Entry Gate

Manual, run once per candidate name. It is cheap on purpose: it decides whether the expensive work is worth doing, and its output routes straight into the full-report mode of the market's analysis skill.

Write memory/entry-gates/<TICKER>-<YYYY-MM-DD>.json. The script computes the registered screening policy's verdict; you decide whether more research is worth it. To disagree, add research_judgment: {"verdict": "pass_to_deep_research", "rationale": "why the concern warrants investigation and what would change my view"}. assess returns your research_verdict and research_departs_from_policy alongside the original policy result. Neither result grants permission to open a position.

What the script verifies

• information.grade (A/B/C) from the evidence source classes you cite;
• quote_freshness from quote.as_of against assessed_at;
• verdict and routing — a stated verdict that disagrees with the computed one is a false claim about the policy; put a differing research choice in research_judgment;
• whether a hard veto's industry exception is actually encoded for this sector.

Vetoes are resolved before any check is counted, so checks_passed can never rescue a vetoed name.

Step 1 — quote and instrument kind

Use the workspace pipelines, never a web price:

clawock fetch-peers --help   # fetch verified quotes for an unheld name; use the documented JSON stdin request

quote.source_class must be one of the pipeline names, or validation fails outright. A quote older than 24h makes the verdict gray rather than pass.

Set instrument_kind. A leveraged ETF has no company fundamentals to grade: it takes the underlying_exposure / decay_and_regime / sizing_limit / liquidity checks and routes to leverage_look_through. When the ticker is in config/instruments.json, the registry's leverage_multiple overrules your declaration.

Step 2 — mechanism and key variables

mechanism is one sentence: who pays, why, and what makes it repeat. If you cannot write it from primary sources, that is the unintelligible_revenue_mechanism veto, not a writing problem.

Then 3–7 key_variables, each with why it decides the outcome — not a metric dump.

Step 3 — checks, each carrying evidence

Company: business_quality, moat, management_governance, valuation, dilution, downside. Every pass/fail needs at least one evidence_ids entry; unknown is the honest answer when you have not verified it, and it makes the verdict gray.

A failed business_quality, management_governance or downside (or, for a leveraged ETF, underlying_exposure or sizing_limit) is a reject. A failed moat, valuation, dilution, decay_and_regime or liquidity is not: an expensive or shallow-moat name is still worth understanding, and sizing stays with the existing risk and decision contracts.

Step 4 — the four hard vetoes

State a status for every veto in config/entry-gate-vetoes.json: integrity/governance, unintelligible revenue mechanism, persistent negative cash generation, destructive dilution. triggered requires evidence.

Two of them encode industry exceptions (clinical-stage biotech, pre-revenue infrastructure). An exception applies only when the artifact's sector matches the encoded one and the exception cites its own evidence row. Integrity and unintelligible-mechanism encode none — there is no sector where they are acceptable.

Step 5 — mirror test and research judgment

Exactly five distinct sentences: what it does, why it earns, what must stay true, what would break it, and what you would do then.

clawock entry-gate validate memory/entry-gates/<TICKER>-<date>.json
clawock entry-gate assess   memory/entry-gates/<TICKER>-<date>.json

Follow assess.research_verdict (your stated judgment, or the policy default):

• pass_to_deep_research → go run us-stock-analysis / hk-stock-analysis Mode 4 (Full Report), or the leverage look-through path for a leveraged ETF. Carry key_variables in as the questions the report must answer.
• gray_needs_evidence → next_evidence must name the question and where to look. State what research would resolve the gap.
• reject → record the research decision and its reason. Thin sourcing alone never lands here; a C grade is gray, because information richness is not investment quality.

Hard limits

• A C information grade describes the sources, never the company.
• A research disagreement preserves the veto evidence and policy verdict; do not erase a concern to justify further research.
• This gate opens no position and changes no thesis. A qualified name goes to full research; exposure still goes through the decision, risk, and settlement chain.