Methodology inspired by publicly taught clinical-trial frameworks; all text is an original paraphrase.
Defaults
| Parameter | Default Value | Rationale |
|---|---|---|
| scrutiny_axes | all six + tolerability | Safety/stats/subgroups/missing data/endpoints/benefit-risk + tolerability/persistence as co-equal axis |
| lattice_comparators | SoC + class peers | Every readout is placed against standard of care and same-class peers |
| safety_imbalance | defer to outcomes | Small-N safety imbalances defer the verdict to a larger outcomes trial |
| conflict_policy | materiality-rated | Conflicting readings surfaced verbatim, rated by materiality or deferred |
| outcome_framing | base/bull/bear | Binary readouts need scenario sizing |
| reaction_context | historical cases | Size moves from past analogues |
Preflight
Run canonical pre-flight per contracts/preflight.md. Include the X-Agentii-Trace header on every tool call per contracts/x-agentii-trace-header.md — carry the _run_id from your first tool result and name yourself (and your parent, if you were spawned).
Triggers
- "Evaluate [ticker]'s upcoming trial readout."
- "What should I look for in [trial]'s data?"
- "Size the readout for [drug] phase 3."
- "What did the AdCom-style scrutiny say about similar trials?"
- "Base/bull/bear for [ticker]'s readout."
- "Which endpoints matter for [trial]?"
- "How has the market reacted to similar readouts?"
- "Readout checklist for [ticker]."
- "Is this trial design adequate?"
- "What are the red flags in [trial]'s design?"
Production Grounding
- Readout ≠ approval: phase-3 success is necessary but not sufficient; FDA re-analyzes sponsor data.
- Apply the six scrutiny axes (safety signals, statistical adequacy, subgroup analyses, missing data, endpoint appropriateness, benefit-risk) plus tolerability/persistence as a co-equal axis — discontinuation rates, dose reductions, and AE-driven dropout often decide commercial uptake. The 道/法 frameworks in
references/knowledge-frameworks.mdare the authoritative checklist. - Cross-trial lattice: every readout is placed against standard of care and same-class peers on aligned endpoints; a readout judged in isolation is incomplete.
- Safety-imbalance deferral: a safety imbalance seen at readout scale defers the verdict to a larger outcomes trial; never over-weight small-N imbalances.
- Materiality-rated conflicts: conflicting characterizations of the same data are surfaced verbatim and rated by materiality or deferred — never averaged.
- Readout framing: readout design, then stock sizing (binary-risk expected value), then historical analogue comparison.
Data Source Priority
search_clinical_trials/get_clinical_trial— design, status, endpoints, dates.search_documents/read_source_*— sponsor disclosure, prior data cuts.search_fda_approvals— regulatory history of the drug/program.- Knowledge layer:
search_investment_cases(event_type=trial_readout|adcom_vote)+ strategies for judgment frameworks.
Methodology
Retrieval Scope
unstructured_document_search
Retrieval Strategy
- Pull the trial record (
get_clinical_trialby NCT id, orsearch_clinical_trialsby drug/ticker). - Assess design + endpoint quality against scrutiny axes, including tolerability/persistence.
- Build the cross-trial lattice: comparator trials for SoC and class peers on aligned endpoints.
- Frame base/bull/bear outcomes with sizing; defer safety-imbalance verdicts to outcomes trials where needed.
- Ground in historical readout/adcom cases via knowledge tools.
Temporal Scope
See frontmatter temporal_scope block.
Tool Allowlist
See frontmatter allowed_tools.
Protocol
- Trial record
- Scrutiny-axes assessment (incl. tolerability/persistence)
- Cross-trial lattice vs SoC and class peers
- Outcome scenarios + sizing
- Analogue grounding
Modes
- Pre-readout (default): design scrutiny + scenario sizing.
- Post-readout: results evaluation + reaction context.
- Program view: multiple trials across a program.
Tool Fallbacks
| Failure | Fallback |
|---|---|
| search_clinical_trials empty | Use filings + press via search_documents; annotate coverage_gap |
| Trial record thin | Note undisclosed fields; do not fabricate |
| Knowledge tools empty | Proceed with structured data + static frameworks |
Output File
{ticker}/{YYYY-MM-DD_HHMM}_trial-readout-analysis_{affix}.md
Output Structure
- Executive Summary — readout stance in 2-3 sentences
- Trial Profile — design, endpoints, status, dates
- Scrutiny Assessment — the six axes plus tolerability/persistence, with evidence
- Cross-Trial Lattice — aligned endpoints vs standard of care and class peers
- Outcome Scenarios — base/bull/bear with sizing; safety-imbalance deferrals flagged
- Historical Analogues — cases with /v/ citations
- Coverage Gaps — degraded flags
Error Handling
| Error | Fallback |
|---|---|
| NCT id unknown | Search by drug/ticker; flag if unresolved |
| Endpoints undisclosed | Flag explicitly; scrutiny limited to disclosed data |
| Conflicting readings of the same data | Surface both verbatim; rate materiality or defer to a larger outcomes trial — never average |
| No comparator data for the lattice | Mark lattice cells unavailable; flag the gap — do not guess |
Memory Load
See contracts/memory-load.md.
Snapshot
See contracts/snapshot-synthesis.md.
Final Summary (TUI)
Include ### Key Citations block with 0-10 clickable /v/ URLs.
References
contracts/citation-and-memory.mdcontracts/output-frontmatter-schema.mdcontracts/memory-load.mdcontracts/snapshot-synthesis.mdcontracts/preflight.mdreferences/knowledge-frameworks.md


