Researcher Dashboard
This is the atlas working surface for M. Gasuns as researcher-builder. It does
not replace .ai_state.md; it turns the current journal into a compact public
map of what is actionable, what is blocked, and which adjacent Sanskrit research
repos currently matter to the atlas.
Trust Block
- Evidence:
.ai_state.mdcurrent queue, atlas review worksheets, generated review packets, and cross-repo session journals inspected during planning. - Limitations: manually summarized v1 dashboard; it does not parse the journal, compute queue counts, or import external repo data.
- Validation: linked atlas pages are checked by
npm run build; status text should be refreshed whenever.ai_state.mdchanges materially. - Owner repo:
csl-atlas. - Next use: pick one row, open its source worksheet or repo, and keep the
resulting decision in
.ai_state.md.
Action Surface
| Lane | Current work | Source | Good next move |
|---|---|---|---|
| Human Sanskrit judgment | H4 semantic-field rows left after deterministic auto-triage | H4 worksheet | Review a small batch by sample type; record decisions before changing generators. |
| Human Sanskrit judgment | Xref shared-core source-check rows left after prefix-control auto-triage | Xref worksheet | Confirm whether exact MW/PWG edges are meaningful or only mechanical overlap. |
| Human action | H5 maker correction divaraTa -> diviraTa |
Correction proposal | Submit to makers; keep source-supported distinct rows out of correction flow. |
| External dependency | Dharmamitra gender and lemma cross-checks are model-pending | GPU runbook | Run pinned local GPU workflow, then rebuild affected queues. |
| Publication tail | P1-P6 are drafted but held for author decisions | Publications | Decide byline, venue, and missing secondary references before submission work. |
Cross-Repo Evidence To Watch
Grammar bridge
WhitneyRoots
Root/class, Whitney grammar sections, MW/Apte senses, DCS frequency, and paradigms can become lemma-dossier context later.
Corpus context
VisualDCS
The 2026 CoNLL-U import and corpus dashboard are relevant only as dictionary-facing lemma context in this atlas.
Correction caution
SanskritSpellCheck
Typo, wrong-reading, and orthographic-drift findings are ideal teaching material for "do not correct too quickly."
Source authority
MWS
Authority records, `ib.`, Register B, and source-link candidates feed atlas citation-trust questions.
Teaching contrast
CommentaryStrategies
Use as examples of how commentary explains differently from dictionaries; do not merge its pipeline into the atlas.
Interoperability
csl-standards
Keep TEI/OntoLex/loss-report work linked as a research appendix, not as the student front door.
Working Rules
| Rule | Why it matters |
|---|---|
Keep .ai_state.md authoritative |
The dashboard is a readable surface, not a second source of truth. |
| Link out before importing | Adjacent repos stay autonomous unless a dictionary-facing artifact is explicitly contracted. |
| Preserve evidence labels | Student-facing simplification must not hide observed, derived, inferred, or reviewed status. |
| Prefer small review batches | Human Sanskrit judgment is the scarce resource; batch size should make decisions easy to finish. |