Skip to content

Scenario book: unexpected finding leads to new hypothesis

Goal

Write a scenario showing an unexpected result during routine analysis that leads the researcher to formulate a new exploratory question, extending the questio data model.

Context

This scenario exercises the discovery workflow — the most scientifically valuable but hardest to automate. The agent detects something anomalous during an iterate loop, flags it (rather than silently continuing), and helps the researcher develop it into a new research question.

The scenario should be written at docs/specs/research-orchestration/scenarios/scenario-unexpected-finding.md.

Pixecog context: - During SWR detection validation (sharpwaveripple flow), the agent notices a bimodal distribution of ripple durations — short ripples (~50ms) and long ripples (~120ms) - Literature (Buzsaki 2015, Fernandez-Ruiz 2019) distinguishes "single ripples" from "ripple complexes" but pixecog's current detection treats them uniformly - This could have implications for H1 (delta-ripple coupling) — do short and long ripples have different cortical correlates? - This is NOT in the current questions.yml — it's a genuinely new observation - cogpy's RippleDetector outputs duration as a feature but it's not currently used for classification - The existing H1-H7 all treat ripples as a monolithic category

Prompt

Write the scenario at /storage2/arash/projects/projio/docs/specs/research-orchestration/scenarios/scenario-unexpected-finding.md.

Step 1: Read context. - Read /storage2/arash/projects/projio/docs/specs/research-orchestration/scenario-book.md for format - Read /storage2/arash/projects/projio/docs/specs/research-orchestration/loop-mechanisms.md - Read pixecog's questions.yml for current hypothesis structure

Step 2: Write the scenario.

Structure: 1. Header with starting state: agent is in the middle of an iterate loop for SWR detection validation (mid-Scenario 1, so to speak) 2. Phase 1: The surprise — during routine SWR validation, the agent reads detection output and notices the duration histogram is clearly bimodal. Instead of silently recording "detection rate: 12.3/min, looks good" and moving on, the agent flags it: "I notice the ripple duration distribution is bimodal (peaks at ~50ms and ~120ms). This wasn't expected from the grounding phase. Want me to investigate?" 3. Phase 2: Investigation — researcher says "yes, investigate." Agent enters investigate loop: checks literature (paper_context for ripple subtypes), checks if this is known in the field (rag_query("ripple subtypes bimodal duration")), checks if it's a detection artifact (inspects waveforms for short vs long events), checks cross-subject consistency. 4. Phase 3: Literature grounding — Agent finds that ripple complexes vs single ripples is a known distinction in literature. biblio_discover_authors("Fernandez-Ruiz") finds relevant papers. biblio_graph_expand discovers more recent work on ripple subtypes. Agent synthesizes: "This is a real phenomenon, not an artifact. Literature suggests short and long ripples may have different functions." 5. Phase 4: New question formulation — Researcher says "This is interesting. It could mean that different ripple types couple differently with cortical oscillations. Let's add an exploratory question." Agent helps formulate a new question: 

Q8:
  text: "Do short and long ripples have distinct cortical coupling patterns?"
  type: exploratory
  pipelines: [sharpwaveripple, coupling_spindle_ripple]
  milestones: [ripple-subtype-classification, subtype-coupling-analysis]
6. Phase 5: Update questio data model — Agent updates plan/questions.yml with Q8, adds new milestones to plan/milestones.yml, proposes that this could modify the analysis plan for H1 (add ripple subtype as a variable). Uses propose-review-confirm. 7. Phase 6: Record and plan — Create observation notes documenting the discovery, a decision note recording the new question, and schedule follow-up analysis as worklog tasks.

Use mkdocs material admonitions: - !!! warning "Surprise detected" for the moment the agent flags the anomaly - !!! info "Behind the scenes" for tool calls - !!! tip "Why flagging matters" for explaining why silent continuation would be wrong - !!! example "Literature says..." for biblio findings - !!! note "Extending the data model" for the questio update

End with: ecosystem coverage (questio extension, biblio discovery, notio recording), loop patterns (iterate → surprise → investigate → extend questio), key insight (answer: the agent's most valuable contribution isn't running pipelines — it's noticing things a human might miss during routine validation and connecting them to literature).

Step 3: Commit with message: "Add scenario: unexpected finding leads to new hypothesis"

Acceptance Criteria

  • [ ] File at docs/specs/research-orchestration/scenarios/scenario-unexpected-finding.md
  • [ ] Shows the iterate → surprise → investigate transition
  • [ ] Agent flags the anomaly rather than silently continuing
  • [ ] Literature grounding confirms it's a real phenomenon
  • [ ] New question added to questio data model
  • [ ] Uses mkdocs material admonitions
  • [ ] Committed

Batch Result

  • status: done
  • batch queue_id: 140acc720c2b
  • session: 49cfe4a4-561d-4bb8-8fb6-d01b2ac17020
  • batch duration: 801.3s