Foundation
One focused full assignment with a supplied neighborhood; approximately 3–4 prepared-learner hours or 5–8 newcomer hours.
THE ANALYST / INSTRUCTOR PLANNING DESK
RETURN TO LEARNER WORKBENCHTHE ANALYST / ASSIGNMENT STUDIO + PRIORITY BRIEF ROTATION
Plan a progression through the work.
This desk helps instructors sequence nine workplace assignments and sixteen compact Priority Briefs across SQL, Python, data judgment, and professional handoff. It contains no sample conclusions. Prepared-hour estimates assume a learner can work independently; newcomer ranges assume active instructor support.
INSTRUCTOR QUICK STARTTHREE STEPS
- 01
Choose the work. Assign one full assignment or a compact Priority Brief.
- 02
Set the support. Select Supported, Guided, or Independent mode, then share that workbench URL.
- 03
Review the handoff. Collect the exported submission and assess it with the judgment rubric.
WORKLOAD BANDSNOT GRADES
Investigation
Several assets and competing explanations; approximately 6–12 hours.
Decision
Multi-stage evidence and several professional artifacts; approximately 10–18 hours.
Practicum
End-to-end build, audit, forecasting, or deployment work; approximately 16–30+ newcomer hours.
COURSE ROUTESHONEST LOAD, NOT COVERAGE THEATER
One semester; mixed-experience cohort with weekly instruction and supported lab time.
- ASSIGNMENTS
- 01 Monday Scorecard · 02 Quarter That Moved · 03 Navigation Vote · 04 Rollback Before Dawn · one advanced elective
- WORKLOAD
- Approximately 30–51 prepared-learner hours or 51–84 newcomer hours, plus instruction and feedback.
- Weeks 1–2: preflight, SQL/Python diagnostic, and a supported Scorecard.
- Weeks 3–5: Quarter investigation; require grain and cutoff review before polishing.
- Weeks 6–9: Navigation experiment with a structured statistics checkpoint.
- Weeks 10–12: Rollback incident in pairs or evidence-lane teams.
- Weeks 13–15: choose FO, PR, NL, MR, or SP to match course outcomes; week 16 is review and revision.
Six-month program; repeated SQL/Python practice, staged feedback, and a larger final practicum.
- ASSIGNMENTS
- 01–05 core · one of 06 Stock / 08 Queue · one of 07 Orion / 09 Model Audit
- WORKLOAD
- Approximately 52–83 prepared-learner hours or 89–138 newcomer hours, depending on electives.
- Weeks 1–4: diagnostic, Scorecard, and recovery week for novice learners.
- Weeks 5–8: commercial data certification and individual feedback conference.
- Weeks 9–12: experimentation and incident decision; switch from supported to guided mode.
- Weeks 13–16: point-in-time modeling and an explicit deployment-boundary review.
- Weeks 17–22: team practicum plus independent red-team audit; weeks 23–24 are revision and handoff.
Prepared learners with substantial outside-work time; not the default promise for a semester.
- ASSIGNMENTS
- All nine assignments in sequence, with 06/08 delivered as teams and 07/09 as independent audits.
- WORKLOAD
- Approximately 74–115 prepared-learner hours or 127–192 newcomer hours before lectures, debugging, and revision.
- Budget one protected recovery/revision week after every two or three assignments.
- Do not compress both 06 Stock and 08 Queue into ordinary homework-sized windows.
- Use checkpoints to stop foundational grain or time errors from propagating into the final handoff.
Hours are investigation time, not a promise of mastery. Add explicit instruction, debugging, feedback, and revision time. A 16-week course should normally choose an advanced elective; a 24-week course can sustain two. All nine is an intensive studio.
RECOMMENDED SEQUENCEREFERENCE ORDER / CHOOSE A ROUTE ABOVE
| # | Assignment | Band | Prepared | Newcomer | SQL | Python | State |
|---|---|---|---|---|---|---|---|
| 01 | The Monday ScorecardCustomer Insights Analyst | Foundation | 3–4h | 5–8h | 2/5 | 2/5 | connected |
| 02 | The Quarter That MovedCommercial Data Transition Analyst | Investigation | 5–8h | 8–12h | 3/5 | 3/5 | connected |
| 03 | The Navigation VoteProduct Experimentation Analyst | Decision | 8–12h | 12–18h | 3/5 | 4/5 | connected |
| 04 | Rollback Before DawnConnected Reliability Analyst | Decision | 6–9h | 10–16h | 4/5 | 4/5 | connected |
| 05 | The 7:30 Capacity CallService Capacity Analyst | Decision | 10–16h | 18–26h | 4/5 | 5/5 | connected |
| 06 | Forty-Eight Hours of StockSupply Planning Data Scientist | Practicum | 12–18h | 20–30h | 4/5 | 5/5 | connected |
| 07 | The Orion RenewalSenior Operations Analyst | Decision | 10–16h | 18–28h | 5/5 | 4/5 | connected |
| 08 | The Queue Nobody OwnsApplied ML Analyst | Practicum | 12–18h | 20–30h | 4/5 | 5/5 | connected |
| 09 | Too Good to ShipModel Risk Analyst | Practicum | 8–14h | 16–24h | 3/5 | 5/5 | connected |
PRIORITY BRIEF ROTATION / 16WEEKLY PRACTICE / EXISTING MERIDIAN TABLES
Priority Briefs are bounded 60–120 minute workplace decisions—not attendance check-ins. Each opens a separate local workbench, reuses an existing governed data neighborhood, requires multiple evidence moves and a polished handoff, and leaves the corresponding full-assignment draft untouched. The sixteen-brief weekly sequence maps directly to a 16-week term and then repeats; permanent URLs let instructors substitute or assign any brief outside the featured window. Debriefs describe defensible approaches and common traps rather than one required answer.
Two satisfaction figures—7.6 and 3.8—are headed for the same executive review. Reconcile their scales, populations and coverage before choosing the headline.
- ROLE
- Performance Reporting Analyst
- TIMEBOX
- 60–75 minutes
- ROTATION
- Slot 1 of 16
- DATA
- 3 tables from The Monday Scorecard
- Governed metric table
- Coverage check
- Executive scorecard note
Two Q2 totals and a fulfillment workbook disagree across tables with different grains and clocks. Decide what Finance can certify today.
- ROLE
- Commercial Operations Analyst
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 2 of 16
- DATA
- 5 tables from The Quarter That Moved
- Grain map
- Close control
- Clock exception view
- Controller handoff
Device readings look calmer as telemetry coverage falls during a storm and firmware change. Decide whether the apparent recovery can narrow containment.
- ROLE
- Reliability Duty Analyst
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 3 of 16
- DATA
- 6 tables from Rollback Before Dawn
- Complete-denominator profile
- Rival-evidence view
- Incident decision card
A navigation experiment reports more pages per session. Decide whether that engagement signal and its commercial guardrails support rollout.
- ROLE
- Experiment Review Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 4 of 16
- DATA
- 5 tables from The Navigation Vote
- Assignment audit
- Effect and guardrail view
- Release-council slide
A workbook claims several warehouse-product pairs have less than 48 hours of cover. Audit the demand signal and choose the constrained actions worth taking now.
- ROLE
- Inventory Planning Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 5 of 16
- DATA
- 8 tables from Forty-Eight Hours of Stock
- Demand correction
- Uncertainty view
- Action portfolio
A retention model posts extraordinary validation results. Determine which features knew the future before a pilot reaches customers.
- ROLE
- Model Validation Analyst
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 6 of 16
- DATA
- 4 tables from Too Good to Ship
- Submitted-result reproduction
- Predictor quarantine
- Governance disposition
Customer text may improve routing, but the corpus contains later messages, internal notes and synthetic PII. Define the useful data boundary without exporting the risk.
- ROLE
- Data Responsibility Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 7 of 16
- DATA
- 5 tables from The Queue Nobody Owns
- Corpus contract
- Coverage and equity audit
- Handoff disposition
- Review-capacity plan
A vendor chart claims technician productivity rose 12% after ORION-2. Reconstruct the metric and replace the board message with what the evidence can support.
- ROLE
- Strategy Analytics Partner
- TIMEBOX
- 60–90 minutes
- ROTATION
- Slot 8 of 16
- DATA
- 6 tables from The Orion Renewal
- Metric reconstruction
- Chart audit and replacement
- Board message
The morning roster, external signals and staffing records do not share one clean decision clock. Reconstruct what dispatch could know before authorizing scarce interventions.
- ROLE
- Dispatch Control Analyst
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 9 of 16
- DATA
- 8 tables from The 7:30 Capacity Call
- As-of packet audit
- Corrected dispatch view
- Capacity-checked action ledger
- Dispatch incident note
A discounting explanation depends on an effective-dated price book that covers fewer channels than the transaction ledger. Separate observed pricing from mapping assumptions and mix.
- ROLE
- Pricing Operations Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 10 of 16
- DATA
- 5 tables from The Quarter That Moved
- Governed pricing fact
- Price-book coverage matrix
- Price-volume-mix bridge
- Pricing disposition
A proactive inspection policy mixes point-in-time health signals with delayed outcomes and current-state fields. Rebuild the decision clock before allocating 500 inspection slots.
- ROLE
- Reliability Program Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 11 of 16
- DATA
- 6 tables from Rollback Before Dawn
- Point-in-time cohort contract
- Retrospective policy evaluation
- Inspection eligibility rule
- Reliability program decision
A product KPI reconciles physically but may not mean what its label claims. Audit the event taxonomy before the metric becomes a permanent dashboard contract.
- ROLE
- Product Analytics Engineer
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 12 of 16
- DATA
- 5 tables from The Navigation Vote
- Session-event reconciliation
- Event taxonomy contract
- KPI comparison
- Dashboard migration note
A supplier scorecard begins from receipts and rewards first arrivals. Rebuild it from obligations before selecting five vendors for remediation.
- ROLE
- Supplier Performance Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 13 of 16
- DATA
- 6 tables from Forty-Eight Hours of Stock
- Receipt-obligation fact
- Supplier scorecard
- Mix sensitivity
- Remediation slate
Orchestration status, SLA performance and data-quality evidence disagree at different grains. Build the release gate before closing the incident.
- ROLE
- Analytics Reliability Engineer
- TIMEBOX
- 75–105 minutes
- ROTATION
- Slot 14 of 16
- DATA
- 3 tables from Too Good to Ship
- Run-level control matrix
- Pipeline incident Pareto
- Composite release gate
- Reliability handoff
A clean resolution clock hides reopen cycles, unresolved tickets and missing ownership history. Decide what the operating review may publish.
- ROLE
- Service Process Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 15 of 16
- DATA
- 4 tables from The Queue Nobody Owns
- Ticket-state reconstruction
- Resolution-definition comparison
- Attribution ruling
- Operations review artifact
Field Operations wants standard pre-call parts kits. Turn historical parts usage into a bounded pilot without confusing recorded use, availability, geography or causal impact.
- ROLE
- Service Parts Analyst
- TIMEBOX
- 90–120 minutes
- ROTATION
- Slot 16 of 16
- DATA
- 6 tables from The Orion Renewal
- Grain-safe service-parts fact
- Candidate kit matrix
- Held-out policy evaluation
- Pilot operating plan
The Monday Scorecard
Reconcile two conflicting satisfaction figures before the executive review.
- SQL CORE
- Profile grains, scales, cohorts, duplicate interactions, and response coverage.
- PYTHON CORE
- Reproduce the profile in Pandas, visualize scale/cohort differences, and create the scorecard artifact.
- PREREQUISITES
- Tables and data types · Basic SELECT/GROUP BY · Pandas introduction
- HANDOFF
- 4 professional artifacts · core-analysis runtime
SQL2
Python2
Data complexity2
Statistical reasoning1
Decision ambiguity3
Deliverable load2
The Quarter That Moved
Certify Q2 orders, revenue, and fulfillment timing after an acquisition cutover.
- SQL CORE
- Construct a stable order fact across headers, lines, events, and shipments.
- PYTHON CORE
- Profile collisions and clock errors, build reusable exception flags, test reconciliations, and certify an extract.
- PREREQUISITES
- Python functions · Pandas joins · SQL joins and aggregation
- HANDOFF
- 5 professional artifacts · core-analysis runtime
SQL3
Python3
Data complexity3
Statistical reasoning2
Decision ambiguity3
Deliverable load3
The Navigation Vote
Determine whether a mobile-navigation experiment warrants rollout.
- SQL CORE
- Build one row per assignment and reconcile exposure, sessions, events, and orders.
- PYTHON CORE
- Run balance checks, estimate effects and uncertainty, inspect distributions, and separate confirmatory from exploratory slices.
- PREREQUISITES
- Probability and sampling · Confidence intervals · SQL cohort construction
- HANDOFF
- 5 professional artifacts · statistics runtime
SQL3
Python4
Data complexity3
Statistical reasoning4
Decision ambiguity4
Deliverable load4
Rollback Before Dawn
Recommend global rollback, scoped containment, or monitored continuation during a storm.
- SQL CORE
- Build point-in-time asset, telemetry, alert, weather, and field-operation evidence lanes.
- PYTHON CORE
- Diagnose missingness, compare regions and periods, quantify counterevidence, and visualize decision thresholds.
- PREREQUISITES
- Missing-data mechanisms · Time-aware joins · Exploratory visualization
- HANDOFF
- 5 professional artifacts · statistics runtime
SQL4
Python4
Data complexity5
Statistical reasoning4
Decision ambiguity5
Deliverable load4
The 7:30 Capacity Call
Build a morning risk view for appointments likely to miss their service window.
- SQL CORE
- Create a point-in-time appointment feature mart at a stable entity and scoring time.
- PYTHON CORE
- Build sklearn pipelines, compare a baseline, use forward validation, inspect calibration/errors, and design an intervention threshold.
- PREREQUISITES
- Supervised learning · Feature pipelines · Temporal validation · Classification metrics
- HANDOFF
- 7 professional artifacts · modeling runtime
SQL4
Python5
Data complexity4
Statistical reasoning4
Decision ambiguity4
Deliverable load5
Forty-Eight Hours of Stock
Forecast 21-day SKU/warehouse risk and recommend transfers, expedites, or substitutions.
- SQL CORE
- Reconcile movement, demand, receipt, transfer, and vendor facts into a complete daily spine.
- PYTHON CORE
- Backtest honest baselines, forecast intermittent demand, simulate lead-time risk, and produce a constrained action file.
- PREREQUISITES
- Time-series validation · Forecast error · Simulation · Operational constraints
- HANDOFF
- 8 professional artifacts · statistics runtime
SQL4
Python5
Data complexity5
Statistical reasoning5
Decision ambiguity5
Deliverable load5
The Orion Renewal
Audit a claimed 12% optimizer gain and make a procurement recommendation.
- SQL CORE
- Define stable route, stop, work-order, visit, and workforce outcomes without fanout.
- PYTHON CORE
- Plot trends and heterogeneity, run sensitivity/placebo analyses, and build the board-ready evidence package.
- PREREQUISITES
- Metric design · Join reconciliation · Quasi-experimental reasoning
- HANDOFF
- 5 professional artifacts · statistics runtime
SQL5
Python4
Data complexity5
Statistical reasoning5
Decision ambiguity5
Deliverable load5
The Queue Nobody Owns
Design a safe shadow-routing system for uncategorized support work.
- SQL CORE
- Construct leak-resistant conversation-level labels and train/validation cohorts.
- PYTHON CORE
- Build TF-IDF and comparison pipelines, inspect class errors, add abstention, package inference, and define monitoring.
- PREREQUISITES
- Text representation · Classification · Error analysis · Model packaging
- HANDOFF
- 8 professional artifacts · nlp runtime
SQL4
Python5
Data complexity4
Statistical reasoning4
Decision ambiguity5
Deliverable load5
Too Good to Ship
Audit an implausibly strong cancellation model and decide its smallest safe path forward.
- SQL CORE
- Trace model lineage, snapshot grain, labels, and point-in-time availability.
- PYTHON CORE
- Reproduce leakage, compare grouped temporal validation, build an allow-listed baseline, and audit calibration and segments.
- PREREQUISITES
- Model validation · Temporal splits · Calibration · Data governance
- HANDOFF
- 5 professional artifacts · modeling runtime
SQL3
Python5
Data complexity5
Statistical reasoning5
Decision ambiguity5
Deliverable load5
CLASSROOM PREFLIGHTRUN BEFORE WEEK ONE
Current laptop or desktop browser. Phones are for brief preview only; tablets should be tested before the course.
Run a live SQL and Python smoke test in the institution’s managed browser image before the first class.
The first workbench load and Python runtime/package setup require reliable network access; allow time for institutional filtering and caches.
Work is local to the browser profile. Private browsing, clearing site data, device replacement, and some managed-lab resets can erase it.
Have learners export a submission at each milestone and move it to the institution’s LMS or approved storage. The site is not an LMS.
Use only the fictional Meridian estate. Do not paste employer, client, patient, customer, or otherwise protected data into the workbench.
Operating and assessment boundaries
- Treat reveals as responses to evidence or timed fallbacks, not hints that dictate one query path.
- Grade the submitted evidence trail and handoff, not time-on-task, query count, model complexity, or resemblance to an instructor conclusion.
- A responsible refusal must name the unsupported boundary, show proportionate evidence, deliver the useful remainder, and propose the smallest ethical next step.
- Machine records are mechanical observations only. At present the site captures successful browser runs with executed-code hashes, flags output that predates later worksheet edits, hashes exported workspace files, and records explicitly bound non-empty artifact files. It does not certify grain, point-in-time validity, inference, artifact quality, or recommendation quality.
- Use the LMS or institutional process for identity, deadline enforcement, accommodations, authorship, collaboration rules, and durable submission retention.
100-POINT JUDGMENT RUBRICHUMAN REVIEW
ONE RUBRIC / MULTIPLE DEFENSIBLE OUTCOMES
Score the evidence and decision, not resemblance to an answer key.
For each dimension, select level 0–4. Points equal the dimension weight multiplied by the level and divided by four. When evidence falls between levels, use the lower level and state what was missing. A qualified conclusion or responsible refusal can earn full credit.
01Business grain and relationship controlDid the learner define what one row represents and control multiplicity?15 PTS
LEVEL 0
Treats physical rows as business entities or events in a way that materially invalidates the result; ignores obvious join multiplication.
LEVEL 1
Names a grain or duplicate concern but does not test it or quantify its effect.
LEVEL 2
Uses a plausible grain and some key checks, but leaves a material relationship, many-to-many path, or reconciliation unresolved.
LEVEL 3
Defines the business grain, profiles key uniqueness, respects relationship cardinality, and reconciles row/entity counts before and after joins.
LEVEL 4
Meets level 3 and stress-tests alternate grains, temporal relationships, partial coverage, or legitimate repeats; explains whether residual grain risk changes the decision.
02Point-in-time validityCould the evidence legitimately have been known at the decision cutoff?15 PTS
LEVEL 0
Uses future outcomes, revised current state, or post-cutoff information as if it were available at decision time.
LEVEL 1
Applies date filters but leaves the analysis cutoff, event-time meaning, or comparison window materially ambiguous.
LEVEL 2
Defines the cutoff and main periods but misses one relevant lag, backfill, revision, snapshot, or temporal-join issue.
LEVEL 3
Makes the cutoff explicit; uses coherent event and availability time; reconstructs historical state where needed; and aligns comparison windows to the decision.
LEVEL 4
Meets level 3 and evaluates sensitivity to availability lag, boundary choices, backfill, temporal coverage, or changing definitions; explains the decision effect.
03Evidence quality and validationIs the evidence relevant, diagnostic, and strong enough for the proposed decision?10 PTS
LEVEL 0
Provides no auditable support, relies on an unrelated metric, or materially misreads the data.
LEVEL 1
Provides a headline result with little validation or treats a catalog reliability label or recorded run as universal certification.
LEVEL 2
Provides relevant evidence and basic diagnostics but leaves a material alternative explanation or data-quality issue untested.
LEVEL 3
Triangulates the main claim, tests decision-relevant data quality, and separates observed facts from assumptions and interpretation.
LEVEL 4
Meets level 3 and actively tests credible rival explanations, negative controls, sensitivity, or robustness; identifies evidence that would change the conclusion.
04Reproducibility and lineageCan another analyst recover the population, transformations, result, and diagnostics?15 PTS
LEVEL 0
Relies on unrecorded manual work, non-executable code, or a result that cannot be connected to source assets.
LEVEL 1
Provides fragments or screenshots but cannot reproduce the reported result end to end.
LEVEL 2
Core code runs, but parameters, exclusions, metric definitions, or validation steps require reconstruction.
LEVEL 3
Executable work reproduces the result and key checks; source assets, cutoff, population, and transformations are explicit.
LEVEL 4
Meets level 3 and adds readable structure, parameters, assertions, deterministic outputs, and concise lineage sufficient for maintenance or peer review.
05Uncertainty calibrationDoes the learner’s confidence match the design, data, and stakes?15 PTS
LEVEL 0
Makes a categorical or causal claim that the evidence cannot support, or conceals decision-material uncertainty.
LEVEL 1
Appends generic caveats without connecting them to the estimate, claim, or action.
LEVEL 2
Identifies material limitations but does not explain their likely direction, magnitude, or consequence.
LEVEL 3
Calibrates language to evidence; distinguishes descriptive, predictive, and causal claims; and explains which uncertainty matters for the decision.
LEVEL 4
Meets level 3 and quantifies, bounds, or stress-tests key uncertainty; states what new evidence would materially change confidence or action.
06Professional judgment, including responsible refusalDid the learner take a proportionate professional position under constraints?20 PTS
LEVEL 0
Recommends a materially unsafe, unethical, or unsupported action; fabricates confidence; or complies with an improper request without challenge.
LEVEL 1
Reports analysis but avoids the decision, or refuses without proportionate investigation and a useful alternative.
LEVEL 2
Gives a plausible recommendation or refusal, but the link between evidence, stakes, alternatives, and action is incomplete.
LEVEL 3
Makes a clear, proportional, evidence-aligned recommendation, qualified conclusion, or responsible refusal; handles stakeholder pressure without overstating certainty.
LEVEL 4
Meets level 3 and anticipates consequences, affected parties, reversible options, escalation paths, and the evidence or threshold that should trigger a different decision.
07Decision communication and governanceIs the work usable by its audience and consistent with data-handling obligations?10 PTS
LEVEL 0
Is materially misleading, exposes prohibited confidential/restricted detail, or obscures the actual decision.
LEVEL 1
Is difficult to act on, omits the population or metric definition, or uses technical output without audience translation.
LEVEL 2
Is understandable and mostly compliant but buries the recommendation, evidence, or limitations.
LEVEL 3
Leads with the decision; defines the metric and population; communicates key evidence and uncertainty concisely; follows catalog sensitivity and export rules.
LEVEL 4
Meets level 3 and anticipates likely misinterpretation, cleanly separates fact from judgment, and tailors detail and escalation to the audience and stakes.
90–100Decision-ready work with strong validation, calibration, and professional judgment.
75–89Credible work that supports action, with bounded gaps or missed robustness opportunities.
60–74Partially defensible work; at least one material issue must be resolved before relying on the decision.
40–59Major analytical or communication weaknesses make the proposed action unreliable.
0–39Evidence and reasoning do not support professional use, or a critical governance failure dominates the work.
Responsible-refusal pathway
Full-credit refusal is not a shortcut. It names the exact unsupported boundary, shows why it matters with proportionate evidence, delivers what can responsibly be concluded now, and proposes the smallest ethical next step that could close the gap or reduce risk.
ASSESSMENT BOUNDARY
The viewer records mechanics. Instructors evaluate judgment.
The current release captures successful browser runs with executed-code hashes, flags outputs that predate later worksheet edits, checks exported workspace files against their saved hashes, and records whether an explicitly bound handoff file is non-empty. It does not verify artifact quality, analytical grain, cutoff validity, inference, or recommendation quality.