If your monitor spent a week buzzing and you messaged almost nobody, the fix is rarely "check Marketplace more." The fix is reading the junk pattern and tightening the filter that actually failed.
Most people cut the price cap first. That often deletes the deals and keeps the parts cars. Start with the failure mode you can name.
Step 1: Sort last week's junk into three buckets
Open the alerts you ignored or regretted. Put each one in one bucket only:
- Wrong identity: wrong model, parts car, accessory, or unrelated listing that matched a loose keyword
- Wrong economics: price or year outside what you will actually buy
- Wrong logistics: too far, dealer lot you will not visit, timing you cannot meet
If a listing fits two buckets, pick the reason you rejected it in the first five seconds. That is the filter to touch.
Step 2: Fix identity noise with excluded terms
Identity junk is where excluded terms earn their keep. Words like "parts," "wanted," "shell," "part out," "motor only," or "down payment" repeatedly waste triage time.
Add the phrases that showed up in real junk, not a fantasy blocklist of fifty words. Then re-check for a day. Why excluded terms beat tighter price caps covers the logic. The short version: exclusions remove known trash without shrinking the price band where underpriced cars appear.
For vehicles, also confirm make and model are constrained when you do not want every listing that mentions the brand once in the description. A make-only monitor is wide on purpose. If you meant Camry, say Camry.
Step 3: Fix economics with year before price
If the bucket was wrong economics, adjust year floor and ceiling before you slash the max price. A $4,500 car from 2004 and a $4,500 car from 2016 are not the same buy. Year cuts age risk without pretending every good deal sits $500 lower.
Use price caps for the money you can actually spend. Use year to stop negotiating with cars you will flip down the road. Concrete setup patterns: price and year filters that cut noise, not deals.
Step 4: Fix logistics with cities and radius
If you keep rejecting cars you will never drive to, shrink the map. Drop metros outside this month's travel radius. Turn off all-cities until the short list is clean. Wider coverage is for scarce inventory after filters are already strict, not for drowning a noisy buy box.
Step 5: Change one dimension, then wait
Do not retune five knobs at once. You will not know what helped.
- Ship one change
- Let new ingests run for at least a day of normal fan-out
- Count alerts that earned a seller message versus alerts you killed in under thirty seconds
- Only then touch the next dimension
Crawlbench match reasons are written at ingest time. Filter edits do not rewrite old alerts. Judge the tune-up on new matches only.
A one-week checklist
Day 1: bucket the junk
Day 1: add two to five excluded terms from real trash
Day 2: raise or lower year bounds if age was the reject reason
Day 3: trim cities you will not visit
Day 4: adjust price only if economics junk remains after year and exclusions
Day 5–7: leave it alone and measure message rate
If you need a clean baseline, rebuild from a minimal vehicle monitor setup and layer constraints back in with evidence.
Closing
Junk week is feedback. Read the rejects, tighten the dimension that failed, and resist the urge to punish the price cap for a parts-car problem. Quiet, useful alerts beat a loud feed you have trained yourself to ignore.