Friction register

Internal working document, compiled 1 Sep 2026 from the eight completed workstreams (10-shopify.md through 80-unit-economics.md). Not the client deliverable. This document merges and prioritises; it does not audit. No finding below was invented here: every row traces to a finding ID already written in its originating document.

Global constraints applied throughout: no em dashes, no emojis, no revenue projections, no PII, no credentials.

Legend used in the register:

• Class (four verbs, business effect not technical category): BLOCKS

(revenue cannot occur), CORRUPTS (decisions rest on false or missing data), WASTES (money leaves, nothing returns), SUPPRESSES (demand exists but cannot find them). A small number of source rows carry class UNKNOWN where the originating document could not determine effect at all (an open access or ownership question, not a business-effect claim). These are grouped after SUPPRESSES rather than forced into one of the four verbs.

• Tier: HARD (pulled from a system, dated) / DERIVED (arithmetic on

HARD) / INFERRED (reasoned from HARD facts, not measured) / UNKNOWN.

• Effort: none (already resolved, informational only) / low (a status

change, a field edit, a single decision) / medium (a build or a multi-step process change) / high (a strategy-level or structural change). Assigned here for prioritisation; not independently audited.

• Blast radius: small / medium / large, this task's judgment of how

much of the business the finding touches, used only as the final sort key.

• Remit: CRO scope (inside Nuuk's signed Stage 1 remit: product page

CRO, add-ons, email lifecycle) / media (advertising execution) / client (the owner's own decision) / UNOWNED (nobody is currently accountable for this under any existing contract).

• Prior-art: No / Partial / Yes, per the originating document's own

check against Nuuk's June 2026 audit, Stage 1 SOW, and July 2026 proposal. Partial means a general topic is Nuuk's but a specific, dated fact recorded here is new. Yes/Partial rows are excluded from client-facing use in their restated form; only the specific new fact in a Partial row may be used, never the general topic.

Sort order: class (BLOCKS, CORRUPTS, WASTES, SUPPRESSES, then UNKNOWN), then effort ascending, then blast radius descending. The logic: spend fixes cannot be evaluated while measurement is broken, and neither is worth doing while the funnel is closed at the end. Within a class, cheapest fixes first; among equally cheap fixes, the ones touching the most of the business first.

---

Six reconciliations

Six known inconsistencies surfaced across the eight workstreams. Each is resolved explicitly below; the register uses the resolution, not the original framing, everywhere it appears.

1. Meta lifetime spend: $4,801.41 vs $4,801.42 (one cent)

00-open-questions.md and an earlier snapshot cite $4,801.41; the 1 Sep 2026 same-day pull in 20-meta-ads.md reads $4,801.42, on the ad account's own amount_spent field. This register uses $4,801.42, the more recent, directly-dated pull, and treats the one-cent gap as rounding noise across snapshots taken on different days, not a finding. Where a range is needed elsewhere in this register (for example ECON-01), $4,801.41-$4,801.42 is shown as the low end of a range, consistent with 80-unit-economics.md's own treatment.

2. Meta amount_spent ($4,801.42) vs three-method Insights ($4,821.05): $19.63 UNRECONCILED

Three independently-shaped queries (account-level single row, campaign-level summed, ad-set-level summed) all agree at $4,821.05. The ad account's own amount_spent field reads $4,801.42 on the same day. Both are HARD, dated, and reproducible; they simply disagree, and no cause was found for the $19.63 gap.

This register uses $4,821.05 for any calculation that needs one number (the cross-checked, three-method figure), because a figure confirmed three independent ways is stronger evidence than a single account-level field. $4,801.42 stays visible everywhere it is the more commonly quoted figure (it is what a client opening the Ads Manager account overview would see first), and META-01 remains open in the register as an unresolved CORRUPTS finding rather than being closed by this choice. Where 80-unit-economics.md already computed a range using both figures (ECON-01, ECON-03), that range is preserved rather than collapsed to one number.

3. Task 3 corrects Task 2's META-02: 11 vs 13 lifetime purchases is not unstable data

Task 2's META-02 recorded this as CORRUPTS/HARD, framed as Meta's own purchase count changing between pulls. Task 3 (30-tracking.md, TRACK-04) re-tested with a second, independently-shaped query and found both counts are correct, stable readings of two different Meta fields: purchase (the standard website-pixel event) = 11, $755.45 lifetime; omni_purchase (a broader metric that additionally counts 2 Messenger-based messaging_order_created_v2 conversions from now-paused MESSAGING_PURCHASE_CONVERSION ad sets) = 13, $770.54 lifetime. December 2025's extra 2 in omni_purchase matches December's messaging_order_created_v2 count exactly.

This register reflects the correction. META-02 does not appear below as an open finding; TRACK-04 does, carrying the resolved explanation. Any other row in this register that needs a lifetime purchase count uses 11 (purchase) as the conservative default and states omni_purchase (13) alongside it where the source document does.

4. Task 8 arithmetic error: $112.02/$17.90 corrected to $112.05/$17.93

80-unit-economics.md Section 2 and the ECON-02 finding row both understated the generous repeat-purchase read. Corrected arithmetic: (25 live orders / 21 distinct buyers) x $94.12 AOV = 25/21 x $94.12 = $112.05 per buyer (not $112.02), and the derived incremental value is $112.05 - $94.12 = $17.93 per buyer (not $17.90). This has been corrected directly in 80-unit-economics.md (Section 2 prose and the ECON-02 finding row) as part of this task, and this register uses the corrected figures throughout. The conclusion is unaffected: $17.93 still closes under 9% of the $201.88 gap between CPA and AOV, and repeat purchase still does not rescue the break-even picture.

5. Meta purchases: 6 (90-day window) vs 7 (calendar-month window) for identical $455.96

30-tracking.md Reconciliation 1 uses a rolling 90-day window (2026-06-03 to 2026-08-31) and reads 6 purchases at $455.96. Reconciliation 2 uses calendar months (2026-06-01 to 2026-08-31) and reads 7 purchases at the same $455.96 total value. The extra purchase in the calendar-month window carries no incremental value, which is consistent with a $0-value purchase landing on 1 or 2 June 2026 (the two days the calendar-month window includes that the 90-day window does not). This was not directly tested in any workstream and is recorded here as UNKNOWN, not resolved. It does not change any conclusion in this register: both windows agree Meta under-claims relative to Shopify's revenue-generating order count in the same period (TRACK-01), and neither the 6 nor the 7 count changes the dollar figure used anywhere else.

6. Task 1 report summary count corrected: 9 findings, not 13

Task 1's own report summary read as 13 findings. The actual count in 10-shopify.md is 9 findings (SHOP-01 through SHOP-09), of which 5 carry a PRIOR-ART note (SHOP-02 Partial; SHOP-06, SHOP-07, SHOP-08, SHOP-09 Yes) and 4 are genuinely novel with no prior-art overlap (SHOP-01, SHOP-03, SHOP-04, SHOP-05). This register uses 9 and 5/4 throughout; the 13 figure does not appear below.

---

The register

58 findings merged from eight workstreams. Rows marked "resolved" in the Effort column are informational corrections carried forward for traceability, not open action items; they sort to the front of their class bucket and are excluded from the top-five selection and from any spend or effort total.

BLOCKS

CORRUPTS

WASTES

SUPPRESSES

UNKNOWN (class not determinable from evidence available)

Totals: 58 source findings across the eight workstreams, presented as 57 register rows: META-02 is not given its own row because Reconciliation 3 folds it into TRACK-04, the same underlying fact, corrected, so it is merged rather than duplicated under two IDs. 8 rows carry a PRIOR-ART flag (5 fully excluded: SHOP-06, SHOP-07, SHOP-08, SHOP-09, SEO-03; 3 Partial, specific new fact only: SHOP-02, CRO-01, CRO-04). 24 findings are marked UNOWNED: SHOP-01, SHOP-04, META-01, GADS-01, GADS-02, TRACK-01, TRACK-04, TRACK-06, SEO-02, SEO-04, SEO-05, SEO-06, SEO-07, SEO-08, GEO-01, GEO-02, GEO-05, COMP-01, COMP-02, ECON-01, ECON-04, ECON-06, ECON-07, ECON-08. This is the largest single remit bucket in the register: SEO, GEO/AEO, and most of the tracking- and economics-diagnostic findings sit here because no signed contract in this engagement covers SEO, AI-search readiness, or measurement work. Nuuk's remit is product page CRO, add-ons, and email lifecycle; media execution belongs to whoever runs the ad accounts; client covers decisions only the owner can make (product status, pricing, inventory, identity, access). Everything else in that gap is UNOWNED.

---

The selection: top five for the client report

HARD or DERIVED tier only. Nothing INFERRED, nothing UNKNOWN, nothing PRIOR-ART (full or Partial). Each is defensible line by line against the client's own systems. Presented in class order (the same logic as the register: the funnel has to be open before spend is worth evaluating, and spend has to be measured correctly before it is worth optimising).

1. SEO-01 / SHOP-03: the flagship product cannot be found by anyone who doesn't already have the link

Finding. The $59.99 case that carries the store's own flagship SKU and top-level brand name (String Protech® Guitar String Protection Case) is set to UNLISTED status. Shopify's documented, automatic behaviour for that status adds a noindex,nofollow tag to the page and strips it from sitemap.xml. The product resolves normally on direct link and gets real traffic (102 sessions in 90 days), but has no path to new discovery through search or any AI system that follows the store's own sitemap.

Evidence. gql products query and live meta-tag/sitemap pull, both 1 Sep 2026, cross-confirmed against Shopify's own documented platform behaviour for UNLISTED products.

Why it matters commercially. This is not a ranking problem to be improved incrementally. It is a structural switch: while the product stays in this state, every dollar spent making it more findable (better titles, better content, better schema) has nothing to attach to, because the page itself is excluded from the index it would need to rank in.

What it would take to fix, conceptually. A decision on the product's listing status, made with visibility into why it was set that way in the first place (the store also holds a near-duplicate DRAFT record on the same SKU, so the status question and the duplicate-record question likely need to be resolved together, not separately).

2. ECON-02: no margin makes the current paid-acquisition math work

Finding. Break-even cost per paying order, expressed as a function of gross margin, tops out at $94.12 (the full average order value) even at a theoretical, not-real-for-a-physical-product 100% margin. The store's actual blended cost per paying order is approximately $296, more than 3x that ceiling. The margin required to make the current spend break even is over 314%.

Evidence. Arithmetic on the store's own 12-month order export and the independently cross-checked Meta and Google spend totals, both dated 1 Sep 2026.

Why it matters commercially. This is the number that makes every other finding in this register urgent rather than merely tidy. It does not depend on knowing the client's actual unit cost, which is why it is presented as a ceiling rather than a single figure: whatever the real margin turns out to be, it cannot close a gap this size on its own.

What it would take to fix, conceptually. The lever that has to move is spend efficiency or average order value, not margin. That points directly at the WASTES-class findings elsewhere in this register (money leaving on non-purchase-capable ad sets, wasted video-campaign spend) as the place to start, once the client's actual unit cost is known so the real target can be set.

3. TRACK-02: the measurement layer is missing roughly a quarter of real revenue

Finding. GA4's own purchase/ecommerce tracking misses real, revenue-generating Shopify orders across multiple months, verified at the individual transaction level, not just in monthly totals: 20 GA4 transactions and $1,353.38 against Shopify's 19 revenue-generating orders and $1,788.30 over the ten months checked, a 24.3% revenue undercount.

Evidence. Transaction-level GA4 pull cross-referenced against the Shopify order export, both dated 1 Sep 2026, with three separate months matching to the exact cent (ruling out a currency or conversion-rate explanation).

Why it matters commercially. Every decision that leans on GA4 revenue numbers, including the Google Ads cost/revenue picture, is working from an undercounted base. This is worth fixing independent of anything to do with Meta: it means the store's own measurement of itself understates its own performance.

What it would take to fix, conceptually. An audit of why real, paid orders are not producing a GA4 purchase event some months and not others, followed by whatever repair that audit points to.

4. META-03: over a third of lifetime Meta spend sits on ad sets that structurally cannot produce a purchase

Finding. $1,900.66 of the account's cross-checked $4,821.05 lifetime spend (39%) ran on ad sets whose optimization goal carries no purchase-capable event at all: POST_ENGAGEMENT (two ad sets), PROFILE_VISIT, and THRUPLAY. This is a platform-structural fact confirmed directly on each ad set's configuration, not an inference from results.

Evidence. Ad-set and ad-level insight pulls, cross-checked at two levels of granularity, 1 Sep 2026.

Why it matters commercially. This is money that was never able to buy a sale, by the platform's own design, independent of creative quality, targeting, or budget. It is the single largest, most directly attributable waste figure in the whole account.

What it would take to fix, conceptually. A decision on which objectives the ad account should be funding going forward, made with the optimization-goal-to-purchase-capability mapping in view before any budget is reallocated.

5. GEO-01: no independent source anywhere vouches for this brand

Finding. A live AI-assistant query for the brand plus the word "reviews" returned only the brand's own three pages; the assistant itself noted it could not find independent coverage and offered to search further. This was tested directly, today, not inferred from organic search rank.

Evidence. Live AI-assistant query, 1 Sep 2026, full transcript saved.

Why it matters commercially. A brand's own product page and patent claims are not evidence to an AI system deciding what to recommend; a citation from somewhere else is. As buyer research increasingly starts inside an AI assistant rather than a search results page, having nothing for that assistant to point to besides the brand's own words is a growing, not shrinking, gap.

What it would take to fix, conceptually. Earning presence somewhere outside the brand's own domain that an AI system's index can find and treat as corroboration, rather than continuing to rely solely on first-party claims.

---

What we could not establish

Every material UNKNOWN surfaced across the eight workstreams, recorded honestly rather than guessed at.

• **Where the purchase-optimized Meta campaigns actually send clicks

(CRO-02).** The [Venda] campaign and the cupom STRING10 remarketing ad sets serve creatives with no queryable destination URL on this engagement's read-only API access. Ad set IDs involved: campaign 6994075424008 ([Venda]) and ad sets 52538349915212, 52538348943212, 52538305228612. This needs a manual check inside Meta Ads Manager, which this engagement's credentials cannot reach; it would take roughly five minutes and would close CRO-02 outright.

• The true Google Ads spend start date. Every Google Ads spend figure

in this register is a floor, not a lifetime total, because the GA4-to-Ads link creation date lives only in the GA4 Admin API's googleAdsLinks resource, which returns 403 SERVICE_DISABLED for this engagement's credential. Google spend is visible in GA4 only from May 2026 onward; whether that is when spend actually started, or only when the link began reporting, cannot be determined without direct Google Ads account access.

• The client's unit cost. Gross margin, and therefore where on the

break-even table the business actually sits, is unknown. This gates the precise version of ECON-02 and the correct next-step spend target; nothing in this register guesses at it.

• Core Web Vitals (LCP, CLS, INP). PageSpeed Insights returned a

daily-quota error on every call for both the homepage and the flagship product page and did not recover on retry. These three metrics are UNKNOWN, not estimated from a proxy; server time-to-first-byte was captured instead and is explicitly not presented as a Core Web Vitals substitute.

• Google Ads account ownership (GADS-04): whether the client's own

business owns the Google Ads account outright, the way Meta's business confirmably does, or an agency manager account holds it. This carries real transition risk and is answerable in five minutes inside Google Ads > Admin > Access and security > Managers, a screen this engagement cannot reach.

• Who Aguiar Prates and Roberta Pereira are (META-08): two named

actors in Meta's own activity log with exact timing correlations to known campaign-creation dates, neither of whom appears in the account's current agency list and neither of whom is named in any document this engagement holds.

• **Whether the 6-vs-7 Meta purchase count for the identical $455.96

figure reflects a $0-value purchase on 1 or 2 June 2026** (Reconciliation 5). Not directly tested by any workstream.

• **Whether the 2,481-unit inventory total is duplicate bookkeeping or

genuine overstock** (ECON-05). The arithmetic is HARD; the interpretation is INFERRED toward duplicate records, not confirmed.

• A cohort/repeat-purchase curve by month of first order (ECON-07).

The PII-free order aggregate never retains a buyer hash against an order date, so only the headline 14.3% repeat rate is available, not a trend.