ct-status-diff

Methodology inspired by publicly taught clinical-trial monitoring frameworks; all text is an original paraphrase.

Defaults

Parameter	Default Value	Rationale
window	latest two snapshots	The diff is always between consecutive stored states
categories	all 11	Complete change report; filter on request
risk_flags	on	Terminated / Downsized are surfaced first
gap_statement	on	Missing snapshot weeks are stated, never inferred

Preflight

Run canonical pre-flight per contracts/preflight.md. Propagate X-Agentii-Trace per contracts/x-agentii-trace-header.md. Confirm the snapshot coverage via get_ct_status_changes before framing any readout.

Triggers

• "What changed on clinicaltrials.gov for my coverage this week?"
• "Which of [ticker]'s trials moved this week?"
• "Any trial delays or accelerations in my watchlist?"
• "Flag every terminated or downsized trial since last week."
• "Did any recruitment complete for [ticker]?"
• "Show the previous and new primary-completion dates for [drug]'s program."
• "Which trials were upsized this week?"
• "Is [trial]'s enrollment collapsing?"
• "Summarize the week's trial-registry changes for the sector."
• "What should I re-check ahead of [readout] based on registry moves?"
• "Compare this week's diff to the month-ago diff."

Production Grounding

• The diff is a stored-state comparison: every row pairs a NEW value with the PREVIOUS value from the prior snapshot — the previous-value column is the discipline; a row without both is a defect.
• The 11-category taxonomy: New / Suspended / Withdrawn / Terminated / Recruiting / Recruitment-Complete / Completed / Ahead / Delayed / Upsized / Downsized.
• Risk signals: Terminated and Downsized (enrollment collapse is the earliest visible distress). Re-timing alerts: Ahead / Delayed primary-completion moves.
• Recruitment-Complete means the readout clock has started — treat it as a catalyst-date confirmation.
• Gap windows: if snapshots are not consecutive archive writes, the gap is stated on the rows — a change is never attributed across a missing week.
• Determinism: identical snapshots yield byte-identical diffs; never invent an intermediate state.
• Badge discipline: [FACT] registry rows, [DEDUCTED] interpretations, [VIEW] positioning, with the summary table.

Data Source Priority

1. get_ct_status_changes — the stored-weekly-snapshot diff (pipeline.ct_status_snapshot).
2. Trial context: search_clinical_trials / get_clinical_trial for design and endpoints.
3. Company context: search_companies / get_company_profile.
4. Historical grounding: search_investment_cases (event_type=termination/delay/safety) + search_by_analogue.
5. Knowledge: search_knowledge_entries for trial-judgment patterns.

Methodology

Retrieval Scope

structured_only

Retrieval Strategy

1. Pull the weekly diff via get_ct_status_changes (ticker-filtered for single names, sector-wide for watchlists).
2. Rank rows: risk signals first, then re-timing, then recruiting/completed, then new.
3. For each risk-signal row, pull the trial's design via get_clinical_trial and state what changed and what it implies.
4. Ground termination/delay reads in search_investment_cases / search_by_analogue (sectors=med); cite /v/ records.
5. Attach the gap-window statement verbatim wherever the tool reports one.

Temporal Scope

See frontmatter temporal_scope block. Historical diff series (month-over-month) available on request.

Tool Allowlist

See frontmatter allowed_tools.

Protocol

1. Snapshot-diff pull
2. Risk/re-timing triage
3. Per-signal trial context
4. Historical grounding
5. Weekly change report + watch flags

Modes

• Watchlist (default): one diff report across the basket, sorted risk-first.
• Single ticker: per-trial change ledger with trial context.
• Signal scan: restrict to risk signals (Terminated/Downsized) or re-timing (Ahead/Delayed).

Tool Fallbacks

Failure	Fallback
get_ct_status_changes empty (<2 snapshots)	State the coverage gap: the diff needs two consecutive snapshot cycles; use search_clinical_trials metadata as a degraded view
Trial not in snapshots	Flag as registry-coverage gap; do not fabricate a previous value
Knowledge tools empty	Proceed with structured diff only; annotate knowledge coverage_gap

Output File

{ticker}/{YYYY-MM-DD_HHMM}_ct-status-diff_{affix}.md (watchlist: watchlist/{YYYY-MM-DD_HHMM}_ct-status-diff_{affix}.md)

Output Structure

1. Executive Summary — the week in trial-registry changes, 2-3 sentences
2. Risk Signals — Terminated / Downsized rows with previous/new values and trial context
3. Re-Timing Alerts — Ahead / Delayed rows with date deltas
4. Flow Changes — New / Recruiting / Recruitment-Complete / Completed / Suspended / Withdrawn rows
5. Gap Statement — snapshot windows with missing weeks stated
6. Watch Flags — what to re-check before the next readout
7. Coverage Gaps — missing snapshots, unregistered trials, degraded modes

Error Handling

Error	Fallback
Fewer than 2 snapshots	coverage_gap with the two-cycle requirement; degraded metadata view only
Empty diff	State it plainly — a clean week is a valid result; never pad
Ticker not in med universe	Resolve via company match; surface the classification gap

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