conjecture-index

---

name: conjecture-index description: Index the project's open conjectures/propositions -- scans sessions + open-question registers, tabulates with priority (EVOI), origin, related tests, and status argument-hint: [--status open|closed|all] [--since ] [--top ] [--update]

/conjecture-index -- Open-Conjecture Ledger

Scans the project for stated-but-unproven conjectures, propositions, and hypotheses. Groups them by topic, attaches metadata (origin session, priority/EVOI if computed, related test IDs, current status), and emits a single ledger table.

The point: any research project of size accumulates dozens of informally stated conjectures across session minutes, working papers, and framework docs. Without a periodic sweep they stay scattered. This skill produces a canonical ledger so a planner (or /rclab-plan) can prioritize them.

Usage

/conjecture-index                    # all open conjectures, default view
/conjecture-index --status all       # include closed + withdrawn
/conjecture-index --since 80         # only conjectures stated or touched from S80 onward
/conjecture-index --top 15           # top 15 by priority (descending)
/conjecture-index --update           # rebuild the cached ledger at sessions/framework/conjecture-ledger.md

Where conjectures live

Source	What to scan for
sessions/session-N/*.md	Headings matching Conjecture, Claim, Hypothesis, Open Question, Open Problem; bullets under Open questions, Carry-forward, OPEN
sessions/framework/*.md	Any statement marked OPEN, UNCOMPUTED, PROVISIONAL
Project results / permanent registry (if present)	Status-field transitions; watch for PROVISIONAL or pending
Priority register (if present, e.g. sessions/framework/priority-register.md or an EVOI table)	The authoritative source for priority/EVOI scores
Knowledge index open_questions entities	Canonical open-question ledger: mcp__knowledge__list_entities(type="open_questions") (if no knowledge MCP, grep tools/knowledge-index.json for "type": "open_questions")
Knowledge index search	mcp__knowledge__search_knowledge("conjecture OR hypothesis OR open") -- full-text across everything (else grep the JSON index)

Execution steps

1. Parse arguments. Extract --status, --since, --top, --update from $ARGUMENTS.

2. Pull authoritative lists first (these are canonical; everything else is supplementary):

   • mcp__knowledge__list_entities(type="open_questions") -- returns all open questions with full metadata (or grep tools/knowledge-index.json).
   • Read the project's priority register / EVOI table, if one exists.

3. Scan session-minute files (widen the net). Use Grep:

   Grep pattern="^##+\s*(Conjecture|Claim|Hypothesis|Open\s+Question|Open\s+Problem)" glob="sessions/**/*.md" output_mode="content" -n=true
   

   Also scan for bullet-level markers: - **OPEN**, - **UNCOMPUTED**, - **CONJECTURE:**.

4. Deduplicate. Many conjectures are restated across sessions; collapse by topic (look for shared test IDs, shared mechanism names, or >70% token overlap in the statement).

5. Enrich each entry with the 7-column ledger schema:

   Column	Source
   id	Open-question ID if present; else a slug derived from the first statement
   statement	One-sentence summary (<=180 chars)
   origin_session	Earliest session where stated
   last_touched	Most recent session that referenced it
   related_tests	Test / gate IDs (from the knowledge index cross-ref)
   priority	Priority/EVOI value from the priority register if present; otherwise --
   status	OPEN / IN-PROGRESS / CLOSED / WITHDRAWN / PROVISIONAL

6. Apply filters (--status, --since, --top).

7. Emit the ledger as a single markdown table, sorted by:

   • Primary: priority descending (empty priority last)
   • Secondary: last_touched descending

8. If --update, write the ledger to sessions/framework/conjecture-ledger.md with a timestamp + regeneration command at the top. Otherwise print to stdout.

Output format

# Open-conjecture ledger
- Generated: <ISO date>
- Status filter: open
- Since session: (none)
- Total conjectures: 23
- Authoritative sources: knowledge index open_questions (N=15), priority register (N=18), grep-only hits (N=8)

| # | ID | Statement | Origin | Last | Tests | Priority | Status |
|--:|:---|:----------|:------:|:----:|:------|---------:|:-------|
| 1 | OQ-A-43 | ... | S43 | S80 | ... | 0.81 | OPEN |
| 2 | OQ-B-38 | ... | S38 | S83 | ... | 0.62 | OPEN |
| ... |

## Conjectures with no priority score (N=8)
These have been stated but not yet prioritized. Consider adding to the priority register if any are decision-relevant.
- `<statement>` (S74, no test link)
- ...

## Regeneration
- Cached at: `sessions/framework/conjecture-ledger.md`
- Rebuild: `/conjecture-index --update`
- Sole authority for status transitions: see `.claude/rules/gate-verdicts.md`

Guard rails

• Do NOT invent conjectures. If a bullet says "could this be X?", that is a question, not a conjecture -- skip unless the session explicitly labels it.
• Do NOT copy long statements verbatim. Summarize to <=180 chars.
• Do NOT promote a conjecture to CLOSED based on agent-memory claims. Per .claude/rules/epistemic-discipline.md + .claude/rules/gate-verdicts.md, only the canonical sources (the knowledge index's closed/results entities, gate-verdict files, synthesis docs) can close a conjecture.
• Do flag provisional/withdrawn entries distinctly -- they're useful for the "why we gave up on X" history.
• Do preserve priority/EVOI scores exactly from the priority register -- never recompute.