Sangeetha-Grantha

Metadata Value
Status Completed
Version 1.1.0
Last Updated 2026-07-09
Author Principal Data & AI Engineering review (for Seshadri)
Priority P0 — do before any new feature work

TRACK-105: Working Tree Reconciliation & Safe Landing

Goal

Bring the repository’s dirty working tree to a clean, intentional state — every uncommitted change is either committed with a proper Ref: line or deliberately discarded. No completed work should be left at risk of loss, and no throwaway artifact should be allowed to drift into the history. This is the prerequisite for every other track: you cannot cleanly land the AI uplift or the convergence cleanup on top of a tree that already contains a finished-but-uncommitted feature.

Context — what the tree actually contains (snapshot 2026-06-06)

git status on main (up to date with origin/main) shows ~2,334 insertions / 2,097 deletions across 10 tracked files plus a pile of untracked files. These are not a single change — they are at least four distinct logical groups that became entangled because work continued without committing. The danger item is Group A: a feature already marked Completed in the registry whose code and track files are uncommitted/untracked and would be lost on a careless git checkout/git restore.

Triage table

Group Items Nature Recommended disposition
A. Indic extraction feature (TRACK-100–104) conductor/tracks/TRACK-100…104-*.md (untracked), tools/krithi-extract-enrich-worker/src/structure_parser.py (+83, modified), tests/test_structure_parser.py (+175, modified), 6 new test fixtures under tests/fixtures/structure_parser/, src/browser_batch_extract.py (untracked), conductor/tracks.md rows for 100–104 Completed feature, never committed. Coherent multi-pass parsing work with tests + fixtures. LAND. Commit as one logical unit, Ref: the TRACK-100 doc. Verify tests pass first.
B. Trinity import data (TRACK-093) database/for_import/*.csv (Dikshitar, Thyagaraja, Syama Sastri, Trinity-Test-60, raga-resolution-map, unresolved-ragas-krithis) — all heavily modified In-flight import data cleanups described in TRACK-093. LAND, but only after confirming the cleanups match TRACK-093’s documented CSV-cleanup step. Commit Ref: TRACK-093.
C. Docs & agent tooling application_documentation/sangeetha-grantha-state-of-nation-july-2026.md, …-uplift-tasks.md, conductor/prompts/*.md, .agents/workflows/agentic-prompt-optimizer.md, AGENTS.md (+2) Intentional documentation/tooling additions. LAND. Commit with doc refs. Low risk.
D. Review → RESOLVED (discarded 2026-06-06) prompt-optimizer-review-iter1.html, engineering.plugin Introspected: engineering.plugin = a Claude Code/Cowork plugin bundle (ZIP: .claude-plugin/plugin.json, GitHub MCP connector, generic engineering skills) — reproducible tooling, not app code. prompt-optimizer-review-iter1.html = a throwaway “Eval Review” benchmark output from agentic-prompt-optimizer experimentation. Neither aligns with the application’s long-term/production goals. DONE — discarded both; added *.plugin and *-review-iter*.html to .gitignore to prevent recurrence.

Note on TRACK-099: its files are already committed (25c1f5b, 6562218). The only thing wrong is the registry status line — handled in TRACK-106, not here.

Implementation Plan

Phase 1 — Verify before touching (read-only)

Phase 2 — Land Group A (the at-risk feature) as one commit

Phase 3 — Land Group B (import data) and Group C (docs)

Phase 4 — Decide Group D ✅ DONE (2026-06-06)

Phase 5 — Push & confirm parity

Acceptance Criteria

Risks

Dependencies

Progress Log

Ref: application_documentation/sangeetha-grantha-uplift-tasks.md