BeforeThe Candle
Solana pAMM graduation research

The market has a moment
that forms before the candle.
We're the ones watching it.

Every pump.fun graduation generates structured, measurable on-chain behaviour in the seconds around pool creation. We capture that tick by tick — bundler patterns, organic arrival timing, recovery profiles, deployer history — and build labelled datasets that don't exist anywhere else.

0+
graduation events
monitored
↑ live & growing
~4%
of graduates sustain
genuine momentum
0+
labelled token
records captured
↑ live & growing
Now
early access
available
↑ taking enquiries

You're not entering bad tokens. You're entering at the wrong moment.

The standard graduation signal fires on transaction count buyRatio. That measurement is wrong — and it's costing you SOL on every session.

The bundler dump problem — why your signal lies
Transaction Action SOL Amount TX 1 BUY 0.08 SOL TX 2 BUY 0.12 SOL TX 3 SELL 4.70 SOL ← bundler TX 4 BUY 0.09 SOL TX 5 BUY 0.11 SOL TX 6 SELL 5.20 SOL ← bundler TX 7 BUY 0.07 SOL TX 8 BUY 0.14 SOL TX 9 BUY 0.10 SOL TX 10 BUY 0.08 SOL buyRatio = 0.80 ← looks strong, signal fires net SOL flow = −9.01 ← SOL is LEAVING the pool

Eight buys, two sells. Standard signal sees buyRatio=0.80 and fires. But the two sells are committing 50× more SOL per transaction than the eight buys combined. The bundler is extracting. You are entering while they are exiting.

This is not bad luck. This is a measurement problem. Transaction count cannot see what SOL-weighted pressure can.

Confirmed across 987 live trades
0.673 SOL
avg sell transaction size at tick 10 — losing trades
Same metric, winning trades
0.127 SOL
avg sell transaction size at tick 10 — winning trades
5× differential. Consistent. Quantified for the first time on T22 PumpSwap pools.

Even when a token is genuine, entry timing determines whether you capture the move or get shaken out before it starts. Every graduation follows the same structural sequence: pool opens, bundlers dump, organic buyers accumulate, momentum builds. The question is where your signal fires in that sequence.

We quantify this with a single field: recoveryRatio — how far the token has recovered from its post-graduation trough at the moment your signal fires.

Trade cluster
Median recoveryRatio at signal
What it means
Losing trades
1.24
Signal fires above graduation price — race already run
Profitable trades
0.36
Signal fires mid-recovery — race just starting
recoveryRatio = (signalPrice − troughPrice) / (graduationPrice − troughPrice)  |  >1.0 = above graduation price
Sample dataset — 10 graduation events · May 29 2026
sol_pressure_at_graduation > 0.70 · net_sol_inflow > 0 · pressure_building = true
Time UTC Session Wallets Buy Ratio Total SOL Pressure (spot) Pressure (15T avg) Net SOL Inflow Avg Buy / Sell SOL Building
21:52US/Asia2380.636422.0 0.9560.854+5.972.086 / 0.145
17:20US1530.601403.3 0.9600.720+2.390.623 / 0.105
20:27US910.636229.0 1.0000.974+7.001.400 / 0.000
22:55US/Asia860.958116.6 0.9580.700+4.724.939 / 0.054
15:43US240.93638.5 1.0000.651+10.472.094 / 0.000
17:16US680.958168.5 0.9700.561+2.840.978 / 0.046
11:54Asia190.65476.6 1.0000.632+7.411.483 / 0.000
17:22US340.80566.5 0.9820.759+6.951.770 / 0.128
04:48Asia570.83084.0 0.9340.703+5.551.989 / 0.211
21:50US/Asia410.61658.0 0.9820.592+6.591.679 / 0.126

Structured, labelled, and not available anywhere else.

The Listener v5 and v6 monitor every pump.fun graduation on PumpSwap in real time. The trading strategy generates structured data as a byproduct — and that data has commercial value entirely independent of the strategy itself.

The strategy is not for sale. The data observations are.

SOL pressure wave archive
Every tick of every position records rollingBuySol, rollingSellSol, solPressureRatio, avgBuySol, avgSellSol, netFlow, and wavePhase. The complete SOL pressure fingerprint of every graduation event — BUNDLER_DUMP → TRANSITION → ORGANIC → REVERSAL — with confirmed outcomes.
Highest value — wave reader
Graduation price journey
graduationPrice, troughPrice, recoveryRatio, priceVsGraduation on every signal from June 2026. First dataset linking the pre-signal price journey to confirmed trade outcomes. Cannot be reconstructed retrospectively — requires a real-time observer at graduation.
New — June 2026
token-context.json
One record per graduated token. Bundler fingerprint, recovery profile, session regime, deployer history, organic arrival timing. Growing daily.
Primary dataset
Labelled signal archive
Every signal — admitted and rejected — with reason codes and confirmed outcomes. Positive and negative examples. ML-ready. The rejected signals are as valuable as the admitted ones.
ML training resource

The same data. Seven different reasons to want it.

We're open to conversations with anyone building in the following areas. You don't need a specific ask yet — if you're curious whether the data is relevant to what you're doing, that's reason enough to reach out.

Solana bot developers
SOL-weighted pressure at graduation. Recovery ratio and graduation price journey fields. Labelled signal archive with negative examples. Stop entering bundler dumps — train your filter on data that measures the right thing.
Quantitative trading firms
Labelled feature matrix for supervised learning — recovery_confirmed, first_organic_buy_ms, bundler_price_impact_pct, session_regime, peak_multiplier. Pre-labelled graduation outcome dataset.
DeFi analytics platforms
token-context.json as a recurring data feed. Pre-labelled, queryable, continuously updated. Build graduation quality indicators without the capture infrastructure.
MEV & sandwich research
Bundler fingerprint dataset at scale. Thin-pool dynamics at graduation across thousands of events. Coordinated wallet behaviour characterised and timestamped.
ML research teams
Outcome-labelled feature matrix as training data. Positive and negative examples with confirmed outcomes. result and moonshot_category as target variables for supervised classification.
Compliance & regulatory
Deployer wallet history with graduation rates and bundler association flags. Empirical, on-chain evidence of coordinated selling patterns at the graduation event.
Crypto research & media
Exclusive access to research findings for a limited window. The T22 organic arrival timing data and bundler fingerprint analysis are both publication-worthy. Co-authored work considered.

There's a person behind the data.

Before The Candle started as a research account — not a data business. The bot was built to trade. The data it generates as a byproduct turned out to have independent value worth developing seriously.

The Twitter account publishes what the data shows — graduation mechanics, bundler patterns, session regime findings, the T22 blind spot that drove a full architecture rebuild. No token calls. No signal reveals. No performance theatre.

The datasets being built here are the same data the research is drawn from. If you follow the account, you already have a reasonable sense of what the data contains and how it was collected. If you don't, that's a good place to start before reaching out.

— @beforethecandle

Let's have
a conversation.

Early access is available now for bot developers and researchers who want to work with the data before the full dataset launch. If the data looks relevant to what you're building, reach out — we'd rather talk early than not at all.

We won't pitch you if it's not a fit. We will respond to genuine enquiries.

Follow @beforethecandle on Twitter / X
Read the research on Substack
Research & data enquiries
Live trading began May 2026.
Data collecting continuously. Early access available for qualified developers and researchers.

Full dataset launch: Q3 2026.
Early access partners receive preferential pricing and schema input.