha alias

Inputs

• Citation Graha: Sun
• Computed canonical Graha: Surya
• Citation position and verdict: exact engine values
• Prose explicitly names Surya

Calculation

1. Canonicalize Sun to Surya.
2. Compare position and verdict exactly.
3. Confirm a recognized Surya/Sun label occurs in the text.

Result

• Citation accepted
• Prose remains labeled model interpretation; only citation fields are engine-verified

Method notes

• The verifier is a factual/safety boundary, not a semantic proof of the prose and not evidence of classical authority.

Contract

Owning layer: interpretation
Claim kind: generated-interpretation
Time basis: Daily generated-data batch for the supplied transit snapshot.

Inputs

• date
• computed graha positions
• model response

Outputs

• verified transit citations
• generated prose

References and evidence

Evidence classes: project_heuristic
Verification states: heuristic

Cited transit triples are checked; prose and advice are explicitly generated interpretation.

daily_horoscope.llm_interpretation

Evidence class: project_heuristic
Verification state: heuristic
Locator: No precise locator is registered.
Last reviewed: 2026-07-21
Scope: The generator verifies the rasi set and every declared graha, house and verdict in transits_cited against computed gochara. The text and advice fields are model interpretation: they are screened for unsupported precision and selected high-stakes directions but are not semantically or scripturally verified. They must never inherit the authority state of their cited transit triples.

Sources

• No external source is registered for this claim.

The complete machine-readable source registry is available in provenance.json. A regression fixture or same-code reproduction is not independent verification.

Reproduce and review

• tests/test_llm_phalalu_verify.py

Run the linked tests, then the complete offline contract:

python tools/verify_project.py

Public surfaces

• website
• generated-data

Limitations

• Verification prevents unsupported transit facts but cannot make generated prose a classical or professional authority.

Implementation

• owner: telugu_panchangam/personal/llm_phalalu.py — generate_rasi_phalalu
• owner: telugu_panchangam/personal/llm_phalalu.py — _verify

This route is generated from computations.json. Edit the registry and canonical reference prose in the same pull request as a behavior change.