Exploratory patterns within the verified error answers
Companion evidence and limitations for the 80-offer consumer report.
This appendix accompanies the Outrigger consumer report. It groups all 36 archived answers already adjudicated as containing a verified material error: 34 primary answers and two technical-retry answers, covering 32 distinct offers. All belong to the other-promotions group; none belongs to the five retailer-labeled Black Friday offers.
These are counts within an already selected error subset, not error rates among all 156 answers or among shoppers generally. The group definitions were developed after the original review. They describe the affected purchase detail, not the cause of a model error. Original outcomes, labels and disagreements remain unchanged.
How the grouping was done
A language-model reviewer read every effective material incorrect record for the 36 responses, its final source review and supporting evidence. Distinct answers are counted once in each applicable group. An error repeated in a price, subtotal and summary does not create three responses. One answer can enter several groups, so the group counts must not be added as though they were mutually exclusive.
The review examined 63 effective claim records for coverage. Those records can restate or split a single substantive contradiction; they are not 63 unique errors. Records coded as unknown, omitted or correct were excluded from these error groups. A known mandatory charge missing from an asserted complete total can instead be an admitted contradiction in that total; those cases retain their original incorrect labels. An unavailable final quote or an honestly acknowledged unknown is not treated as an error. Historical visual findings retain their archived screenshot-review attribution; this thematic review did not claim a new direct image verification. This is post-unblinding model-assisted analysis, not human verification or a new blind replication.
Groups and counts
| Affected area | Primary answers | Retry answers | Total answers in group |
|---|---|---|---|
| Price / promotion | 10 | 1 | 11 |
| Extra charges / total | 10 | 0 | 10 |
| Shipping / fulfillment | 9 | 0 | 9 |
| Product / contents | 9 | 0 | 9 |
| Returns | 7 | 1 | 8 |
All group counts use the same scope of 36 error responses. Group sizes depend on the chosen definitions, and several products came from the same retailer. For example, six mandatory-fee cases came from four DreamCloud offers, and three appliance-delivery cases came from three Home Depot offers. Those are clustered observations, not independent replications across the retail market.
Price / promotion
Incorrect current merchandise/membership price, advertised comparison amount, active promotion/mechanism, or discount calculation. Derived totals based on that same price misunderstanding remain here; errors about additional shipping/state charges are separately grouped.
Included public case summaries: Case 04 (Gemini, initial answer), Case 05 (Gemini, initial answer), Case 06 (Gemini, initial answer), Case 09 (Gemini, technical retry), Case 10 (Gemini, initial answer), Case 12 (Gemini, initial answer), Case 13 (ChatGPT, initial answer), Case 17 (ChatGPT, initial answer), Case 26 (Gemini, initial answer), Case 30 (Gemini, initial answer), Case 34 (Gemini, initial answer).
Extra charges / total
Wrong or absent additional mandatory state fee or shipping charge while presenting a complete/known expected before-tax total. Includes the associated incorrect charge claim. Does not absorb every wrong item-price input merely because it appears again in a total.
Included public case summaries: Case 18 (Gemini, initial answer), Case 19 (ChatGPT, initial answer), Case 20 (Gemini, initial answer), Case 21 (ChatGPT, initial answer), Case 22 (Gemini, initial answer), Case 23 (ChatGPT, initial answer), Case 28 (Gemini, initial answer), Case 31 (Gemini, initial answer), Case 35 (Gemini, initial answer), Case 36 (Gemini, initial answer).
Shipping / fulfillment
Incorrect shipping charge/free-delivery eligibility, service eligibility, timing assigned to a service/market, or seller-versus-shipper identity. This factual-domain grouping does not assert a common model failure mechanism.
Included public case summaries: Case 06 (Gemini, initial answer), Case 07 (ChatGPT, initial answer), Case 15 (Gemini, initial answer), Case 16 (Gemini, initial answer), Case 28 (Gemini, initial answer), Case 31 (Gemini, initial answer), Case 33 (Gemini, initial answer), Case 35 (Gemini, initial answer), Case 36 (Gemini, initial answer).
Product / contents
Mismatch in depicted selected variant/configuration, wrong included component/material, wrong exclusion of included goods/protection, or optional styling incorrectly called required.
Included public case summaries: Case 01 (ChatGPT, initial answer), Case 02 (ChatGPT, initial answer), Case 11 (Gemini, initial answer), Case 14 (Gemini, initial answer), Case 23 (ChatGPT, initial answer), Case 24 (Gemini, initial answer), Case 25 (Gemini, initial answer), Case 27 (Gemini, initial answer), Case 29 (ChatGPT, initial answer).
Returns
Incorrect selected-item return duration, start anchor, return-shipping/restocking claim or component eligibility. Scope is the admitted contradiction, not every unresolved return detail.
Included public case summaries: Case 03 (Gemini, initial answer), Case 05 (Gemini, initial answer), Case 08 (Gemini, technical retry), Case 18 (Gemini, initial answer), Case 20 (Gemini, initial answer), Case 22 (Gemini, initial answer), Case 24 (Gemini, initial answer), Case 32 (Gemini, initial answer).
Case level explanation
The following table preserves the full 36-response set. “Primary” identifies the initial attempt and “retry” a permitted technical retry. These explanations summarize admitted contradictions; they do not certify every other statement in the answer.
| Public case summary | Affected areas | Supported contradiction and relevant qualification |
|---|---|---|
| Case 01 (ChatGPT, initial answer) | Product / contents | The answer card showed a cream lower body with a broad bright stripe instead of the selected navy polo with narrow stripes; the correct price was not made false by the image error. |
| Case 02 (ChatGPT, initial answer) | Product / contents | The answer card showed a pale-blue lower panel instead of the selected predominantly navy crewneck with warm broad stripes. Original misrouted citation artifacts and corrected supplements remain separate. |
| Case 03 (Gemini, initial answer) | Returns | The answer gave the selected Newegg computer a 30-day return period; the exact item showed 15 days for refund and replacement. |
| Case 04 (Gemini, initial answer) | Price / promotion | The answer denied an active advertised promotional code although the exact contemporary listing advertised $30 off with SSF74363. This establishes an advertised-code contradiction, not checkout acceptance. |
| Case 05 (Gemini, initial answer) | Price / promotion, Returns | The answer asserted a current $1,649 price instead of the selected $1,599, gave 30-day returns instead of 15 days, and universally denied restocking fees although the official policy allowed a 15% fee in select opened categories. No fee for this exact laptop is inferred. |
| Case 06 (Gemini, initial answer) | Price / promotion, Shipping / fulfillment | The answer stated $1,114.99 after a $45 reduction from $1,219.99, instead of the conditional arithmetic result $1,174.99. It also identified AOACE as both seller and shipper; the listing distinguished sold by AOACE from shipped by Newegg. |
| Case 07 (ChatGPT, initial answer) | Shipping / fulfillment | The answer assigned 2–3 business days to standard delivery; the archived U.S. table assigned 3–4 to standard and 2–3 to expedited. The policy capture preceded the answer by one day; no actual delivery was measured. |
| Case 08 (Gemini, technical retry) | Returns | The answer promised a free prepaid return label; the ordinary direct-purchase policy and case-specific pre-answer attestation recorded a $4.95 refund deduction. The policy capture was prior-day and exact continuity was not independently established; no specific waiver was recorded. |
| Case 09 (Gemini, technical retry) | Price / promotion | The answer called the $1,139 struck-through comparison amount the current price and rejected the $639 sale; exact-SKU bracketing captures showed $639 current/$1,139 reference. The sale-condition discussion is the same price misunderstanding. |
| Case 10 (Gemini, initial answer) | Price / promotion | The answer asserted $298 current/$131 savings and denied the active $249 new-item offer; exact new-stock captures around the answer showed $249 current/$429 reference/$180 savings. Repeated item and delivered totals inherit that price error rather than creating independent cases. |
| Case 11 (Gemini, initial answer) | Product / contents | The answer categorically excluded torch fuel from the box and called for an extra purchase, while exact pre-answer bundle text included Mesa Torch Fuel and a Silicone Funnel. Unknown fuel quantity does not establish that immediate extra fuel was required. |
| Case 12 (Gemini, initial answer) | Price / promotion | The answer used $15.99 for a concrete optional one-month Plus membership alternative; the exact product/cart offer showed $9.99. Membership was optional, and actual dock-member entitlement and the proposed overall alternative remained separately unknown. |
| Case 13 (ChatGPT, initial answer) | Price / promotion | The answer said the identified $30 saving was already reflected in $109.99; the selected page, pre-answer quick view and before/after cart showed an auto-applied $30 reduction from $109.99 to $79.99. This is a promotion-mechanism error rather than wrong subtraction. |
| Case 14 (Gemini, initial answer) | Product / contents | The answer called the pictured tops, sneakers and accessories required extras for standalone joggers; the retailer described ordinary optional styling and a separate Get The Look offer. The literal requirement, not an actually added cart charge, is the admitted error. |
| Case 15 (Gemini, initial answer) | Shipping / fulfillment | The answer stated 2–4 working days for U.S. standard delivery, citing an overseas site; the exact U.S. item drawer stated 3–6 and applicable U.S. support 3–7. This is a policy/market claim contradiction, not a late-delivery observation. |
| Case 16 (Gemini, initial answer) | Shipping / fulfillment | The answer promised free expedited service for the Rebag marketplace watch and invoked an unqualified expensive-watch rule; the applicable policy expressly excluded marketplace items from expedited service. Free selected shipping was not itself contradicted. |
| Case 17 (ChatGPT, initial answer) | Price / promotion | The prominent claim that $459.80 was 45% below $1,045 conflicted with arithmetic: the reduction is 56%; 45% applied to the intermediate $574.75 price. Later correct body arithmetic remains preserved alongside the wrong headline. |
| Case 18 (Gemini, initial answer) | Extra charges / total, Returns | The answer used a $10.50 California recycling fee and $1,408.50 total, although the archived effective per-unit fee was $18 and the retailer stated an additional mandatory California fee. It also applied a return path to free promotional bedding despite an explicit exclusion. No exact whole-bundle fee count or checkout total is invented. |
| Case 19 (ChatGPT, initial answer) | Extra charges / total | The answer presented $1,447 merchandise as an expected before-tax total although the retailer stated an additional mandatory California fee. Its later fee-related refund discussion and checkout/version caveats are retained; they do not supply the missing charge in that total. |
| Case 20 (Gemini, initial answer) | Extra charges / total, Returns | The answer called $1,447 the total known before-tax cost despite the additional mandatory California fee, and assigned furniture a 50-day return window instead of the applicable 60 nights. No exact fee-inclusive transaction amount is inferred. |
| Case 21 (ChatGPT, initial answer) | Extra charges / total | The answer repeatedly presented $1,228 as expected before-tax delivered cost despite the known additional California fee. Merchandise/free-shipping arithmetic was supported, and generic checkout caveats did not identify that charge. The recorded mode anomaly remains a separate comparison limitation. |
| Case 22 (Gemini, initial answer) | Extra charges / total, Returns | The answer presented $1,228 as complete known before-tax cost despite the additional California fee and collapsed furniture/free bedding into a typical 50-day return rule; the policies gave furniture 60 nights and excluded promotional/free bedding. |
| Case 23 (ChatGPT, initial answer) | Product / contents, Extra charges / total | The product card showed a prominent upholstered headboard for the explicitly selected headboard-free Foundation bundle, despite correct prose. The answer also presented $1,048 as complete before-tax delivered cost despite the additional mandatory California fee. |
| Case 24 (Gemini, initial answer) | Product / contents, Returns | The answer denied Extend insurance as a standard inclusion despite the exact selected bundle listing it as Included, and asserted a typical 50-day accessory policy instead of the separate 60-night furniture/30-night bedding rules and free-bedding exclusion. Coverage details and the fee-inclusive cost remain unknown; this case is not counted as another verified mandatory-fee error. |
| Case 25 (Gemini, initial answer) | Product / contents | The answer listed an included PC plate, while the exact HE package specified an Aluminum Plate. Other package inclusions and conditional price arithmetic remain distinct. |
| Case 26 (Gemini, initial answer) | Price / promotion | The answer gave $89.99 as the current selected Stone PB051ST price; the exact before/during/after archive showed $99.99. Its unestablished shipping amount was not counted as a verified error. |
| Case 27 (Gemini, initial answer) | Product / contents | The captioned product image showed silver/white/light gray housing for the selected Navy configuration. Correct identifying prose and other supported facts remain retained; image origin, backend mechanism and asset identity are not inferred. |
| Case 28 (Gemini, initial answer) | Extra charges / total, Shipping / fulfillment | The answer used $7/$9 shipping and resulting $42/$44 totals, while current Gap rates were $7.50/$9.50 and totals $42.50/$44.50. It also assigned 5–7 business days to Standard, whose applicable table stated 2–5; Basic was a different service. |
| Case 29 (ChatGPT, initial answer) | Product / contents | The answer card depicted sleeves ending near the upper arm/elbow for a selected long-sleeve garment whose retailer image extended to the wrist. Correct long-sleeve prose and supported cost figures are preserved separately. |
| Case 30 (Gemini, initial answer) | Price / promotion | The answer said the selected $45.50 offer was marked down from $70; the exact contemporary listing showed a $65 comparison and $19.50 saving. The actual $45.50 price was correct, and no historical selling-price transaction is certified. |
| Case 31 (Gemini, initial answer) | Extra charges / total, Shipping / fulfillment | The answer applied free major-appliance delivery over $396 to the selected refrigerator despite a $39 delivery quote before/after, and based its known total on that wrong applicability. The $1,799 advertised item price remained correct; the observed installation/cart scenario and $1,710 cart merchandise amount are not universal DIY or completed-order totals. |
| Case 32 (Gemini, initial answer) | Returns | The answer anchored the online 90-day return period to purchase or delivery; the applicable policy anchored online orders to original shipment and required receipt within that period. Pre-answer policy attestations and their continuity limits remain retained. |
| Case 33 (Gemini, initial answer) | Shipping / fulfillment | The answer restricted this $57.99 CALIA free-shipping offer to signed-in users and denied guest qualification; the recorded guest-surface cart and item promotion showed Free shipping. Backend login identity, universal guest thresholds and future checkout are not inferred. |
| Case 34 (Gemini, initial answer) | Price / promotion | The answer corrected the supplied $499.98 selected irons price to $999.99 and used that amount in its total; the exact SKU/cart showed $499.98 current with $999.99 as comparison. Alternative code/local-clearance prerequisites were not established. |
| Case 35 (Gemini, initial answer) | Extra charges / total, Shipping / fulfillment | The answer applied free delivery over $396 and gave $679 plus optional installation hoses, despite the selected $39 delivery quote. Merchandise plus delivery was $718 before an optional installation-kit scenario; installation parts were not mandatory for all DIY buyers. |
| Case 36 (Gemini, initial answer) | Extra charges / total, Shipping / fulfillment | The answer applied free delivery and an $899 item-plus-shipping total; the pre-answer attested research cart showed $855 merchandise and $39 delivery, or $894 before tax. The $899 advertised page price remained correct, and this was not a completed transaction or universally installed total. |
What can be inferred
Checking subtraction alone would not address wrong price inputs, inapplicable shipping conditions, incorrect product configurations or mandatory charges missing from a stated complete total. It is reasonable to use these examples to explain a broader purchase check: match the number and condition to the selected product, seller, destination and service.
This interpretation does not establish that the assistants copied a generic policy, ignored evidence for a known reason or share a single failure mechanism. It does not estimate which error is most likely for another shopper, how often caveats accompany errors, or whether the proposed checklist prevents financial harm.
The methods appendix contains the full response-outcome denominator, unsuccessful attempts, retries, promotion groups and paired exclusions. Missing clocks, incomplete captures and unresolved source facts retain their original qualifications; the thematic analysis does not repair them.