med-event-takeaways

Methodology inspired by publicly taught sell-side note-writing practice; all text is an original paraphrase.

Defaults

Parameter	Default Value	Rationale
note_anatomy	9 sections	headline → per-stock → KOL → lattices → deltas → vectors → risks → trackers
delta_framing	peak × POS	Modeled deltas vs consensus are unadjusted peak × POS
conflict_policy	materiality-rated	Surface verbatim, rate materiality, or defer to an outcomes trial
vector_set	estimates/thesis/positioning	The three what's-changed vectors per covered name

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). Confirm ticker resolution via search_companies before event queries.

Triggers

• "Write the note on [conference / KOL event]."
• "What are the takeaways from [event] for my coverage?"
• "Distill the [conference] KOL panel on [drug/class]."
• "What changed for [ticker] after [event]?"
• "Model the delta to consensus from [event]'s data."
• "Which stocks move on [event] and why?"
• "Conflict check: what did KOLs disagree on at [event]?"
• "What do I update first — estimates, thesis, or positioning?"
• "Summarize the cross-trial comparisons from [event]."
• "Upside and downside risks after [event]."
• "Link the living trackers that [event] should update."

Production Grounding

• Note anatomy is fixed: headline verdict (paragraph one) → per-stock takeaways with model numbers → KOL distillation → raw data + cross-trial lattices → modeled deltas vs consensus (unadjusted peak × POS) → three what's-changed vectors (estimates / thesis / positioning) → risk bullets → living-tracker links.
• Conflict rule: conflicting views surfaced verbatim, rated by materiality ("incremental positive, not narrative-changing") or deferred to a larger outcomes trial — never silently averaged.
• Consensus delta: model-vs-consensus always explicit; POS changes logged with reasons; abstain where unquantifiable.
• Buy-side lens: client-question section, where investor attention sits, per-name bull/bear pivot conditions.
• Badges: [FACT]/[DEDUCTED]/[VIEW] with the summary table.

Data Source Priority

1. Event content: transcripts/slides via search_documents + list_sources + read_source_outline / read_source_pages.
2. Trial context: search_clinical_trials / get_clinical_trial for the studies discussed.
3. Consensus: search_earnings_calendar (company-level); product consensus derived and labeled [DEDUCTED].
4. Knowledge: search_investment_cases / search_investment_strategies / search_by_analogue (sectors=med) for pattern grounding and /v/ citations.

Methodology

Retrieval Scope

unstructured_document_search

Retrieval Strategy

1. Enumerate the event's covered names, sessions, and data presentations.
2. Retrieve transcripts/abstracts; extract per-presentation data with sources.
3. Pull each discussed trial's record for lattice construction (vs SoC and class peers).
4. Model deltas: unadjusted peak × POS per asset vs consensus, assumptions stated.
5. Ground patterns via search_investment_cases / search_by_analogue (sectors=med); cite /v/.

Temporal Scope

See frontmatter temporal_scope block. Notes cover the meeting window plus the forward quarters its data affects.

Tool Allowlist

See frontmatter allowed_tools.

Protocol

1. Event inventory
2. Data extraction (raw numbers + sources)
3. KOL distillation + conflict capture (verbatim)
4. Cross-trial lattices
5. Model deltas vs consensus
6. What's-changed vectors + risk bullets + tracker links

Modes

• Conference note (default): multi-day meeting, per-session distillation.
• KOL / site visit: single-source distillation with conflict emphasis.
• Single-name note: one covered name with deeper model treatment.
• Basket: multi-ticker note with per-name sections.

Tool Fallbacks

Failure	Fallback
Event transcript unavailable	Degrade to abstracts/slides; annotate coverage_gap
No consensus figure	State required inputs; label the delta [VIEW] — never fabricate
Trial record missing	Lattice cell marked unavailable; do not guess
Knowledge tools empty	Event data only; annotate knowledge coverage_gap

Output File

{ticker}/{YYYY-MM-DD_HHMM}_med-event-takeaways_{affix}.md (multi-name: watchlist/{YYYY-MM-DD_HHMM}_med-event-note_{affix}.md)

Output Structure

1. Headline Verdict — the event in 2-3 sentences
2. Per-Stock Takeaways — 2-4 lines each, ≥1 model number
3. KOL Distillation — themes + conflicting views verbatim with materiality ratings
4. Cross-Trial Lattices — vs standard of care and class peers
5. Model Deltas vs Consensus — peak × POS arithmetic, [DEDUCTED] labels
6. What's Changed — estimates / thesis / positioning per name
7. Risk Bullets — upside and downside
8. Living-Tracker Links — trackers to append, never rewrite

Error Handling

Error	Fallback
Conflicting KOL views	Surface both verbatim; rate materiality or defer to an outcomes trial — never average
Consensus absent	coverage_gap with required inputs; keep the internal model consistent
Event scope unclear	Narrow to named sessions/presentations; flag what is out of scope

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