Setting Sniping Filters That Catch Deals (Not Noise)
Every sniping setup lives or dies on one config screen. Set your filters too tight and they never fire — you'll conclude sniping is dead while other people fill deals daily. Set them too loose and your phone becomes a slot machine that pays out in false alarms. Filter design is the actual skill. Here's how to do it deliberately.
The two failure modes: silence and noise
A filter is a bet about what a real deal looks like. It fails in exactly two directions. Silence: your threshold demands a discount so deep it appears a few times a year on your watchlist, and when it does, a faster buyer takes it anyway. Noise: your threshold is shallow enough that ordinary price wobble trips it, and within a week you've learned to ignore your own alerts — which is worse than having none, because now the one real ping of the month dies in a muted channel.
Most beginners fail toward noise, because a busy feed feels productive. It isn't. The purpose of a filter is to make alerts rare enough to act on every single one. If you aren't willing to open your wallet for a given ping, the filter that produced it is miscalibrated — a mistake covered at length in the beginner mistakes list.
Price-vs-reference: the threshold that does the work
Everything else is decoration; the core rule is "alert me when the listed price is X% below reference." Both halves need care.
The reference must be a number you trust: a liquid mid across venues, not the marketplace's own "suggested price," which is routinely inflated to manufacture discount badges. Deal scanners differ mostly in how honestly they compute this baseline, so audit yours against a few items you know cold before trusting its percentages.
The threshold X depends on what the discount has to pay for. Commonly observed bands, as of late 2026, look roughly like this — treat them as starting points, not gospel:
| Threshold vs reference | What it typically catches | Verdict |
|---|---|---|
| 0–8% below | Ordinary spread, venue fee differences, price wobble | Noise. Fees eat it on resale. |
| 8–15% below | Motivated sellers on liquid items | Real for accumulation; thin for flipping. |
| 15–30% below | Quick-sell listings, cross-venue lag | The working zone. Expect competition. |
| 30%+ below | Mistakes, mispriced specials — or defects and scams | Verify before you celebrate. |
Note the last row cuts both ways: the deeper the apparent discount, the higher the odds something is wrong with the listing rather than right with your filter.
Constraints that cut noise without cutting deals
Once the price rule is set, every additional constraint should remove alerts you would have ignored anyway:
- An item allowlist, not a category. "All knives under reference" is a firehose. Ten to thirty items you personally know how to value is a watchlist. If you can't state an item's fair price from memory within a few percent, you can't act on its alerts fast enough to matter.
- Float windows only where float pays. On finishes where wear is visible and priced, a float band keeps float-tax listings from polluting your feed. On flat finishes, a float constraint just deletes deals.
- Seed and sticker rules as a separate hunt. Pattern sniping needs its own filters with its own references; mixing paint-seed rules into your price feed multiplies noise in both directions.
- A price floor. Alerts on $0.50 items cost the same attention as alerts on $200 items and pay a fraction as much. Set a minimum ticket so every ping is worth the interruption.
- Venue selection. Two or three marketplaces you keep funded beat eight you'd have to top up first. An alert you can't execute inside the deal's lifespan is decoration.
Alert fatigue has actual math
Give yourself a budget: decide how many alerts per day you can genuinely evaluate — for most people with jobs, that's commonly somewhere between three and ten — then work backwards. If your current rules fire fifty times a day, they are, functionally, spam, and your true filter becomes "whichever alert I happen to see," which is the worst filter ever designed.
The discipline that follows: every alert gets a verdict, buy or pass, and every pass gets a one-word reason. A week of reasons tells you exactly which rule to tighten. If most passes say "discount too shallow," raise X. If they say "don't know this item," shrink the allowlist. If they say "was already gone," your problem isn't filters at all — it's latency, and the honest fixes are a faster alert channel or handing the rule-clean cases to automation that executes as well as watches.
Tune like an engineer, not a gambler
Filters are never finished. Prices drift, references move, and a threshold calibrated in a quiet month misfires through a Valve update. The maintenance loop is simple: review weekly, change one parameter at a time, and judge each change by fills and near-misses rather than by vibes. Keep a scratch log of deals you missed and deals you're glad you skipped; the boundary between those two piles is where your threshold should sit.
And separate the two jobs cleanly. Filters that page a human should optimize for judgment calls — ambiguous listings where knowing the market earns the margin. Rules with no ambiguity left in them ("this item, under this price, on this venue") don't need your phone to buzz at all; they need standing enforcement. That split — human filters for judgment, machine rules for policy — is the whole architecture of a modern setup, and it's the same conclusion the broader sniping guide arrives at from the other direction.