CP3405DT3 · System Intelligence Platform
Prof. Dr. Tan · TR2 '26
SPRINT W3 · ROLES & RESPONSIBILITIES · Scrum does not change. The complexity of the system changes.
SCRUM + SYSTEM ROLES

Roles & Responsibilities

Scrum does not change. The complexity of the system changes. System roles are internal specialisations within the Dev Team — not replacements of the Scrum framework.

"Scrum does not change. The complexity of the system changes." — This is not a cliché. Every role on your team has two lenses: a Scrum responsibility and a system intelligence responsibility. Both must be fulfilled.
Diagram 4 — Role-System Mapping

Scrum Roles → System Responsibilities

Each standard Scrum role carries additional system intelligence responsibilities in CP3405. The Dev Team roles (R3–R10) are internal specialisations, not separate entities.

SCRUM ROLES
Product Owner
R1
Guards sprint value. Confirms Definition of Done. Owns the prediction call narrative.
↔ Sprint Goal Owner
Scrum Master
R2
Removes blockers. Protects team flow. Facilitates retrospective. Does NOT command the team.
↔ Process Health Guardian
Development Team
R3 – R10
Self-organising. Cross-functional. Builds and delivers the sprint increment. Internal specialisation below.
↔ Intelligence Pipeline Operators
SYSTEM RESPONSIBILITIES
Sprint Intelligence Lead
System Layer: Output
Defines the prediction quality bar. Approves the final locked prediction. Narrates the system story on Monday.
↔ Final release sign-off
Pipeline Integrity Lead
System Layer: Process
Ensures the pipeline runs end-to-end. Identifies where the process broke down. Proposes system improvements in retro.
↔ Process → System feedback loop
Agent + LLM Operators
System Layer: Processing
Almanac R3, Macro R4, Technical R5, Data R6, LLM R8, Calibration R10 — each owns one stage of the pipeline.
↔ Manual → Automated pipeline
All Ten Roles — Sprint W3 Responsibilities

Every Role Has a Weekend Job and a Monday Speaking Point

R1 — Product Owner
Sprint Goal Guardian
Weekend: Confirm the W3 sprint goal. Validate the Definition of Done. Make the final call on which prediction represents the team's best work.

Monday: State the sprint goal. Announce DoD status. Deliver the final W3 prediction with confidence level.
R2 — Scrum Master
Process Flow Protector
Weekend: Coordinate deadlines across the team. Identify and remove blockers. Ensure every role participates. Run a brief Sunday retrospective check.

Monday: Present the retrospective: what failed, what improved, and one specific process change for Sprint W4.
R3 — Almanac Agent
Historical Pattern Analyst
Weekend: Pull seasonal patterns, historical analogues, and calendar-based signals (Fed meetings, options expiry, earnings seasons). Produce a written analysis before LLMs are queried.

Monday: Present key seasonal insight and confidence contribution to final call.
R4 — Macro Agent
Macro-Economic Analyst
Weekend: Analyse macro environment: inflation signals, Fed language, bond yields, dollar strength, sector rotation from Yahoo 5D / Finviz. Produce written output independently.

Monday: Present the macro thesis supporting or undermining the team prediction.
R5 — Technical Agent
Chart & Pattern Analyst
Weekend: Analyse SPX, NDX, IWM charts. Identify key levels, trend structure, momentum signals, and volume patterns. Write independent technical output before LLMs.

Monday: Present chart evidence and explain why the technical picture supports (or contradicts) the prediction.
R6 — Data / Actuals Agent
Evidence & Calibration Tracker
Weekend: Record W2 actual outcomes against predictions. Calculate accuracy. Identify model disagreement vs. actual. Feed actuals into the calibration file.

Monday: Present the W2 outcome vs. prediction and calibration score update.
R7 — Human Score Analyst ⭐ Most Important
Wild Card & Override Lead
Weekend: After all agents and LLMs have reported, the R7 lead reviews the synthesis. Applies the Human Score. Decides whether to confirm, adjust, or override the AI consensus. Must justify the Wild Card observation.

Monday: Present the Human Score reasoning. This is the primary marks differentiator. Weak R7 = weak team score.
R8 — LLM Operator
Multi-Model Comparison Lead
Weekend: Query all four LLMs (ChatGPT, Claude, Gemini, DeepSeek) with identical prompts. Save raw responses. Build comparison table showing agreement, disagreement, and credibility assessment.

Monday: Present which models agreed, which diverged, and what that reveals about the prediction uncertainty.
R9 — GitHub Lead
Evidence & Release Manager
Weekend: Commit all evidence files. Ensure folder structure is clean. Create the correct release tag: vW23. Verify the release link is accessible before the Sunday deadline.

Monday: Open GitHub live in the presentation. Show the release tag. Walk through the evidence folder structure.
R10 — Calibration Lead
Accuracy & Learning Tracker
Weekend: Maintain the calibration suite. Track directional accuracy per model and per agent over time. Identify systematic biases. Feed findings into R7's Human Score input.

Monday: Present calibration trend. Are predictions improving? Which model has the best track record so far?

Monday 8 June — Who Says What

RoleSpeaking PointEvidence to Show
R1 Product OwnerSprint goal, DoD status, final W3 callPrediction file, DoD checklist
R2 Scrum MasterRetrospective: fail / improve / changeSprint board or retro notes
R3 AlmanacSeasonal pattern and calendar signalWritten Almanac output file
R4 MacroMacro thesis supporting the callMacro analysis file with sources
R5 TechnicalChart evidence and key levelsChart screenshots + technical file
R6 DataW2 outcome vs. prediction, accuracyActuals vs. predictions comparison
R7 Human Score ⭐Human Score reasoning + Wild CardHuman Score file with justification
R8 LLM OperatorAgreement, disagreement, credibilityLLM comparison table (all 4 models)
R9 GitHub LeadLive repo walk-through + vW23 tagGitHub open on screen
R10 CalibrationAccuracy trend + model biasesCalibration suite file