agentleFS
DeltologicDeltologic/datadoe-ai-skills/skills/buy-box-loss-root-cause/SKILL.md

buy-box-loss-root-cause

Skill8 starsChanged 45 days agoMITView on GitHub

Find the Amazon SKUs losing the Buy Box (Featured Offer) and the most likely reason - price vs the featured offer, out of stock, or fulfilment - so you can win it back. Live from DataDoe. Use for "buy box", "featured offer", "lost the buy box", "buy box percentage", "why did sales drop but traffic did not", or "someone else has my listing".

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 5 are distinctive, and 5 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After a competitor/reseller appears on your listing.

    line 29
  2. Distinctive:

    Check Featured Offer eligibility and whether Amazon judges the price uncompetitive.

    line 58
  3. Very specific:

    Treat your_price = 0 as missing (it is 0 on half to two thirds of the rows), not as a price: for that SKU take …

    line 113
  4. Very specific:

    One marketplace per run: every export in this step, (a) to (d), also filters marketplace_country_code to the scanned marketplace.

    line 158
  5. Distinctive:

    Run these exports instead, all of which fit the cap:

    line 165
  6. Very specific:

    traffic-complete: when asins_with_traffic >= 80% of the median of the 7 complete days before it.

    line 170
  7. Very specific:

    Candidates for (d): every selling ASIN from (c) with an anchor-day row, bb_now < 95, and anchor-day total_page_views >= the candidate floor …

    line 195
  8. Distinctive:

    no such seller -> box not always shown.

    line 261
  9. Distinctive:

    no offers data -> matched price, offers not checked.

    line 264
  10. Very specific:

    Read-only. No auto-repricing (a price change would be bulk-price-update, dryRun-gated).

    line 385

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Buy Box Loss Root Cause

Finds the SKUs where you're losing the Featured Offer (Buy Box) and tells you the most likely reason - price vs the featured offer, out-of-stock, or fulfilment - so you can win it back. DataDoe has no single "buy box" table, so this skill synthesizes it from the buy-box % and price columns.

When to use this

  • Sales dropped on a SKU that still gets traffic (classic buy-box loss).
  • Weekly check on hero SKUs' buy-box ownership.
  • After a competitor/reseller appears on your listing.
  • Trigger phrases: "buy box", "featured offer", "lost the buy box", "buy box percentage", "why did sales drop but traffic didn't", "someone else has my listing".

The framework. The Buy Box gates (check in order)

  1. Ownership - the anchor-day avg_buybox_percentage below the thresholds in Configuration (< 95 = candidate, < 80 = loss) on an ASIN with real traffic = you are sharing or losing the box on the anchor day. Use the last traffic-complete day (the anchor, step 3a) plus a mandatory 7-day trend, never a 30-day average - averaging masks a dip-then-recovery and flags ASINs that already hold the box again. Traffic floor: flag an ASIN only when its anchor-day total_page_views >= 20 (or its per-day total_page_views summed over the 7-day window >= 50) AND bb_now < 95. A low or 0% read on 0-5 page views is "no data", not a loss (on one re-test 52 of 70 ASINs flagged without the floor had zero page views that day). avg_buybox_percentage is never null - it is 0 when there was no featured-offer impression. An ASIN with no row on the anchor day has "no data that day", not 0%. Both columns are per-ASIN values repeated on every SKU row of the ASIN - read them once per ASIN-day (max), never summed across SKUs.
  2. Price - compare your_price with featuredoffer_price (or lowest_price_new_plus_ shipping) on an in-stock SKU, then confirm with the offers on the listing (step 4b).
    • your_price > featuredoffer_price = priced out (+gap).
    • your_price == featuredoffer_price and 0 < bb_now < 100 = matched price, box not always yours. A gap of 0 does not clear you - when you hold the box, featuredoffer_price is your own price - but it does not prove a rival either. The offers decide which of two causes it is:
      • another seller's offer at a landed price (price + shipping) at or below yours -> another seller at your price - an offer/enforcement question, not a price cut;
      • you are the only seller, or every other offer is dearer -> box not always shown - on those page views Amazon showed no Featured Offer (or featured your other offer). Check Featured Offer eligibility and whether Amazon judges the price uncompetitive; do not tell the seller someone shares the listing. On one UK account all six flagged ASINs read gap 0, and the offers showed the seller alone on two and the only rival GBP 2-19 dearer on the rest.
    • your_price == featuredoffer_price and bb_now == 0 with real traffic = lost at matched price - confirm with the offers who holds the box. The offers table is a snapshot that can be one or two weeks old (one account: 11 days before the anchor). Use it whatever its age and print its date; when it is more than 7 days older than the anchor, prefix the cause with "likely". Only with no snapshot at all for the ASIN say "offers not checked" and give both possible causes.
  3. Stock - available = 0 = Amazon cannot feature the offer. Judge it on the latest snapshot (today), not only the anchor day: the buy-box read is lagged, a stockout is a today-problem. An ASIN that ran out often loses its page views too, so the stock check also sweeps every selling ASIN, not only the flagged ones (step 4c).
  4. Fulfilment/health - FBM vs FBA and account-health issues also cost the box (flag as "check" - not in these tables). Residual only: no price row at all, or price and stock both fine and bb_now == 0 with no offers data. Report the first gate that fails, biggest-revenue SKU first.

Configuration

  • MCP base: https://mcp.datadoe.com/mcp/v1
  • Data sources (no dedicated buy-box table - synthesize):
    • Profit by SKU & Date (amazon_profit_by_sku_and_date) [premium] - avg_buybox_percentage (a per-day series), total_sales, total_page_views, total_sessions, total_units_sold, sku, child_asin, product_name. One row per SKU per day; requiresDatePeriod: true (every export needs from/to); refreshed intraday with sales about 1 day behind. Two rules apply to every export in step 3:
      • Per-ASIN traffic aggregation: total_sessions, total_page_views and avg_buybox_percentage describe the child ASIN on that date and are repeated on every SKU row of the ASIN. Group by child_asin (plus date where needed) and take max() of those three; sum() only sales, units and orders. Summing sessions or buy box across SKU rows double-counts.
      • Traffic completeness, per ASIN: the traffic columns come from Amazon's Sales and Traffic report and lag from 2 days to two weeks (Amazon's own note says up to 3 days), with holes of several days, while sales are ~1 day behind. Judge completeness on the per-ASIN rows of this table (step 3a). Do not use the account-level amazon_profit_by_date.total_sessions: it can show full sessions on days whose per-ASIN rows are missing (one UK account: 6,000-7,000 account sessions a day on 26-29 Sep 2026 with no per-ASIN traffic rows at all), so a hole in the data you actually read stays invisible. A child ASIN with traffic but no SKU activity that day comes back as a row with a null sku - variation parents do, but so does a selling ASIN that sold nothing that day, which is exactly what a total buy-box loss looks like. Keep those rows. avg_buybox_percentage is nullable: false: expect zeros, not nulls, when there was no featured-offer impression. Export cap: 5,000 rows CSV / 1,000 JSON - a 30-day daily per-SKU export does not fit on a catalog of more than ~160 selling SKUs, so use the export shapes in step 3.
    • FBA Inventory Health (amazon_fba_inventory_health) [premium] - sku, child_asin, your_price, featuredoffer_price, lowest_price_new_plus_shipping, available, inbound_quantity, units_shipped_t30. A daily snapshot that keeps daily history (requiresDatePeriod: true; from = to = one day gives that day's snapshot). FBA SKUs only - an own FBM SKU on the same ASIN is invisible here. Price lives here, not in the profit table. One day is ~7,500 rows on a UK account, over the 5,000-row CSV cap, so always filter (child_asin in (...), or the stockout filter in step 4c). Treat your_price = 0 as missing (it is 0 on half to two thirds of the rows), not as a price: for that SKU take amazon_listings_with_cogs.listing_price_value ([premium]) and say it is today's listing price, not the anchor-day price. Leave used-condition and amzn.gr.* Grade and Resell SKUs out of the price gate - they are priced below new, so their negative gap means nothing.
    • amazon_item_offers - the offers on the listing: per-offer SellerId, IsBuyBoxWinner, IsFulfilledByAmazon, ListingPrice, Shipping inside the offers JSON, plus summary (offer count, Buy Box prices) and last_seen_at. Your own offers carry your marketplace_seller_id as SellerId. Not premium, no date range; a current snapshot per ASIN that can be a week or more old - state last_seen_at. Filter asin in (...) for the flagged ASINs. Your own FBM offer can be missing from it; the ASIN's non-FBA SKUs and their listing_price_value in amazon_listings_with_cogs show it. Premium note (applies to both premium tables above): a premium export costs 5 AI Tokens instead of 2 - nothing else differs, and the tables are part of the always-on default dataset, so they are never disabled. A 0-row export means no data in the window or an initial load still in progress - say which, and for price fall back to amazon_fba_stranded_inventory.your_price or amazon_listings_with_cogs.listing_price_value (with fba_quantity_available for stock); never render zeros.
  • Thresholds (defined once here; use these words in the output):
    • bb_now < 95 = flag candidate; bb_now < 80 = loss; 80 <= bb_now < 95 = sharing.
    • 7-day trend decides the label: bb7 >= 90 and bb_now < 80 = "just lost it"; bb7 < 90 = "chronically low".
    • Traffic floor to flag: anchor-day total_page_views >= 20, or 7-day >= 50.
    • Candidate floor for the 7-day pull: anchor-day total_page_views >= 5 (the 7-day floor divided by 10, rounded up - low enough that an ASIN with a quiet anchor day but a busy week still reaches the 7-day check).
    • When the user sets a different floor X: X replaces 20, the 7-day floor becomes 2.5 x X, and the candidate floor becomes the new 7-day floor divided by 10, rounded up (floor 10 -> 7-day 25 -> candidate 3). Say which floors you used.
    • 30-day sales window: the 30 calendar days ending yesterday (sales are ~1 day behind), independent of the anchor.
  • Currency/marketplace: localise (e.g. a German marketplace = EUR). The seller list can name the UK marketplace UK while table columns use GB (both seen) - accept either.

Step-by-step workflow (MCP-native)

  1. sellers_and_vendors_list -> pick the seller.
  2. exports_sources_get -> resolve amazon_profit_by_sku_and_date, amazon_fba_inventory_health (both premium, never disabled; if inventory health returns 0 rows, use the price fallback above) and amazon_item_offers.
  3. Buy-box reads from amazon_profit_by_sku_and_date - do not pull the full 30-day daily per-SKU series (groupBy [date, child_asin, product_name] is ~46k rows on a 1,500-ASIN catalog vs the 5,000-row cap; the export truncates inside day 4). Every per-ASIN export below takes max() of the traffic columns and sum() of sales/units. One marketplace per run: every export in this step, (a) to (d), also filters marketplace_country_code to the scanned marketplace. A seller with several marketplaces shares most child ASINs across them, and Amazon can publish their traffic on different days, so an unfiltered probe lets another marketplace's traffic mark a day complete - bb_now is then read on a day with no traffic data for the scanned one. The column holds the ISO code (GB for the UK, not UK); an in filter with the value GB,UK works when unsure. Run these exports instead, all of which fit the cap:
    • (a) Calibration, per ASIN: last 45 days, filters total_page_views > 0 and the marketplace, groupBy [date], countDistinct(child_asin) as asins_with_traffic (~45 rows). A date that is missing from the result had no per-ASIN traffic at all - count it as 0. A day is traffic-complete when asins_with_traffic >= 80% of the median of the 7 complete days before it. Anchor day = the newest complete day that ends a run of at least 3 consecutive complete days; a day standing alone or two days after a hole are not the anchor (one re-test: 09-25 sat alone after a 4-day gap; the right anchor was 09-20). Rows present with total_sessions = 0 are not completeness; the traffic value is. Name the anchor, the traffic lag in days from today, and every hole (incomplete day) in the 30 days before the anchor; every day after the anchor is provisional.
    • (b) Anchor day per ASIN: from = to = anchor day, groupBy [child_asin], max(avg_buybox_percentage) -> bb_now, max(total_page_views) -> anchor-day page views, sum(total_sales), sum(total_units_sold) (one row per ASIN; take product_name from (c)).
    • (c) 30-day sales per ASIN (ranking input): from/to = the 30 calendar days ending yesterday, independent of the anchor - say the dates; groupBy [child_asin, product_name], sum(total_sales), sum(total_units_sold); sum per child_asin in code if an ASIN comes back on two rows (one page on a ~3,300-ASIN account; a larger catalog runs it as groupBy [child_asin] with orderByColumn child_asin - unique - and takes product names from the flagged-ASIN exports). Keep ASINs with total_units_sold > 0 - a flag precondition, not just a ranking input: variation parents carry traffic and never sell, and without this filter 3 of the 4 ASINs that cleared the traffic floor in one test were parents showing a fake 100% loss. No traffic columns in this shape - a max over 30 days returns the busiest day, and a sum double-counts SKU rows.
    • Before the floor (report it): of the selling ASINs from (c) that have an anchor-day row in (b), how many read bb_now < 80, and how many of those had 0 page views.
    • Candidates for (d): every selling ASIN from (c) with an anchor-day row, bb_now < 95, and anchor-day total_page_views >= the candidate floor (5 by default). Pull (d) for this set, not for "the flagged ASINs": the 7-day floor needs (d) first, so pulling it only for flagged ASINs is circular and drops the ASINs that clear only the 7-day floor (one re-test: 12 anchor-day page views / 56 over 7 days, and 10 / 69 - both real; another run's only real loss had 14 / 84). Size on a 1,200-selling-ASIN account: 7-23 ASINs, ~50-160 rows.
    • (d) 7-day trend for the candidates (mandatory): the 7 most recent complete days up to and including the anchor from (a) - skip hole days, and say so when the window had to reach back past a hole - groupBy [date, child_asin], filters: child_asin in (...) and the marketplace, max(avg_buybox_percentage) + max(total_page_views) per ASIN-day -> bb7 = page-view-weighted bb over those 7 days (sum(bb x page_views) / sum(page_views)), to tell "just lost it" from "chronically low" and to confirm the anchor-day read is not a one-day dip on a handful of page views.
    • Flag an ASIN only if it is a candidate AND clears the traffic floor (anchor-day total_page_views >= 20, or the 7-day sum from (d) >= 50). bb_now < 80 = loss; 80 <= bb_now < 95 = sharing, listed below the losses. Label by trend: bb7 >= 90 and bb_now < 80 = "just lost it", bb7 < 90 = "chronically low". Never flag on the 30-day average alone.
    • No anchor-day row: selling ASINs with no row on the anchor day are "no data that day", not 0%. Report their count and 30-day sales as one line - they are unmeasured, not losses (observed: 495-525 ASINs, GBP 9-13k). Stockouts among them are caught by step 4c.
  4. Inventory and offers (amazon_fba_inventory_health: sku, child_asin, your_price, featuredoffer_price, lowest_price_new_plus_shipping, available, inbound_quantity, units_shipped_t30):
    • Latest date first: from = 3 days ago, to = today, groupBy [date], count(sku) -> the newest snapshot date (usually today).
    • (a) Anchor-day snapshot: from = to = anchor day, child_asin in (...) for the flagged ASINs and the stockout-sweep ASINs from (c) - the price pairs with bb_now (same moment), and the stock tells whether a stockout already existed on the anchor day.
    • (b) Offers: amazon_item_offers, asin in (...) for the flagged ASINs, item_condition New - parse offers per ASIN: who holds the Buy Box, every other seller's landed price (price + shipping) next to yours, and any second offer of your own (an FBM SKU on the same ASIN, sometimes cheaper than the FBA one). State last_seen_at.
    • (c) Stockout sweep, latest snapshot: from = to = latest date, filter available = 0 and the marketplace - no velocity predicate: Amazon's units_shipped_t30 can read 0 on a SKU whose ASIN really sold (on one UK snapshot 43 selling ASINs were fully out of FBA stock with units_shipped_t30 = 0) - columns sku, child_asin, available, inbound_quantity, CSV limit 5000, orderByColumn sku (one row per SKU per snapshot, so the order is unique; page with skip if it fills - 3,827 rows = 1 page on that account). Keep the ASINs that sold in step 3c (total_units_sold > 0) - the real sales decide, not Amazon's velocity - then pull the latest snapshot for those ASINs and the flagged ASINs (child_asin in (...), in chunks if the list is long) to check their other SKUs: an ASIN is out of stock when every SKU with a non-null available is 0. Report those as stock rows whatever their buy-box read: an ASIN that ran out often drops to 0 page views (one account's best stock catch went from 30-45 page views a day to 0), so no traffic floor can ever admit it, and a 100% read on the anchor day can sit on the last unit. If amazon_listings_with_cogs shows an Active non-FBA SKU on the ASIN, say "FBA out of stock; FBM SKU {sku} still listed".
  5. Join inventory onto the ASIN with matching time bases: pair the anchor-day price snapshot with bb_now (same day), never the 30-day average or a snapshot taken days later. amazon_fba_inventory_health can have several SKUs for one child_asin - do not collapse those rows or average your_price; show each in-stock SKU's price row. Sales stay per ASIN (the 30-day export in step 3c is grouped by child_asin).
    • Price (anchor-day snapshot): for each in-stock SKU (available > 0; a null available is unknown, not 0) with a price (your_price > 0, else the listings price as above), gap = your_price - featuredoffer_price. Then:
      • gap > 0 -> priced out (+gap); fix: match/beat featuredoffer_price.
      • gap == 0 and 0 < bb_now < 100 -> read the offers from 4b:
        • another seller at or below your landed price -> another seller at your price; fix: check that seller in the offers list - an enforcement/offer question, not a price cut;
        • no such seller -> box not always shown; fix: check Featured Offer eligibility and whether Amazon considers the price uncompetitive; if your own FBM offer is cheaper than the FBA one, say that Amazon may be featuring it instead;
        • no offers data -> matched price, offers not checked; give both causes.
      • gap == 0 and bb_now == 0 (traffic floor cleared) -> lost at matched price; name the Buy Box holder from 4b if there is one.
    • Stock (latest snapshot, 4c): out of stock today -> stock (as of {latest date}) with the full 30-day sales exposed - even when the anchor-day read was fine. If the anchor-day snapshot already showed 0, say "out of stock since at least {anchor}"; otherwise "ran out after {anchor}". Show inbound_quantity. List the top 10 stock rows by sales exposed and give the rest as one line with their count and total (one account had 79 stock rows, 53 of them under GBP 50).
    • Fulfilment/health check: residual only - no price row at all, or price and stock both fine and bb_now == 0 with no offers data.
  6. Rank by sales at risk - 30d sales x (1 - bb_now/100), or the full 30d sales on a current stockout - across losses, sharing and stock rows, and render.

Output format

Buy Box Loss - {marketplace} - buy-box as of {anchor day} (traffic lag {n}d{, holes: dates with no per-ASIN traffic}; later days provisional)  (sales: 30 calendar days ending {yesterday})
Traffic floor: {pv} page views on the anchor day / {pv7} over 7d{, 7-day window reaches back past a hole} · thresholds: < 80 loss, 80-94 sharing
Before the floor: {b} selling ASINs read < 80% on {anchor day}, {z} of them on 0 page views. After the floor: {l} losses, {s} sharing.

SKU                 BB%(now) BB%(7d) PV   Sales    Likely cause                               Fix
{sku}               {bb}%    {bb7}%  {pv} {cur}..  priced out (+{cur}gap)                     match/beat {cur}{feat}
{sku}               {bb}%    {bb7}%  {pv} {cur}..  another seller at your price ({seller})    check that offer (enforcement, not a price cut)
{sku}               {bb}%    {bb7}%  {pv} {cur}..  {likely }box not always shown (you are the only/cheapest seller{, offers of {date}})   check Featured Offer eligibility
{sku}               {bb}%    {bb7}%  {pv} {cur}..  lost at matched price                      check who holds the box
{sku}               {bb}%    {bb7}%  {pv} {cur}..  fulfilment/health (residual)               check FBM/AHR

Sharing (80-94%), by sales at risk: {sku} {bb}% / {bb7}% - {cause} · {cur}{at risk}
Out of stock today ({latest date}) - full 30d sales exposed, whatever the buy-box read (top 10):
{sku}  {product}  0 units (inbound {n})  {cur}{sales 30d}  {out since at least {anchor} | ran out after {anchor}}{ · FBM SKU still listed}
+ {n} more out-of-stock ASINs, {cur}{total} of 30-day sales
Today's stock for the flagged ASINs: {all in stock | list}. Offers checked: {last_seen_at}.
Unmeasured: {k} selling ASINs ({cur}{sales}) had no row on {anchor day} - no read, not losses.
Biggest sales at risk: {sku} ({cur}.. exposed).

Worked example (illustrative)

The calibration counts ASINs with traffic per day: complete through yesterday-minus-two, with four days missing entirely the week before - so the anchor is two days back, and the 7-day trend reaches back past the hole. Reading each ASIN's anchor-day buy-box % (not a 30-day average), confirmed by the 7-day trend and the traffic floor, surfaces the ones losing the box with real traffic - an ASIN that dipped mid-month but sits at 98% on the anchor day is not flagged, and neither is one at 75% on 4 page views. An ASIN with three SKUs shows the same 41% and 310 page views on each SKU row: read once (max), not summed to 930 page views. The skill then joins the anchor-day prices: if your_price 12.90 > featuredoffer_price 11.95, the cause is "priced out by 0.95". If your_price 23.49 == featuredoffer_price 23.49 at 69%, the offers decide: here the only other offer is an FBM seller at 26.48 + 15.50 shipping, so nobody shares the listing at your price - "box not always shown", a Featured Offer eligibility check, not an enforcement case. And the stock sweep finds an ASIN that held 100% on 45 page views with its last unit on the anchor day and 0 today - a stock row with the full 30-day sales exposed, which no buy-box threshold would have flagged. Same signal, different fix - the skill picks the right one.

Quality self-check

  • Did I judge completeness on the per-ASIN rows (countDistinct(child_asin) with page views > 0, missing dates = 0) for the scanned marketplace only, anchor on the newest complete day ending a run of 3+, and name the anchor, lag and every hole?
  • Did I apply the traffic floor (anchor-day total_page_views >= 20 or 7-day >= 50, or the scaled floors when the user set one) so a 0% read on a handful of page views is "no data"?
  • Did I take max() of avg_buybox_percentage / total_page_views / total_sessions per ASIN-day and sum() only sales and units?
  • Did I keep null-sku traffic rows of selling ASINs (a total loss looks like that) and use 30-day units sold to drop parents?
  • Did I report the before-the-floor count on a stated base (selling ASINs with an anchor-day row)?
  • Did I pull the 7-day trend for the candidate set over the 7 most recent complete days, skipping holes?
  • Did I check the offers before naming a cause on a matched price - "another seller at your price" only when such an offer exists, "box not always shown" when I am alone or cheapest?
  • Did I run the stockout sweep on every zero-stock SKU without a velocity filter, keep the ASINs that really sold (step 3c), check their sibling SKUs, and expose the full 30-day sales on a current stockout?
  • Did I use the defined thresholds and the pinned 30-day window (ending yesterday)?
  • Did I check price, offers and stock before blaming "fulfilment"?
  • Did I rank by revenue at risk across losses, sharing and stock rows, not by lowest bb%?
  • Did I keep margin in mind (winning the box below cost isn't a win)?

Common mistakes

  • Chasing buy-box on tiny long-tail ASINs (bb 0 but 1 unit/mo) - not worth it.
  • Recommending a price cut when the real cause is a stockout.
  • Averaging buy-box % across the window - avg_buybox_percentage is a daily series; an ASIN that dipped and recovered reads as a chronic loss.
  • Summing total_sessions, total_page_views or avg_buybox_percentage across the SKU rows of one ASIN - they are per-ASIN values repeated on every SKU row.
  • Running the step-3 exports without a marketplace filter on a multi-marketplace seller - another marketplace's published traffic marks a day complete, and every ASIN of the scanned one reads as a buy-box loss on it.
  • Comparing a 30-day average buy-box to a single-day price snapshot - two different time bases; pair the anchor day with the anchor day.
  • Pulling the full 30-day daily per-SKU series - it blows the 5,000-row cap and silently truncates after a few days.
  • Calibrating on amazon_profit_by_date.total_sessions - the account total can look complete on days with no per-ASIN rows, so the 7-day trend silently loses days and the header says "no gap".
  • Anchoring on sales completeness, or on the latest day present when it stands alone after a hole.
  • Telling the seller "another seller shares your listing at your price" from a gap of 0 alone - on one account the seller was alone or cheapest on every such ASIN. Check the offers first.
  • Checking stock only for flagged ASINs - an ASIN that ran out loses its page views, so it never reaches the flag list; sweep today's stockouts on every selling ASIN.
  • Pre-filtering the stockout sweep on units_shipped_t30 > 0 - Amazon's velocity can read 0 on a SKU whose ASIN really sold; let the profit-table sales decide.
  • Dropping null-sku rows on the anchor day - a selling ASIN with traffic and no sale that day comes back that way, and a total loss is exactly that.
  • Pulling the 7-day trend only for already-flagged ASINs - circular with the 7-day floor.
  • Choosing your own "low" threshold - < 95 candidate, < 80 loss, as defined in Configuration; a different cut gives a different list.
  • Expecting null avg_buybox_percentage - it is 0 when there was no featured-offer impression. A sole-seller ASIN can read below 100% when Amazon shows no Featured Offer on some page views - that is not a rival.
  • Reporting an ASIN that has no row on the anchor day as 0% - it has no data that day.
  • Racing a hijacker to the bottom instead of enforcing (flag for enforcement).

Notes

  • Read-only. No auto-repricing (a price change would be bulk-price-update, dryRun-gated).
  • Buy box is synthesized (no first-class table) - state assumptions in the output.
  • Inventory Health covers FBA SKUs only; own FBM SKUs (often suffixed _MFN / _Virt) show up in the offers and in amazon_listings_with_cogs, not in the price and stock gates.
  • A DataDoe skill, built on DataDoe buy-box %, price, offers and inventory columns.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.