Three weeks into a 7-month engagement, I was paginating Walmart's /v3/items endpoint to refresh status on a few thousand listings, and the data kept looking weird. Same SKU appearing on page 4 and page 7. Item counts that didn't reconcile against the Walmart dashboard.
I ran the pagination twice, byte-compared the responses, and confirmed it: roughly 30% of the pages Walmart returned were duplicates. The documented endpoint was broken. The client had been syncing prices against this for an unknown number of months.
We re-pushed 2,399 corrected prices once I rewrote the sync to use /v3/reports?reportType=ITEM&reportVersion=v4. I never found out exactly how much that bug had cost in lost Buy Box and mispriced sales, but the catalog was systematically wrong against what Walmart actually showed on the seller dashboard for as long as the bug had been in place.
That was one finding. There were 38 more.
The engagement was 7 months on an Odoo 18 platform for a U.S.-based made-to-order manufacturer. About 7,000 SKUs across Amazon, eBay, and Walmart. A pre-engagement Claude-driven code review had surfaced 41 findings: 8 critical, 33 high/medium.
On Amazon: 703 variation families were drifting. I reconciled all 703 to 100% catalog match. 7,644 SP-API submissions accepted with no rejections needing follow-up. I also found a recurring lead-time-to-ship bug that was wiping shipping windows on every fulfillment patch. 6,838 active FBM listings had had their lead times zeroed at some point in the prior year.
On eBay: 708 multi-variation listings throwing [21916735] errors. Built a GetItem comparison to determine source-of-truth, fixed ReviseInventoryStatus container behavior.
The image work nearly broke things. Converted 8,623 WebP/PNG-alpha masters to PNG-RGB. Conversion triggered Odoo's image-derivative chain to cascade-delete 34,488 derivative attachments. ORM recovery: 11 minutes per 1,000 records. Wrong path. PIL bulk-insert direct to ir_attachment while still triggering the derivative-compute chain in the same transaction. ~32,500 derivatives recovered without service disruption.
Underneath all of that: snapshot/diff watchdog (11 health checks), MD5-grouped error monitor, 143-method test suite. None of the marketplace work would have stayed fixed without them.
By handoff: 8 of 8 critical findings closed, 30 of 33 high/medium findings closed. Production cutover unblocked.
The transferable lesson: most of what was wrong here was silent. The variation drift had been there long enough that the dev team had stopped looking at it. The Walmart pagination bug was burned into the documented endpoint. The client's sync code was correctly implemented against an incorrect API spec.
An Odoo + Marketplace Reconciliation Audit is a 3-business-day snapshot that surfaces it specifically: affected SKUs, dollar-impact estimates, fix-effort estimates. The audit reconciles your Odoo catalog against each connected marketplace and tells you where they disagree, which side is right, and what the disagreement is costing. $1,500 single-marketplace. $2,500 all three. Book a 15-min call or email david@ai-frost.com to scope one.
One more thing this engagement produced that doesn't show up in the SKU counts: the AI-augmented practice stack itself. Knowledge vault, persistent memory, module-lock hooks, watchdog snapshots, hard rules tied to real incidents, the 143-method test suite. 5 of the 7 months were spent building that stack alongside the data work, learning Claude Code (early-tooling era), and grinding through stakeholder alignment and Odoo data hygiene that should have been handled before I arrived.
The stack is the reusable artifact. New engagements get it deployed in week 1 instead of built across 5 months. With a passed Reconciliation Audit handling the discovery and a client providing decision-maker access within 24 hours, the engineering sprint that traditional consulting would estimate at 18-24 months compresses to roughly 1 month. That's the offer. 15-20x calendar compression, comparable per-hour rate, materially lower total project cost.
Full case study with the practice-stack walkthrough, piece by piece, plus the cost-comparison math and what new engagements actually look like in practice: 7 months of AI-augmented Odoo engineering, decomposed.
| Year | Engagement |
|---|---|
| 2025–2026 | Multi-marketplace Odoo 18 engagement 703/703 reconciled · 38/41 findings closed · 7,644 SP-API submissions · 32,500 derivatives recovered |
| 2025 | MountainPumpkins.com Solo e-commerce build · Stripe · Next.js · deploy pipeline |
| 2023–2025 | T2 Engineering · Minuteman III · Hill AFB Senior electrical engineer · weapons-system sustainment |
| 2021–2023 | USAF 309th SMXG · Hill AFB TPM · RFID asset mgmt · 12-dept · 7,200 hrs + $500K/yr saved |
| 2017–2021 | USAF SEON · Hill AFB Lead electronics eng · telescope-network DevSecOps |
| 2016–2017 | NXP Semiconductors Electronic test engineer · mixed-signal characterization |