docs: PARKED Zacks MCP capability note — no research authorization

[Trujillofa]

Summary

• Rewrites docs/specs/zacks-mcp-listed-equity-overlay-v0.md as a PARKED capability / data-availability note.
• Posture: PARKED capability note — no research authorization.
• The crypto research program and public-data book remain sealed (June 23 capstone).
• Acquiring dated / PIT data does not produce DATA_PASS and does not authorize a probe, sweep, extract ingestion, or overlay.

What this is / is not

• Is: a record that a generally licensed Zacks MCP exists, plus schema facts and must-nots.
• Is not: a research-lane spec, Gate-0 brief, KEEP path, paper/live path, or agent re-arm.

Reopening (all five required)

1. Explicit human authorization
2. A separately scoped new program
3. A specific named differentiated / non-public advantage (not access to a generally licensed dataset)
4. Proof the thesis is distinct from the closed mNAV / equity-fundamentals lane, including the prior MSTR failure
5. A complete Gate-0 brief before code, probe, sweep, or extract ingestion

A future Gate-0 brief must preregister the null, multiple-testing correction, concentration limit, expected regime/failure, symbol/timeframe, trade density, independence, trend-filter choice, and validation plan.

Restrictions preserved

• Current MCP holdings are not historical backtest inputs
• No live extracts or tokens committed
• No agent re-arm; no paper/live path
• Desk snapshots must not leak into later research inputs

Test plan

• Docs only; no code, config, dependency, extract, or MCP credential changes
• File cannot be read as automatically unblocking after data acquisition
• Terminal research seal preserved

Docs only. No research execution, market-data backtests, paper/live changes, merge, or deploy from this PR.

[Trujillofa]

Review: no findings, ready to merge

Docs-only, one new file, 68 lines. CI green, mergeable_state: clean.

Checked it against the standing program state in CLAUDE.md and docs/reports/research-consolidation-2026-06-19.md, which is where a new lane spec is most likely to go wrong right now, and it lands on the right side of every line:

• Doesn't reopen the closed surface. The consolidation closed OHLCV structural probes on majors and said the next primitive should be data-first. This is an exogenous listed-equity overlay, explicitly "not a perp OHLC family and not CVD" — a different instrument class, not a closed lane in new wrapping.
• No probe opened, no Gate 0/1 consumed. It stops at a spec and stays BLOCKED, which is the correct state given the finding that drives it: current MCP holdings carry no as-of history, so the question is unmeasurable today. Writing that down instead of probing anyway is the whole point.
• DATA_PASS is falsifiable and pre-registered — dated holdings or ≥10y PIT statements, a separate perp cost book, and a written overlay contract naming the host family. That's a gate someone can fail, not a vibe.
• The "must not" list covers the real failure modes, including the subtle one: treating today's holdings print as dated flow. That's exactly how this kind of feed leaks lookahead into a backtest.
• Correctly stays out of docs/reports/autoresearch-candidate-ledger.md — per CLAUDE.md the ledger is for final promotion/reject decisions only, and this is neither.
• No tokens or live extracts committed; the MCP snippet is a bare URL pointed at ~/.cursor/mcp.json with an explicit "do not commit project .mcp.json".

docs/specs/ is the right home and the -v0 naming matches the neighbours.

One thing to carry forward rather than change here: the spec authorizes "desk-only current snapshots with attribution", and that's the one activity that can quietly turn into a backtest input later. Whenever the licensed dated extract does arrive, the first thing to confirm is that no desk snapshot from the interim leaked into a config — the spec forbids it, but nothing enforces it yet.

---

Generated by Claude Code