Retail Lab · Agent Experiment
Agent-Based Search
Marketplace filters trust the spec table. An agent doesn't — it reads the spec table, then goes and checks whether the spec table is lying. This experiment runs one real, deliberately hard shopping request end to end and shows the difference.
Why this is a different kind of search
Keyword and filter search on Amazon/Wayfair/Walmart answers "what's listed as matching?" Agent-based search answers "what actually matches, and how do I know?" — by treating every retailer spec field as a claim to verify, not a fact to filter on.
Marketplace filter search
- Trusts the "Product Dimensions" field as-is
- Can't tell an overall measurement from an interior one
- Has no way to catch a scrambled W×H×D field
- Returns everything that matches the filter, verified or not
Agent-based search
- Treats the listed spec as a hypothesis, not an answer
- Hunts for an independent source — dimension diagram, assembly PDF, cut sheet, buyer review — before trusting an interior measurement
- Separates verified results from inferred ones, and shows the arithmetic behind every inference
- Actively looks for the specific failure mode that burned the requester before, in every candidate
Example prompt
This is the actual request run through the agent for this experiment — a buffet/sideboard/credenza search with a binding interior-dimension requirement, and an explicit warning about a prior bad purchase caused by a scrambled spec field.
I need to find a buffet/sideboard/credenza. Hard requirements:
RETAILER (scope restriction):
- Amazon, IKEA, or Wayfair only. No Amish/solid-wood makers, office
furniture dealers, or vintage dealers this pass — just these three.
DIMENSIONS (interior measurements are the binding constraint):
- Interior cabinet compartment: minimum 25" WIDE and minimum 14" DEEP.
These are INSIDE-the-cabinet usable dimensions, not overall furniture dimensions.
- Overall depth will therefore need to be ~17-20" (carcass walls and door
thickness typically eat 3-4" of depth and ~3" of width).
- Total wall run: 120" maximum, OUTSIDE dimension, across a combination of
matching-style base cabinets at various widths — not
limited to two identical units.
- Buffet/counter height preferred: roughly 32-36" tall.
- Must be generally available to buy now. Temporarily out of
stock is fine. Amazon's "Currently unavailable. We don't know when or if
this item will be back in stock." reads as discontinued, not a short
restock wait — disqualify regardless of how well it otherwise passed.
- Color: black or dark grey only. Check the product line itself for
a black/dark-grey variant rather than assuming one exists — if a line only
ships in one finish and it isn't black/dark-grey, it's disqualified even
if every dimension passes.
PRIMARY CRITERIA — maximize total storage units, not just pass/fail:
- Count how many 4.4"W x 14"D x 4.4"H storage units fit inside the
combination. A bay contributes floor(width/4.4) width-multiples times
floor(depth/14) times floor(interior-clear-height/4.4). Sum this across
every cabinet in the combination; that total is the number to maximize
within the 120" run, not the count of cabinets or the count of
qualifying bays.
STORAGE TYPE:
- Cabinet space with adjustable shelves is the priority. Drawers are a
negative — zero drawers preferred, minimal drawers acceptable.
- A 25"+ wide bay means the piece must NOT have a center divider behind
each door pair. Typical 4-door/3-divider layouts yield ~14" bays and FAIL.
Two-door cabinets with a single full-width adjustable shelf usually have
no divider and PASS if wide enough.
CRITICAL METHODOLOGY — READ THIS:
Retailer spec tables in this product category are frequently WRONG, and
marketplace filter facets inherit those errors. I have already been burned:
a cabinet listed and filtered as 18" deep is actually 11.8" deep with a
10" interior shelf. Its spec field read "8"W x 14"H x 18"D" — axes scrambled.
So do NOT trust:
- Amazon/Wayfair/IKEA depth or dimension filter facets
- The "Product Dimensions" field in a marketplace spec table
- Marketing copy adjectives like "deep" or "spacious"
DO verify each candidate against at least one of:
- A manufacturer/IKEA dimension diagram image showing interior measurements
- An assembly instruction PDF (IKEA publishes these for nearly every
product, usually with interior callouts)
- The named brand's own spec sheet, where Amazon/Wayfair source from one
- Customer Q&A or review text where someone reports measuring the interior
- Cross-checking the same SKU across 2+ of these three retailers for
dimensional agreement
OUTPUT FORMAT:
A table with one row per candidate: product+brand, overall W×D×H, interior
bay W×D, doors/drawers, price, retailer+link, and interior dims labelled
PUBLISHED or INFERRED. Unverified candidates go in a separate section.
Prioritize PUBLISHED. 3 verified options beat 15 guesses.
Results — Amazon / IKEA / Wayfair only
Restricting the search to just these three retailers made the category noticeably harder — most of what's sold there is flat-pack and shallower than the 14" interior floor — but 1 candidate currently passes every requirement including color, on IKEA, with 2 more runner-ups. Adding a black/dark-grey requirement cost this page its previous #1 pick: TONSTAD Cabinet with Doors was the cleanest dimensional match found, but it only ships in off-white — checked the line directly, no black or dark-grey variant exists, so it's disqualified on color alone (see Ruled Out). A live per-zip stock check on one runner-up (VIHALS) also exposed that this page's earlier availability check — a generic catalog flag — doesn't reflect real per-location stock, so every remaining IKEA item's "Stock" status now reads UNCONFIRMED rather than a false AVAILABLE. Two Amazon candidates and that same VIHALS unit have already come back confirmed UNAVAILABLE — see the dedicated section below. Wayfair produced zero verified passes this pass (see the note at the bottom). The trap example below is still worth knowing even though it no longer involves the current winner: two cabinets share the same IKEA product family name but one hides a center divider.
✓ Best match — verified
HAVSTA Sideboard, Black — IKEA
$400
Overall
47⅝"W × 18½"D × 35"H
Storage
2 sliding doors, 1 shelf, 1 drawer
New #1 pick once black/dark-grey became a hard requirement: the previous winner, TONSTAD Cabinet with Doors, only ships in off-white — checked its product line directly and no black or dark-grey variant exists, so it's disqualified on color alone (see Ruled Out below) despite otherwise being the cleanest dimensional match on this page. HAVSTA's black finish carries the identical price and dimensions as the white version IKEA's own measurement diagram confirmed: 45⅝" interior width straight across where the doors meet, 1.1" over the depth floor. Dimensionally solid — but check the live delivery/store widget for your own zip before ordering, since this page's stock status for it is UNCONFIRMED, not verified AVAILABLE.
View at IKEA →
Working combinations — stacking multiple cabinets to fill the 120" run
A single cabinet is one candidate; a family like HAVSTA — which sells more than one width in the same finish — supports several different end-to-end stackings. Each combination below only uses candidates with zero deficient criteria (a unit that already fails on its own, like VIHALS' height or the Prepac console's depth, can't be fixed by combining it with another), sums their 4.4"×14×4.4" storage-slot counts, and shows the literal width arithmetic that keeps the total at or under the 120" exterior run. Every cabinet name links to its listing. Only combinations that are practically different are shown: the max-slot plan a family can fit, plus an alternate only when swapping in a different model actually changes the cost/verification trade-off or gets you to zero unverified risk — not "3 fit, here's 2 and 1 too."
Candidates
Verified and runner-up candidates in one table, best match first — a runner-up is simply a row with something short on Deficient Criteria or a less-than-PUBLISHED Basis, not a separate class of result.
| Product + Brand | Overall W×D×H | Interior Bay W×D | Doors / Drawers | Price | Retailer | Storage Units (4.4×14×4.4) | Deficient Criteria | Basis |
| Loading… |
⚠ The exact trap this prompt warned about — same product family name, one cabinet passing: IKEA's TONSTAD Sideboard (a separate SKU from TONSTAD Cabinet with Doors, 47⅛"W overall, same collection) looks like a wider version of the same piece — but its own measurement diagram shows one door compartment is only 22" wide, with drawers occupying the rest of the case. Same failure pattern as the requester's prior 18"-listed/11.8"-actual purchase: a product that visually reads as "the bigger version of the one that passed" and isn't. (TONSTAD Cabinet with Doors itself passed every dimensional and divider check — it's disqualified separately, on color, not on this trap.) This is not a fluke — IKEA's BESTÅ storage combination with doors (47¼"W, height also fails at 29⅛") shows the identical pattern: a wide-looking cabinet with a genuine center partition, confirmed in IKEA's own diagram, not inferred.
↳ Why didn't HAVSTA Cabinet with Base come in the first pass? It's not disqualified on dimensions — it was simply never searched for by that name. The prior pass searched "HAVSTA sideboard" and IKEA's larger catalog lines, but not the separate "HAVSTA Cabinet with Base" product within the same series. Checked directly:
31⅞"W × 18½"D × 35"H, 2 doors, 1 adjustable shelf, 0 drawers, $270, black ↗ — an
identical overall footprint to TONSTAD Cabinet with Doors (the strongest dimensional match on this page, though now disqualified on color, not dimensions), and "1 adjustable shelf" (singular) matches the passing pattern rather than the "per door" divided pattern. IKEA embeds its actual interior measurement in a diagram image rather than page text, and this session's search budget is now spent, so the interior figure above is an
inference from the verified HAVSTA Sideboard's own published wall-offset, not an independently confirmed number.
Currently unavailable — passed the dimension check, failed the availability check
A live stock check against each listing found two candidates that no longer look purchasable. Both show Amazon's id="outOfStock" block with the exact same wording — distinct from a normal "temporarily out of stock, notify me" restock message — which reads as discontinued rather than a short-term gap.
Ruled out in this pass — confirmed divided-bay or depth fails
⚠ On Wayfair specifically: zero verified passes this pass. Wayfair's own product pages returned HTTP 429 (rate-limited) on every direct-fetch attempt, so its search results leaned on brand sites and cross-retailer corroboration rather than Wayfair's own diagrams — thinner coverage than Amazon or IKEA got. Wayfair's "sideboard"/"buffet"/"credenza" results were dominated by the same SimpliHome lines that fail on Amazon. Not yet checked: Wayfair's traditional solid-wood buffet lines (Birch Lane, Sunset Trading, Jofran, A-America, Powell), which typically run 19-21" deep — deeper than the flat-pack MCM lines that surfaced first. A follow-up pass targeting those brands specifically is the natural next step if more Wayfair coverage is wanted.
All figures on this page are generated from agent-based-search-data.json — the single data file backing these tables.
How the verification actually worked
- Even inside a fixed retailer scope, hunt for the exceptions — within just Amazon/IKEA/Wayfair, most listings are shallow flat-pack, but a handful (IKEA's TONSTAD/HAVSTA lines, Amazon's Costzon buffet) run deep enough to pass. Restricting scope doesn't mean lowering the bar on verification.
- For every promising candidate, look past the marketplace spec table for an independent source: IKEA's own measurement diagrams, an Amazon listing's in-image dimension infographic, or a named brand's spec sheet where Amazon/Wayfair source from one.
- Sanity-check the numbers against each other, not just against the requirement — e.g. does overall height minus interior height leave a plausible amount for the base/top/drawers, or does the ratio look scrambled the way the prior bad purchase was?
- Actively hunt for the specific failure mode named in the prompt (a center divider hiding behind a wide-looking cabinet) in every candidate — which is exactly what surfaced IKEA's TONSTAD Sideboard as a divided-bay trap sharing a product family name with TONSTAD Cabinet with Doors, itself a passing cabinet later disqualified separately on color.
- Separate what's confirmed from what's inferred, and show the arithmetic for every inference, rather than presenting a best-guess number with the same confidence as a published one.
- Say plainly where coverage is thin — Wayfair's own pages rate-limited every direct fetch this pass, so its results lean on secondary sources; that gap is reported, not papered over with an unsupported "no results" or a false pass.
Agent ResearchVerification-FirstRetail Lab