Public trace · no login
One bid.
Every step, all shown.
Most PPC tools give you a number. This is the whole decision behind one, from our demo account: the conversion estimate, the limits it had to stay inside, and the checks that let $1.60 become $1.97.
Read the traceAccount settings read at decision time
- Target ACoS
- 30%
- Set by the seller for this marketplace. Strategy: hit my ACoS target.
- Average order value
- $129.99
- Trailing 14 days for this product on Amazon.com.
- Clicks and orders
- 128 clicks · 9 orders
- Trailing 14 days for this keyword.
- Spend and sales
- $182.25 · $1,169.91
- ACoS 15.6%, well under the 30% target.
- Days of supply
- 48 days · lead time 14
- From Amazon inventory data. Feeds the stock check.
- Previous live bid
- $1.60
- The bid live at Amazon before this decision.
agent
Tactician
marketplace
US
keyword
noise cancelling headphones
campaign
[AI-EXACT] B0CW8ZTK4P
The seller picked "hit my ACoS target". The strategy sets how strongly bids react and how far one run may move a bid: at most ±50% on this strategy (±30% on max profit, ±70% on grow).
Tactician1 ÷ (1 + e^(−0.5 × (48 − 14))) = 1.00
The stock check. It compares days of stock (48) with the time a new shipment takes (14). With plenty of stock the multiplier is 1.00, so the bid is not slowed. At 9 days of stock against the same 14-day lead time it would be 0.08, and the bid would fall to a fraction until stock is back.
Sentinel9 orders ÷ 128 clicks = 7.03% · estimate 7.0%
How likely a click is to turn into an order. The estimate starts from the product’s own rate and moves toward what this keyword has actually done. With 128 clicks the keyword’s own record carries it.
TacticianHow sure that estimate is. With 128 clicks the range around 7.0% is narrow, so the engine trusts it fully. With only a few clicks this number is lower, and the bid moves in smaller steps until there is more evidence.
Tactician30% target ACoS × $129.99 order × 7.0% conversion = $2.73
The most a click may cost and still meet the 30% target. No bid for this keyword goes above it, whatever else the model says.
Tactician5% × $129.99 = $6.50
A second, wider fence at 5% of the selling price. It only matters if an input is wrong: a broken conversion rate would make every check above agree with itself, and this one would still stop it.
Tactician($1.38 click today ÷ $1.46 product average) × ($7.42 ÷ $9.14 revenue per click) = 0.77
Is this keyword a good deal right now? It compares what a click costs today with the product’s usual price, adjusted for how much revenue each click brings. This keyword earns $9.14 per click against $7.42 for the product, so its clicks are worth more than they cost. Below 1 means lean up; above 1 means lean down.
Tactician1 + (1 − 0.77) × 0.30 = 1.07
Turns that comparison into a lean of +7%. The 0.30 comes from the strategy: a calmer strategy reacts less, a growth strategy more.
TacticianThe bid live at Amazon for this keyword, read at decision time.
ExecutorThe bid the engine proposes this run. It blends where the bid was heading, the +7% lean above and what a click is worth to this product, and stays under the $2.73 ceiling. The checks below can only stop it, never raise it.
Tactician$1.60 × 0.5 = $0.80 · $1.60 × 1.5 = $2.40 · $1.97 is inside
The step limit: one run may move a bid at most ±50% on this strategy. A bigger change would be cut to the edge, and the rest would have to wait for the next run.
Tactician$1.97 ≤ $2.73
The proposal is checked against the ceiling from the target ACoS. It is $0.76 below it.
Tactician9 orders in 128 clicks
A keyword that keeps taking clicks without a sale is capped and never bid up. This one sells, so the check does not apply.
Tactician$1.97 ≤ 2 × $1.60 = $3.20
A last check against the bid that is live now. A proposal more than twice the live bid would be treated as a fault and cut back.
TacticianTwo switches, one for all marketplaces and one per marketplace. If either is on, no decision is made at all. Checked at the start of every decision.
Tactician($1.97 − $1.60) ÷ $1.60 = +23.1%
Sent to Amazon. The reason the seller reads: "Raising the bid from $1.60 to $1.97. ACoS is 15.6%, well under your 30% target, so there is room to win more clicks." The trace is kept for 30 days and exports as CSV.
Executor
16 steps · 30-day retention
click any row to expand · blue rows changed the outcome
What you are looking at
Mirox re-prices keywords on a cycle. Each time it considers one, the Tactician writes a structured record of the decision: the inputs it read, the intermediate values it computed, every safety gate it checked and in what order, and the action it submitted. That record is the trace. It is stored for 30 days, exports as CSV, and is what you read in the dashboard the moment after a bid posts.
The 16 lines above divide into four kinds. Inputs are read, not decided: the strategy, the stock, the bid currently live. Computations derive one number from others, and each shows its working. Gates can only stop a bid, never raise one. The single decision at the end is what went to Amazon.
The order matters as much as the content. Gates run after the proposal and before the write, so a bid above a limit is cut back before it reaches Amazon, and the trace names the check that did it. The safety page explains each of those checks in plain words.
Why this page exists
Every tool in this category says its AI is explainable. Ask one why a specific bid moved on a specific Tuesday and you will usually get a chart of outcomes rather than an account of the decision. The distinction is not academic: a chart tells you what happened, and only a trace tells you whether it should have.
So this is the version that can be checked. The numbers reconcile, the formulas are printed next to the values they produce, and the CSV is the same export a paying customer gets. If the arithmetic is wrong, it is wrong in public, which is a stronger commitment than any adjective we could put on a landing page.
Want to price the gap between where your account is and where you want it? The waste calculator does that from three numbers, and the ACoS calculator works out the break-even the target ACoS above is measured against.
Questions this raises
- Is this a real seller’s data?
- No. It is our demo account, with sample data, and the same example our films and screenshots use. The steps and checks are the ones Mirox runs on every bid, under the names it writes into the trace. Publishing a real seller’s keywords, conversion rates and order values is not a trade we get to make on their behalf.
- Can I check the maths?
- Yes. Every check shows its arithmetic with this example’s numbers, and every input is listed at the top: the stock multiplier, the $2.73 ceiling from your target ACoS, the step limit, the fence against the live bid. The proposed bid itself blends several signals, so the page states it rather than squeezing it into one line; your own dashboard trace shows every term. If a line does not add up, tell us.
- What happens if a gate fails?
- A check can only lower or stop a bid, never raise it. A proposal above a limit is cut to the limit, and the trace records which check did it and why. The kill switch is checked at the start of every decision, so no decision is made while it is on.
- Do I get this for my own account?
- Yes, from the first day of your free trial — for every bid Mirox places, or, in Simulation Mode, every bid it would have placed without placing it. Traces export as CSV.
- Why publish this at all?
- Because "explainable AI" is a claim every vendor in this category makes and almost none of them will show you. A screenshot proves nothing; a complete trace with checkable arithmetic and a download button is either right or it is publicly wrong. We would rather be the second than be vague.
Your account, your traces
Read one of these
for every bid on your account.
Try Mirox free for 14 days on your real keywords. Every bid gets a trace like this one — and in Simulation Mode, every bid it would have placed, before a cent moves.
14 days free · cancel anytime · connect and disconnect in one click