Candidate C
Primary comparison107.2Klost-sales units
4.4Mnormalized total cost
FP&A / Business Analytics · Validation replay
15.6% lower normalized inventory exposure—with an 8.2% service trade-off.
Under the predeclared L3_A0.90_I1.0 policy, Candidate C reduced normalized total cost and exposure versus Seasonal Naive S7, while lost sales increased. It is a trade-off—not a universal winner.
Business question
How do demand-forecast errors translate into operational and inventory exposure under explicit stock policies, and which decisions change under asymmetric costs?
Executive view
One policy. Three forecast choices.
Total-level inventory simulation for d_1914–d_1941. Costs are normalized proxies; lost sales use observed sales as imperfect demand.
Seasonal Naive S7
Benchmark99.1Klost-sales units
5.2Mnormalized total cost
Last-value naive
Benchmark121.9Klost-sales units
6.6Mnormalized total cost
Inventory and policy are simulated. Values are aggregate-only and do not represent Walmart internal inventory, service targets, or monetary cost.
What changes the decision
Two findings that keep the headline honest.
Seasonal Naive S7 won the official evaluation.
0.847 WRMSSE
On d_1942–d_1969, the frozen seasonal baseline outperformed Candidate C. This post-competition local reproduction was not used for tuning or selection.
The preferred fixed forecast changes with asymmetric costs.
4 of 6 ratios favor Candidate C
Last-value naive ranks first when under-forecast cost reaches 5:1 and 10:1. These are normalized scenarios—not calibrated Walmart costs.
Context, not explanation
Corporate and macro data remain separate from the M5 result.
SEC and latest-revised FRED observations help orient the portfolio. They do not explain the 28-day inventory result and are not joined to the official evaluation.
Derived as operating cash flow less CAPEX; not a directly reported SEC concept.
Skills demonstrated
Evidence before tool names.
Each capability maps to a frozen artifact, receipt, or reconciliation check.
Forecast evaluation
Frozen official M5 evaluation · WRMSSE across 12 levels
Decision analytics
Asymmetric cost ratios · item-store-day loss before aggregation
FP&A
Forecast-to-actual PVM bridge · validation-window controls
Inventory policy
27 predeclared scenarios · lost-sales and holding-cost proxy
Corporate finance
SEC/XBRL statements · lineage and reconciliation checks
Data governance
Frozen manifests · SHA-256 gates · reviewer-approved evidence
Technical details
Built deep. Presented with boundaries.
The home imports seven reviewer-approved aggregate JSON files at build time. SHA-256 verification runs before the build; the deployed page needs no parent-repository path, runtime API, database, secret, or network request.
Review the evidence summary- Validation replay
- d_1914–d_1941 · headline and policy proxy
- Official evaluation
- d_1942–d_1969 · separate accuracy result
- PVM
- Estimated retail revenue — M5 sample; not equivalent to corporate accounts
- Claims
- No causal, optimality, operational, investment, or deployment claim
Project status
Frozen portfolio, with visible limitations.
Full raw-to-public reconstruction, two-run determinism and valid cold/warm runtime were not completed.
The three-person Career Signal Gate was not executed and is not PASS. The owner accepted the omission for time constraints; no feedback, people or timings were fabricated.
- Core evidence
- Stable
- Home
- Reviewer-approved
- Release review
- External review pending · [65] not permitted
- Publication
- Published