Skip to main content
How it works · signal pipeline

How PumpStriker works: from wallet buys to a graded signal

PumpStriker's Engine watches what wallets GMGN labels smart money or KOL are buying. A signal needs three or more tracked wallets buying one token within 30 minutes, with real size or a real burst. The Engine checks a fresh price, runs eleven safety kills, grades the signal A, B or C and sends eligible cards to your Telegram chat.

By Mezyan D, Bathyr Devs Last reviewed

GMGN OpenAPI data

Your computer or VPS

PumpStriker Engine

  • Watches tracked wallets
  • Safety kills and grades
  • One decision per token
  • Local dashboard
127.0.0.1 link
shared secret

PumpStriker Telegram bot

  • Up to 3 encrypted wallets
  • One-tap trades
  • One exit engine
  • Cards and alerts
Telegram: your private chatSolana RPCs · Jupiter · pump.fun
Keys are stored encrypted on your machine and every trade is signed there. Signed transactions and your public address go to Solana RPCs and Jupiter, and your bot's messages go through Telegram.

What are the key facts about PumpStriker signals?

Key factsas of November 2026
Data sourceGMGN OpenAPI: the smart-money trade feed every 10 s, the KOL trade feed every 15 s; needs a GMGN API keyLiveNeeds setup
LaunchpadsPump.fun (with PumpSwap and Mayhem launches), Bags, LetsBonk and Jupiter Studio; nothing known to be more than a day old
Trigger3 or more distinct tracked wallets buying one token inside a trailing 30-minute window
Fire linePlus at least 20 SOL bought, or a strong burst (5-minute volume at least 0.15 of 1-hour volume): the same as a score of 85 or more
Price checkA quote at most 30 s old, a trade in the last 10 minutes and a price that is not frozen
Safety killsEleven hard rules, K1 to K12 (there is no K7); an unknown value never kills
GradesA, B or C: the auto-buy safety tier the signal passed; labels, not forecasts
TelegramGrade B or better by default, then your own filters
DeliveryA local, authenticated outbox retried for up to 10 minutes; LATE when sent more than 30 s after the decision
Second lookUnknown safety facts re-checked for 5 minutes after the signal
RecordOne frozen decision per token, with its detection time, inputs, sources, ages and rule version
Live
In the current build and on by default, or on when you turn it on.
Needs setup
Built, but off until you configure it.

What are the steps from a wallet buy to a signal card?

Fourteen steps, in the order the Engine runs them. Each line links to the details below.

  1. Watch

    PumpStriker watches what wallets GMGN labels smart money or KOL are buying, using GMGN data.

  2. Scope

    Only supported launchpads, and nothing known to be more than a day old on its curve or pool.

  3. Cluster

    A token needs at least three different tracked wallets buying within 30 minutes.

  4. Near the line

    Close calls are re-checked quickly.

  5. Fire line

    The rule is published: three wallets, real size or a real burst.

  6. Fresh price

    No decision on stale prices.

  7. Safety kills

    Tokens that GMGN's data flags as honeypots or unsellable, or with live mint or freeze authority, high sell tax or other red flags, are removed before a card is sent. A check with missing data is marked unknown, not safe, and re-checked for 5 minutes.

  8. Labels

    Every signal shows what was checked and what was unknown.

  9. Tiers and grade

    Grades tell you which safety tier a signal passed. They are labels, not forecasts.

  10. Telegram eligibility

    You choose the lowest grade that reaches your chat.

  11. One frozen decision

    Every decision is stored and auditable.

  12. Delivery

    The card arrives in your private Telegram chat; a card sent more than 30 seconds after the decision is marked LATE.

  13. Second look

    If a safety check fails late, the signal is withdrawn.

  14. Afterwards

    Every signal stays on the record.

What does PumpStriker watch?

The Engine reads GMGN's smart-money trade feed every 10 seconds and its KOL trade feed every 15 seconds. It also scans GMGN's new, near-graduation and graduated token lists every 30 to 180 seconds and reads GMGN token signals every 45 to 180 seconds; both scans pause at the leanest budget level. Token signals are recorded as extra facts and never fire a signal by themselves.

Only tokens from Pump.fun (including PumpSwap and Mayhem launches), Bags, LetsBonk and Jupiter Studio are considered. A token out of scope, or whose age is known to be more than one day, is dropped for good. For a migrated token the age counts from its pool, so an older token on a new pool can still qualify.

When does a token become a signal?

Distinct buyers of one token are counted inside a trailing 30-minute window. One buyer is noted, two trigger a first token check, and three trigger a full check. Tokens with three buyers that are just short of the bar get their data refreshed more often, every 30, 60 or 180 seconds, and a new buy pulls the next read forward.

How is the signal score worked out?

Every candidate gets a rule score from 0 to 100, built from published points; unknown inputs never earn points. The score and the tier are frozen the first time the fire line is met. A token under 60 minutes old always scores 85.

Score points (unknown inputs never earn points)
TermPoints
Cluster15 points for 3 or more wallets, 8 for 2; at least 6 when GMGN counts one or more smart or renowned wallets
Quiet and aged8 points: pool at least 165 minutes old and 1-hour volume under $10,000 (24-hour volume when the 1-hour figure is unknown)
Burst and aged12 points: at least 60 minutes old and a 5-minute / 1-hour volume ratio of 0.43 or more; 6 points at 0.25 or more
Launchpad3 points when the launchpad name contains "pump"
Ecosystem3 points for the pump ecosystem
  1. NONE64 or less
  2. WATCH70 to 79
  3. STRONG80 to 84
  4. STRONG85 to 89
  5. ULTRA90 to 99

Below the fire line: recorded, never sent to Telegram.

Fire line at 85: these tiers can reach Telegram when the grade and your filters allow it.

Score tiers from NONE to ULTRA, with the fire line at a score of 85.
Score tiers and what reaches Telegram
TierScoreCan reach Telegram
ULTRA90 to 99Can, if the grade and your filters allow it
STRONG85 to 89Can, if the grade and your filters allow it
STRONG80 to 84No: below the fire line, recorded only
WATCH70 to 79No: recorded only
NONE64 or lessNo: recorded only

Which checks run before the decision?

Fresh price. Before deciding, the Engine takes a fresh snapshot: the pool, the last hour of 1-minute candles, security data if needed, and a quote. It decides only on a quote at most 30 seconds old, for a token that traded in the last 10 minutes and whose price is not frozen. If it cannot get one, it retries for up to 3 minutes and then records the token as unpriceable.

Safety kills. Eleven hard rules (K1 to K12; there is no K7) remove tokens that GMGN's data flags as honeypots or unsellable, or with a blacklist function, live mint or freeze authority, a sell tax of 10% or more, a Mayhem launch, a dev wallet over 25%, top-10 holders at 50% or more, wash trading or a market cap under $5,000. A killed token is recorded and never sent.

Unknown is not safe. An unknown value never kills. If mint authority, freeze authority, sell tax or the top-10 share is unknown, the signal is marked safety pending, which closes every auto-buy tier (grade C). Other unknown checks (honeypot, cannot-sell, blacklist, dev wallet, wash trading, Mayhem) are recorded as data gaps, do not change the grade, and are re-checked for 5 minutes after the signal.

Labels. Every decision also gets named true, false or unknown labels: acting labels used by the auto-buy tiers, measured labels that are recorded only, and quality labels for data problems. Beyond the hard kills and the safety tiers, every decision records more than 30 named risk and data-quality labels, such as insider and bundler share, copycat names and late clusters. They are kept for transparency and auditing and do not change the decision.

The full kill list, with notes

What do grades A, B and C mean?

Four nested auto-buy safety tiers sit on top of the kills: BS, B0, B1 and B2. A tier is open only when none of its blocks is true and every needed fact is known. The grade records which tier a signal passed, frozen at the decision.

Tier BSBase blocks: safety pending, fire line unconfirmed, liquidity unknown, liquidity under 2% of market cap, pool under $1,000
Tier B0Adds: market cap under $10,000, falling knife
Tier B1Adds: 5-minute spike, young pump, launch run
Tier B2Requires: market cap from $10,000 to under $25,000

Grade A · passed every tier, inside the $10k to $25k band

Grade B · passed B1, not B2

Grade C · everything else, including safety pending (hidden in Telegram at the default setting)

A tier is open only when none of its blocks is true and every needed fact is known. Grades are labels, not forecasts.
  • Grade A: passed B2, so every tier, with a market cap from $10,000 to under $25,000.
  • Grade B: passed B1 but not B2.
  • Grade C: everything else, including safety pending. Not shown in Telegram at the default setting.

A grade is a sorting label. It is not a forecast and not a ranking of how a token might trade.

Every block in the tier matrix

Which signals reach my Telegram chat?

By default only grade B or better is marked for Telegram, and never a token under 60 minutes old that already trades 3x or more above its launch price. The bot then applies your own filters: Signals Active, Min Score, the market-cap and liquidity bands and the lowest card grade.

The signal goes to your own bot through a local, authenticated outbox that retries for up to 10 minutes. A signal that cannot be delivered in that time is recorded as expired: it is never shown late or bought, and none is sent twice.

The card shows the grade, the score, the entry price, market data, the safety status, smart-money and KOL counts, the reasons, the exit plan and buy buttons. At most 8 signal cards go out per 10 minutes; the rest wait and none are dropped, while cancels and alerts are not paced. Signal cards arrive around the clock while your machine runs; the optional session window limits only auto-buy, and exits always run.

The filters you control

What is the detection time of a signal?

The detection time is the moment the Engine makes its one decision on a token. It is stored with the decision, together with the inputs, their sources and their ages and the rule version, and every later step is timed from it.

  1. T

    Detection time

    The Engine makes its one decision on the token and stores it with the inputs, their sources and ages, and the rule version.

  2. +30 s

    LATE mark

    A card sent later than this after the decision is marked LATE.

  3. +5 min

    Second look ends

    Safety facts that were unknown are re-checked until here; a late failure sends one cancel.

  4. +10 min

    Delivery window ends

    A signal not delivered by now is recorded as expired: never shown late, never bought, never sent twice.

  5. about +1 h

    First re-read

    Grade A and B signals are re-read for auditing.

  6. +6 h

    Tracking ends

    Live prices for grade A and B signals stop, and every decision is re-read.

  7. about +24 h

    Last re-read

    The last audit read of the decision.

Rules and timings measured from the decision, not to scale. None of them is a result.

Because the decision is frozen with its inputs and their ages, you can always see what the Engine knew at that moment and how old each value was. Selected examples from PumpStriker's signal records, each with its detection time, are on the Hall of Fame page.

What happens after a signal?

  • Second look. For 5 minutes after a signal, safety facts that were unknown keep being re-checked. If one now fails, one cancel goes out and a position that auto-buy opened is sold. Your own buys, including Buy taps on a signal card, are never touched.
  • Tracking. Grade A and B signals are tracked live for up to 6 hours. Exit alerts follow the default plan and never sell by themselves.
  • The record. Each decision is re-read about 6 and 24 hours later (grade A and B signals also after about 1 hour) so the rules stay auditable. Every signal stays on the record, and the dashboard lists every signal with CSV download.

What are the limits of the signal pipeline?

  • A signal or a grade is information, not a forecast, and no tool can make an outcome certain.
  • The trigger is built on public GMGN data that other GMGN users can also see.
  • Data can be missing, late or wrong, and a token can change faster than any check.
  • The checks rely on third-party data and services: GMGN, Solana RPC providers and Jupiter.
  • What your Engine sees depends on its GMGN key, its budget level and your settings.
  • Nothing runs while your machine is off.

Frequently asked questions

Is a PumpStriker signal a forecast?

No. A signal means a token met published rules on GMGN data at one moment: three or more tracked wallets buying within 30 minutes, real size or a real burst, a fresh price and no failed safety kill. The grade shows which safety tier it passed. Neither is a forecast, and no tool can make an outcome certain.

Is the signal score a probability?

No. The score is a 0 to 100 rule score built from published point rules, and it is frozen the first time the fire line is met. A signal needs 85 or more, which is the same as meeting the fire line. The rules are versioned and stored with each decision.

Why did smart money buy a token but no signal came?

A signal needs three or more distinct tracked wallets within 30 minutes plus real size or a real burst, a supported launchpad, an age not known to be over one day, a fresh price and no failed safety kill. A token that misses one of these never becomes a signal; a killed or unpriceable token is still recorded.

What does LATE mean on a signal card?

The card was sent more than 30 seconds after the Engine's decision. Signals wait in a local outbox that retries for up to 10 minutes; a signal that cannot be delivered in that time is recorded as expired and is never shown late or bought.

Can a signal be withdrawn after it is sent?

Yes. For 5 minutes after a signal, safety facts that were still unknown keep being re-checked. If one now fails, one cancel goes out and a position that auto-buy opened in that token is sold. Your own buys, including Buy taps on a signal card, are never touched.

Can other traders see the same data?

Yes. The trigger is built on public GMGN data that other GMGN users can also see. PumpStriker applies fixed, published rules to that data on your own machine.

Sources and review

Last reviewed by Mezyan D, Bathyr Devs. Facts are as of November 2026 and follow the current build; release notes are on the changelog.

Risks and disclosures

  • High risk: Memecoins are extremely high-risk and can move very fast in either direction. Only trade with money you can afford to put at risk.
  • No guarantees: PumpStriker does not promise profit, a win rate or any trading result. A signal or grade is information, not a prediction, and no tool can make an outcome certain. Anything promising guaranteed returns is a scam.
  • Experimental auto-buy: Auto-buy is an experimental, opt-in feature. It is off by default, starts in dry run, and live buys stay inside caps (by default 3 per hour, 20 per day, 5 open positions, a daily SOL cap, a 0.03 SOL reserve and a 15% price-impact cap). The caps are set in your own configuration; a value above the default is accepted and logged as a deviation. Turning it on is your decision.
  • Third parties and partners: PumpStriker relies on services run by others (GMGN, Helius, Jupiter, pump.fun, Telegram), and their uptime, fees and terms apply. UnclaimedSol is a partner: rent-recovery claims made through the in-bot button may earn Bathyr Devs a referral share.
  • No token: There is no PumpStriker or BATHYR token, presale or airdrop. Anyone selling one is running a scam.

Not financial advice: PumpStriker is software, not financial advice. Nothing on this site is a recommendation to buy or sell any token. Do your own research (DYOR). Read the risk disclaimer and the terms of use.