Live API verified The queue grouped work by buyer. A buyer who has a message thread and received one delivery link reads as finished. A second receipt from that same buyer inherits the finished state and never surfaces. One buyer holding two orders was covered by a single card, and the safety-net table built to catch exactly this listed nine buyers without listing them.
Live API verified The fix is to key on receipt id and join the Etsy username. Each paid receipt becomes its own row, so a delivered first order cannot absorb an undelivered second one. The username comes from the convo_history-url attribute on the Orders page and survives display-name changes, which is what defeated name matching when five different buyers all appeared as "Sign in with Apple user".
Live API verified Nine buyers hold more than one receipt. Three of them had no row before this pass. Coverage stated plainly: the username map resolves 47 of 63 receipts because it was read from three pages of Shop Manager, and the remaining 16 are joined on receipt name, which is the weaker test.
Shop Manager observation Buyer intake also arrives on the receipt, not only in messages. Etsy exposes a buyer note field at checkout. A queue built from message threads cannot see it. Two buyers wrote their full intake there and were treated downstream as having supplied nothing, so one card sent the seller to ask a customer for something he had already provided 33 days earlier.
Shop Manager observation Two documents generated the same day disagreed on six rows. The later one had settled those rows by opening each buyer's own thread through the order, a test that returns either the thread or the string "0 conversations between you and X" and cannot be fooled by a display name. Acting on the earlier document alone would have meant redoing six deliveries that had already gone out. When two builds disagree, the one carrying per-order thread evidence wins.
Live API verified 69 of 71 units sold at $1. The other two were one $13 and one $10 purchase. Merchandise revenue across all units is $92.00. Verified Jul 16, 2026 at 7:48 PM ET.
Live API verified 3 reviews across 63 orders, which carried 71 units. Stating it as "3 reviews from 71 orders" conflates two units and overstates the denominator.
Live API verified 7 receipts carry is_shipped=false. The flag records whether a button was pressed. One order reads Completed with 2 of 3 readings undelivered, and another reads unshipped after delivery by message, so the flag is used here only to decide which orders are worth opening.
| Category | Open items | Basis for its rank |
|---|---|---|
| Live complaint, never delivered | 1 | |
| Never delivered | 13 | |
| Paid, undelivered | 4 | Money taken, work outstanding. Ordered oldest first. |
| Contacted, delivery unproven | 4 | |
| Admin close | 1 | Delivered, one artifact outstanding. |
| Courtesy, unpaid | 1 | Goodwill, not an obligation. |
| Awaiting buyer | 1 | The next move belongs to the buyer. |
| Uncertain paid fulfillment | 1 | Money taken, delivery unproven. Outranks a known complaint. |
| Open complaint | 2 | The failure is stated and unresolved. |
| Evidence retrieval | 1 | Blocks the recoveries beneath it. |
| Account enforcement | 1 | Consequences reach past one conversation, so it leads. |
Ranking basis. Account enforcement outranks everything because its consequences reach past one conversation. Uncertain paid fulfillment outranks visible complaints, because a complaint is already known and a silent unresolved order is not. Evidence retrieval outranks the recoveries that depend on it. There is no score: the data carries no fulfillment minutes, labour cost, or per-order profit, and a number would imply precision that does not exist.
Each row is the live value or the report's own arithmetic, not a re-reading of the report. Two of these correct the previous version of this page, not the PDF.
| Earlier report said | Verified value | Consequence |
|---|---|---|
| Earlier report said, incorrect a customer is owed a paid reading | Her paid order was delivered Jul 14. The a third party reading is unpaid courtesy, offered in exchange for a review. | Recategorised from Paid-undelivered to Courtesy-unpaid, and dropped from rank 4 to 12. The review condition is removed: a customer prohibits review manipulation. |
| Earlier report said, incorrect Are They Watching You now sells at $16 | $13.00, verified Jul 16, 2026 at 7:48 PM ET | Prices were inherited rather than re-pulled. Every price on this page now carries a verification time, and historical paid price is a separate field from current listing price. |
| Earlier report said, incorrect a customer: two orders at $1 each | Three purchased units across two orders. She bought the same listing twice. | Her open Help Request may be literally true. Reconciliation outranks interpretation, so she moves to rank 3. |
| Earlier report said, incorrect a customer older order dated Jul 12 | Jul 10, 11:10 PM ET | Does not change the recovery, and it weakens trust in every hand-copied date. All timestamps here come from the live receipt. |
| Earlier report said, incorrect Old API-open orders, all resolved in Shop Manager | 7 receipts carry is_shipped=false. A $10 order sits unshipped at 14 days. | The heading and its own 'Verify in inbox' row contradict each other. The $10 order moves from a footnote to rank 2. |
| Earlier report said, incorrect Every corrected file normalised to -16 LUFS | The corrected exports measure -16.5 to -19.1 LUFS | The report's own table contradicts its claim. The measured range replaces the asserted target. |
| Earlier report said, incorrect Recordings run 12 to 19 dB below target | 14 to 19 dB below, computed from the same table | Arithmetic check against the report's own numbers. |
| Earlier report said, incorrect Sharing permissions are fine, no Drive problem exists | 10 of 10 links passed anonymous access testing at the recorded verification time | The test proves present accessibility of those links. It cannot prove what a buyer met at delivery. |
| Recording | Length | Measured | Below the -16 LUFS spoken-word target | Corrected export | Buyer response on record |
|---|---|---|---|---|---|
| a customer - a third party reading | 7:23 | -34.9 LUFS | 18.9 dB | -17.9 LUFS | Quietest of the ten. She complained. |
| a third party | 5:39 | -33.7 LUFS | 17.7 dB | -16.5 LUFS | No complaint on record |
| a customer - a third party | 5:13 | -33.1 LUFS | 17.1 dB | -17.7 LUFS | No complaint on record |
| a customer - original | 2:34 | -32.5 LUFS | 16.5 dB | -19.1 LUFS | She complained. |
| a customer - reading 1 of 3 | 12:25 | -32.0 LUFS | 16.0 dB | -18.2 LUFS | No complaint on record |
| a third party | 2:08 | -31.8 LUFS | 15.8 dB | -16.6 LUFS | No complaint on record |
| a customer - REDO | 11:11 | -31.0 LUFS | 15.0 dB | -16.8 LUFS | 1.5 dB above the file she could not hear |
| a customer - prior | 9:47 | -30.9 LUFS | 14.9 dB | -17.0 LUFS | No complaint on record |
| a customer | 4:52 | -30.0 LUFS | 14.0 dB | -16.9 LUFS | No complaint on record |
| a third party | 5:04 | -30.0 LUFS | 14.0 dB | -16.7 LUFS | No complaint on record |
Mac2 audio evidence All 10 recordings measured between -34.9 and -30.0 LUFS, which is 14 to 19 dB below the -16 LUFS spoken-word delivery target. 2 of 10 buyers independently reported they could not hear the recording.
Mac2 audio evidence The corrected exports measure -16.5 to -19.1 LUFS. The earlier report describes them as normalised to -16 LUFS. Its own table does not support a single target, so the measured range is what this page reports. Every corrected file is substantially louder and more usable than its original, and none of them sits exactly at -16.
Unverified hypothesis That low volume explains the review rate. 3 reviews across 63 orders does not establish why other buyers declined to review. To settle it: repeat-purchase behaviour, buyer replies, how many buyers request their money back, and audio complaints, compared before and after corrected files reach buyers. With 3 reviews, this will likely stay unsettled.
Unverified hypothesis That silent buyers strained through it or quietly stopped returning. 8 of 10 said nothing about audio. Silence evidences neither strain nor satisfaction. To settle it: the corrected re-send is a natural test. A response from a previously silent buyer is signal. Nothing before that is.
Inbox verified All 10 Drive links passed anonymous access testing at the recorded verification time. That establishes present anonymous accessibility for those specific links. It does not establish what any buyer encountered at the moment of delivery.
Mac2 audio evidence A -31 LUFS export means recording at roughly a sixth of the intended level, so the gain staging is the defect and normalisation is the patch. Fix the source, then apply on every export:
ffmpeg -i input.m4a -af loudnorm=I=-16:TP=-1.5:LRA=11 -c:a aac -b:a 160k output.m4a
build_v2.py. Companion page: drwu-htmls.vercel.app/etsy-truth.