trial-readout-analysis

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.md are 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

1. search_clinical_trials / get_clinical_trial — design, status, endpoints, dates.
2. search_documents / read_source_* — sponsor disclosure, prior data cuts.
3. search_fda_approvals — regulatory history of the drug/program.
4. Knowledge layer: search_investment_cases(event_type=trial_readout|adcom_vote) + strategies for judgment frameworks.

Methodology

Retrieval Scope

unstructured_document_search

Retrieval Strategy

1. Pull the trial record (get_clinical_trial by NCT id, or search_clinical_trials by drug/ticker).
2. Assess design + endpoint quality against scrutiny axes, including tolerability/persistence.
3. Build the cross-trial lattice: comparator trials for SoC and class peers on aligned endpoints.
4. Frame base/bull/bear outcomes with sizing; defer safety-imbalance verdicts to outcomes trials where needed.
5. Ground in historical readout/adcom cases via knowledge tools.

Temporal Scope

See frontmatter temporal_scope block.

Tool Allowlist

See frontmatter allowed_tools.

Protocol

1. Trial record
2. Scrutiny-axes assessment (incl. tolerability/persistence)
3. Cross-trial lattice vs SoC and class peers
4. Outcome scenarios + sizing
5. 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

1. Executive Summary — readout stance in 2-3 sentences
2. Trial Profile — design, endpoints, status, dates
3. Scrutiny Assessment — the six axes plus tolerability/persistence, with evidence
4. Cross-Trial Lattice — aligned endpoints vs standard of care and class peers
5. Outcome Scenarios — base/bull/bear with sizing; safety-imbalance deferrals flagged
6. Historical Analogues — cases with /v/ citations
7. 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.md
• contracts/output-frontmatter-schema.md
• contracts/memory-load.md
• contracts/snapshot-synthesis.md
• contracts/preflight.md
• references/knowledge-frameworks.md