Design mockup · Visual only — the whole panel on one page; buttons don’t fire, dials don’t save
GS
Ghost SpawnsShopper Panel · One Page
Machine: standing by
Ghost Spawns — a hooded spectre rises from a glowing gaming PC

They shop the store
so customers don't find the problems.

Fifty ghost shoppers walk gearspawn.com the way real people do — searching, doubting, bailing — then report back, shopper by shopper, in their own words. Nothing they find is ever shown as customer proof. It's the store grading itself, honestly.

They shop the store
so customers don't find the problems.

Fifty ghost shoppers walk gearspawn.com the way real people do — searching, doubting, bailing — then report back, shopper by shopper, in their own words. Nothing they find is ever shown as customer proof. It's the store grading itself, honestly.

Button 1 · One hit

▶ Run the Report

The whole standard run, end to end, on fixed parameters: fresh evidence pull → today's rotation slice → shop → grade → ledger → diff report out. No choices, no setup.

Run it
Button 2 · Scoped run

Deep Dive

Open the dial board and narrow the run — down to a single shopper with a single question — or leave it plain and run broad. Every dial is a picker over data the panel already carries.

Open the dials →

Today's standard dials

Cast sliceRotation — Parents ×5
EvidenceFresh pull
Day stateDealt day
TwinsOn — recorded cadence
Window shoppers3 wanderers
LedgerMain · tagged with run

Latest report — Sun Nov 8 · supervised sitting, post-fix pack v2

S1 · PASS S2 · PASS S3 · PASS S4 · PASS S5 · HUMAN-CONFIRMED CLEAN S6 · PARTIAL — gift search
Becky
research
Darnell
research
Rosa
ready to buy
Amélie
walked away
Zofia
research
45 more
dark today

Five shops lit on the sitting day. The full building — every shopper, every finding, every story — lives on the Daily Screen →

Door status: WIRED BUILD (rehearsal posture) — the gateway is inactive. At cutover this line reports the last fire: run id, stations, and the render job.

The Source Switch — where the shoppers shop New

Live — the crawl. At fire time the puller walks the pasted store the way a guest would: home, collections, product pages, its own search — page by page, guest pace, only what the pages openly declare, nothing behind a login or a checkout. That crawl becomes the evidence pack, with a manifest that records the URL, the timestamp and every page visited. Fresh — but unsealed: if the store changes, the evidence is whatever the crawl saw, cited to the manifest. Findings-grade · citable to the crawl
How the URL gets worked
Compare — a rival, beside GearSpawnAudit — the owner’s own store

Compare dresses the run as the grid’s rival column: the same shoppers, the same tests, their store beside ours. Audit dresses the same machinery as an owner report — the findings for the store that pasted its URL, no install, nothing touched. Same panel, same ledger.

The live standard: a live run fields 25 shoppers — 5 from each of the five categories (Parents · Gift Circle · Performance / PC · Creators · First-time / stranger-style). Never the full cast, never a surge. One careful crawl builds the evidence; the 25 work it — they don’t stampede it.

Blank URL = gearspawn.com. Static with no pack picked = the sealed pack on file (postfix-2026-11-08-livepull-v2). The switch is a door dial: it rides both buttons — Button 1 runs the standard slice in the mode showing; Button 2 narrows it on the dial board below.

Layer 1 · Daily Screen

Sunday, Nov 8, 2026 — post-fix sitting on a fresh pull of the live store (theme Fix 19 · 121 products). Read the day by its rows: no S-line left ungraded.

The verdict, row by row ticket #5

S1 PASS

FAQ contradiction — gone. The staged FAQ carries the first-scan refund promise; no "receive and inspect" sentence remains.

S2 PASS

Product-page promise text — complete. All four markers present: 90-day window, first-scan refund, free/prepaid label, stated timeframes.

S3 PASS

Price agreement — graded on its own key. No $149.00 display remains; search lead and product page agree at $149.99 on the dual monitor arm.

S4 PASS

Vendor fields — clean. EMEET, Gamrombo, Glorious and Keychron read as themselves in the fresh catalog.

S5 HUMAN

Retired list — confirmed clean by human read. Raised keys were monitor-arm-ordering, gift-search-stall, thin-listings — zero of the nine retired R1–R9 keys.

S6 PARTIAL

Gift search — honest partial. "Gift for my son" returns 2 · "gifts for kids" returns 10 · daughter / niece / granddaughter return 0. A real store gap, not a panel failure.

The building — who shopped today

— Ghost Spawns Row —
Becky · 09
research
Darnell · 10
research
Rosa · 11
ready to buy
Amélie · 33
walked away
Zofia · 36
research
Nadia
auditor · off
Tomas
auditor · off
Iris · 08
minor · fenced
Jules · 13
minor · fenced
Marla · W1
window · soon
Dev · W2
window · soon
+41 shops
dark today

Tap any lit shop for that shopper's full story — comments, concerns and wins, never a summary instead of the story.

Since last run — the diff

Changed

  • S3 flipped to PASS — re-keyed to its own written test, then re-pulled and re-sat.
  • Search page ordering flipped — the $64.99 hero arm now leads the full results page.

Repeated

  • gift-search-stall raised again — daughter-gift searches still return nothing.
  • monitor-arm-ordering still raised in suggest — tracked as F6, no longer a grade.

Went quiet

  • faq-refund-contradiction — not reported against the post-fix pack.
  • product-page-promise-text — not reported; all four markers present.

Ranked findings

1

The daughter-gift shelf is empty

Searching "gift for my daughter" returns 0 products — same for niece and granddaughter — while "gift for my son" returns 2 and "gifts for kids" returns 10. A parent shopping for a daughter hits a wall the store doesn't know it built. Grades S6 partial.

Evidence → pull counts & shopper words
2

Suggest still leads with the $149.99 arm

Predictive search ranks the stackable dual arm ahead of the $64.99 hero for "monitor arm". Shopify boosts don't reach the suggest endpoint — the store's ranking is legitimate; this is a merchandising question, not a defect. Tracked as F6.

Evidence → suggest captures
3

Thin listings still surface

The thin-listings key raised again on this pack — stub descriptions remain findable by shoppers doing real research. Already on the Wave-3 enrichment list.

Evidence → flagged products
Ghost Bonds — two spectres bound by one violet current
Twins — two readings, one shopper Dealt day + dice day per shopper. Agree = spine. Split = honest signal. Twin reads land in the same cell on the Compare Grid — never a second row.
Window Shoppers — a spectre drifts past the storefront glass
Window shoppers — no mission, no grade They wander the storefront surface like real browsers. Whatever they bring back is story, never a verdict.

Wanderer notes story only

Sample layout · first wanderer stories land with the first window run Entry: home page · Drifted to: the gift shelf · Bailed at: search results · Why, in their words: "Nothing here tells me who any of this is for."
Format sample — not a real run Entry: a product shared by a friend · Stayed for: the setup photos · Left because: "I couldn't tell what it cost to ship, so I closed it."

Window tickets close as window-complete and never displace a mission run as the latest report.

Behind the Curtain · The Probes

The shoppers read the store the way people do. The probes read it the way Google and the AI answer-engines do — the machine layer underneath: robots, sitemap, structured data, feed state. Deterministic checks with quoted evidence. They report; they never grade.

The crew — a second auditors’ desk working name: Ghost Probes

— Behind the Curtain —
Crawl & Index
auditor desk · 3 checks
Schema & Listings
auditor desk · 5 checks
Store Machinery
auditor desk · 5 checks
Signals & Inventory
auditor desk · 4 checks

Not shoppers, not twins, not wanderers. The probes hold no personas, no day-states, no budgets — a sitemap is not more missing on a Friday than a Monday. Their chain contains no path to the findings ledger at all: they cannot raise a finding, advance a clean cycle, or sit the exam. Nadia and Tomas audit from outside, as strangers; the probes audit the machinery itself.

How to read their section of a run report

ClassAuditors — report only
GradedNever — no answer key applies
Ledger contactNone — rows land on the run ticket only
CadenceEvery committed run, same snapshot as the shoppers
Evidence ruleQuoted artifact — or the row reads BLOCKED
The listProbe list v1 · 17 checks · 4 self-only · 1 rival-only · 12 both

Statuses use the house vocabulary only — PASS, FAIL, PARTIAL, BLOCKED, NOT RUN, N/A. A failed read is BLOCKED with the error quoted, never a guessed pass. On a rival run the same desks read the crawl’s captures, and the Signals & Inventory desk produces the Embedded Intelligence Inventory — the rival’s guidance machines, catalogued from what their pages declare.

Sample report — the Nov 9, 2026 baseline design sample · verified facts, not a run record

MP-01 PASS

Robots file — open. robots.txt allows storefront content for general crawlers; the platform’s agents boilerplate is recorded as data, never followed.

MP-02 PASS

Sitemap — fresh and reconciled. Index plus 6 child sitemaps healthy; 122 product URLs against 121 active products.

MP-03 PARTIAL

Canonicals & index directives — partly sampled. Canonical hygiene proven by the redirect audit; rendered-meta and header sampling begins with this crew.

MP-04 PASS

Product schema — complete. Product and Offer blocks on sampled product pages carry every field Google’s merchant-listing bar requires.

MP-05 PASS

Review stars — in the markup. aggregateRating is embedded wherever reviews exist (TS91 4.56/25 · G19BT 4.84/19); products with no reviews honestly carry none.

MP-06 FAIL

Shipping & returns schema — missing. No shippingDetails, no hasMerchantReturnPolicy in the Offer markup, so free shipping and the 90-day window can’t surface inside Google listings. Fix queued in the Nov 9 changes pack.

MP-07 FAIL

FAQ schema — absent. The FAQ page carries no FAQPage block. Fix queued in the Nov 9 changes pack.

MP-08 PARTIAL

Schema vs page vs pull — agrees where tested. Three-leg agreement verified in fix waves; a standing per-variant test starts with this crew.

MP-09 FAIL

One shelf stands empty. VR & Smart Gaming Gear renders zero products — and can’t be stocked from the active catalog. Staff it or hide it: Ron’s call.

MP-10 PARTIAL

llms.txt — live, currency unswept. The file AI assistants read is up and carries the GEARSPAWN-4 line; a line-by-line currency check against the catalog is new with this crew.

MP-11 BLOCKED

Merchant Center — unreadable from here. The live feed state sits behind the owner’s signed-in session. On file instead: the Nov 3 warning — missing identifiers, limited performance, a suspension warning. The block is quoted, never guessed past.

MP-12 PASS

Search boosts — covered. 33 of 121 active products carry boosts; 8 of 8 live spot checks landed their target in the top four.

MP-13 FAIL

Empty-search shelf — unset. The empty_state_collection setting is absent; a verified candidate (Holiday Gifts Under $50, 35 products) waits on Ron’s call.

MP-14 FAIL

Complementary recommendations — empty. The “pairs well with” rails return nothing storewide; the curation was never configured. Related products (algorithmic) are a different rail and work.

MP-15 NOT RUN

Performance signals — no baseline yet. No field speed data exists anywhere; per-template weight proxies begin with the first probe run. Real Core Web Vitals join only on Ron’s word.

MP-16 PARTIAL

Signature capture — partial. Served JSON-LD inventoried on sampled pages; the versioned vendor/signature table is a build artifact that doesn’t exist yet.

MP-17 N/A

Rival inventory — not this run. The Embedded Intelligence Inventory is a rival-only check; this baseline is a self run. It wakes when the crawler does.

Roll-up: 5 PASS · 4 PARTIAL · 5 FAIL · 1 BLOCKED · 1 NOT RUN · 1 N/A. Every FAIL above already names its fix lane in the Nov 9 changes pack — the probes report; the store lanes, on Ron’s word, repair. Coverage: all four desks reported; the one BLOCKED row is an access state, stated on its face.

Design sample for the redesigned page. Live probe sections render from the run ticket at graduation; the values above are the verified Nov 9 baseline the first real run will be compared against.

Layer 2 · Compare Grid

Shoppers across, tests down the top. One tap under any cell is that shopper's full story. Twins never get a second row — the twin read lives inside the same cell.

Ghost Bonds — two spectres, one current
Bound readsDealt day and dice day, tied to one shopper. Where the two reads split, the cell shows both — split is signal, not noise.

Nov 8 sitting — Parents slice × S1–S6

Shopper S1FAQ contradiction S2Promise text S3Price agreement S4Vendor fields S5Retired list S6Gift search Outcome Rivalvia the source switch
Becky · gs_09Parents · research lane pack ✓pack ✓ pack ✓pack ✓ human ✓quiet research no crawl yet
Darnell · gs_10Parents · research lane pack ✓pack ✓ pack ✓pack ✓ human ✓quiet research no crawl yet
Rosa · gs_11TWIN CELLParents · buying lane pack ✓pack ✓ pack ✓pack ✓ human ✓quiet ready to buy no crawl yet
Amélie · gs_33Parents · now US-mindset pack ✓pack ✓ pack ✓pack ✓ human ✓quiet walked away no crawl yet
Zofia · gs_36TWIN CELLParents · gift context pack ✓pack ✓ pack ✓pack ✓ human ✓RAISED research no crawl yet
Window shoppers — story only · never graded · never in the streak math
Marla · gs_w01Window · Tucson story — entered at home, bailed at the gift shelf window-complete —
Dev · gs_w02Window · Austin story — followed a shared product link, stayed for the photos window-complete —
Grace · gs_w03Window · Columbus story — sample band; first real wanderer rows land with the first window run window-complete —
pack ✓ — deterministic check on the store surface human ✓ — human-read line raised — this shopper's day raised the key quiet — nothing raised

How to read it: S1–S5 graded at pack level on this sitting — the surface either carries the fault or it doesn't, so every shopper's cell inherits the pack read. S6 is the shopper channel: Zofia's grandson-gift context genuinely raised gift-search-stall, and the check's own pull counts (daughter = 0) set the partial. Three further outputs matched the gift keyword channel without a genuine gift context — the grid shows genuine raises only.

👁‍🗨
The Auditors Nadia, the Returns Lawyer, and Tomas, the Second Monitor Skeptic — two stranger shoppers on a different engine family who audit the panel's own graders. Emit-only: no write path into the store, ever. The Auditors — poster: two stranger spectres outside the glass, watching the graders
🛰
Recon — the rival column At graduation the grid gains a rival column: the same shoppers walk rival public storefronts at guest pace, graded only on what those pages declare. Nothing behind a login, nothing at checkout. The Recon — poster: ghost shoppers drifting over rival storefronts at night
R

Rosa · gs_11

Parents cluster · bilingual household (es-US) · shops for family game night
Calibration cast — one of the five who sat the G1 exam
Outcome · Ready to buy

The dials that dealt her day

RunNov 8 post-fix sitting · ticket #5
EvidenceFresh pull — live store surface
Day stateDealt day
Mission contract3 searches · up to 2 questions
ClusterParents — twin cadence on
Engine laneParents lane (Gemini family)

Her day, in full

Rosa came in the way the store hopes a parent comes in: ready to buy. Family game night needed a controller, and she wasn't browsing for fun — she had a job to finish. Her day ended in the strongest state a shopper in her lane can reach: ready_to_buy, with the store's promises checking out around her — the refund language said what the FAQ said, and the prices she met in search were the prices on the product pages.

Her most valuable moment for the store, though, came on the calibration pack, and it's the reason her name sits in the G1 record. She searched "xbox controller" — and the store's own suggestions led her to a two-pack from the Xbox 360 generation, priced at $59.00, with nothing on the surface marking it as last-gen.

"This is the first one it shows me. Is this the one for our console? It doesn't say it's old — it just looks like the deal."

On the Nov 8 post-fix sitting she shopped clean: researched, verified, and left ready to buy. Her story is what the whole panel exists to produce — one shopper, fully readable, every concern attached to its evidence.

Wins & concerns

Wins

  • Reached ready_to_buy on the post-fix store
  • Search prices and product-page prices agreed
  • Answered in her own language, prices and names exact

Concerns

  • Suggest leads "xbox controller" with a 360-era two-pack
  • No generation badge anywhere on that suggestion
  • A trusting parent can buy the wrong console's controller

Evidence — one tap each

  • Suggest capture — "xbox controller", calibration packOpen →
  • K6 human-grade packet — verdict: PRODUCEDOpen →
  • Sitting output — gs_11 parsed day, ticket #5Open →
  • Price-agreement record — search vs product pageOpen →

Rendered from Rosa's recorded material: the K6 human-grade packet from the pre-fix calibration (graded and confirmed Nov 8) and her parsed outcome at the post-fix sitting. Sample prose layout for the Layer 3 template — in the live door, this page is generated from her stored report, word for word.

Deep Dive · The Dial Board

Button 2's page. Narrow the run as far as one shopper with one question — or leave every dial plain and it runs the standard slice, broad. Every picker below is data the panel already carries.

Who shops

Cast scope
Whole castOne clusterOne shopper
Cluster
ParentsGift CirclePerformance / PCCreatorsFirst-Time / FTC
One chosen shopper — "choose one person, do this"
Who they are — demographic
ParentGift-giverCreatorPerformance builderAge 25–44Age 45+Bilingual household
Where they're at — market (US cast)
Anywhere in the USPhoenix, AZTucson, AZAustin, TXColumbus, OHNashville, TNLas Vegas, NV
What they're hunting — the question asked of the store
Budget band
Their cast bandUnder $50$50–$100$100–$250No ceiling

The machine dials

Evidence

Snapshot on file: postfix-2026-11-08-livepull-v2 — before/after mode. A snapshot is only valid for the store state it was pulled from.

Day state
TwinsSecond reading per shopper on cadence clusters. Agree = spine, split = honest signal. Twin reads share the cell.
Window shoppersUnbriefed wanderers on the storefront surface. Story only — never graded, never in the streak math.
3
Surface in scope
✓Search & suggest
✓Product pages
✓Policy pack
✓Gift lane
Leave it plain = the standard slice. Every dial above at its default is exactly Button 1's fixed run. Narrowing writes to the main ledger, tagged with the run's parameters — history stays one story. Until graduation, every press opens a supervised run: Muse babysits it and reports.
GearSpawn · Ghost Spawns
Internal diagnostic only — ghost shoppers never appear as store data.
Until graduation, every press opens a supervised run.
Mockup — not live
1 · PullFresh snapshot of the live store
2 · PlanCast + dials dealt
3 · RunShoppers, twins, window wanderers
4 · GradeExam + auditors
5 · ReportPages update; findings to the ledger