Gradbook ($GRADS) · tekmaker

TL;DR

On pump.fun, "graduated" is the single most-watched label in the trenches. Every terminal (Axiom Pulse, GMGN Trenches, Photon Memescope, BullX Neo Vision) has a column of coins that just completed their bonding curve and migrated to PumpSwap, and migration snipers fire within seconds of the pool appearing. But graduation stopped meaning "organic demand" a long time ago. On 2026-10-09, about half of the graduates we enrolled bonded within two minutes of launch, most of them in the same block they were created. Only around one in ten was still at or above its graduation cap a few hours later.

Nobody shows traders what happens after graduation, as base rates. The terminals show a live list. Studies publish a one-off number. Gradbook is the yearbook for pump.fun graduates:

Every coin that bonds is enrolled in that UTC day's class.
Each one is graded against the fixed cap it graduated at (410.9 SOL, about $45K today) at +1h, +6h and +24h, from pump.fun's own 5-minute candles.
Each gets a status: Top of class, Standing, Slipping, Dropped out, Paper cap or Off-curve.
Each class gets superlatives (Valedictorian, Fastest dropout, Most creative accounting), a survival rate, and a page in a class archive.
The Odds desk turns every graded graduate into base rates: "of graduates that were at 0.4–1× their cap one hour in, X% were still above it at +6h". These are split by speed-bonded vs organic.

Token: $GRADS (reserved, not launched). Creator fees pay the data bill. Holding it unlocks alerts and the deeper odds tables, checked read-only. The token is also the brand of a meme-able format ("Class of Oct 9: 596 graduates, 11% still standing").

The user and the pain

Who: the pump.fun and PumpSwap trader who works the "Migrated"/"Graduated" column. They buy coins minutes to hours after migration on Axiom, GMGN, Photon, Padre or a Telegram bot, usually at $40K–$500K market cap with 0.2–3 SOL. A second user is the creator or KOL who wants to show their coin "made it": still standing a day later, honor roll, valedictorian.

The pain:

Graduation is a gamed signal. A coin can be created and bonded in one transaction by its own deployer. The "graduated" badge then appears in every terminal and the deployer sells into whoever trusts it. We watched this live: one j7tracker deployer created a coin, bonded it with about 8 SOL, sold from 5 wallets within 3 seconds, and repeated an hour later.
No base rates. A trader looking at a 40-minute-old graduate at 0.6× its graduation cap has no idea how often that recovers. Terminals show the snapshot (holders, top-10 %, bundle %), never the outcome history of coins that looked the same.
No memory. The Migrated column scrolls away. Yesterday's graduates and what happened to them aren't kept anywhere a normal trader can browse.
Wrong numbers everywhere. pump.fun's own usd_market_cap freezes at the curve value after migration. DexScreener lags on new pairs (we saw $53K displayed while pump.fun trades were at $3K). Raw ATHs are painted by single trades or wash volume ($35M "ATH" on a coin that never held $250K).

Evidence

Post-migration survival is low and measurable. Scorp Labs tracked 15,548 graduations from 2026-05-31, counting a coin alive if it held ≥$5K liquidity. Survival was 56.5% at +5 min, 43.8% at +30 min, 25.8% at +6h and 19.7% at +24h. Median liquidity fell 57% between minute 5 and minute 30. (Scorp Labs, 2026-07-17) This is a one-off blog table, not a product.
Most graduates end far below migration. Belphor data on 2,891 migrations, cited by Memecoinist, found only 6.2% above their migration price at the latest reading, and about 91% at ≤20% of it. (Memecoinist, 2026-09-25) The Belphor primary page is a JS app we could not fetch, so we treat it as secondhand.
Self-bonding is common and measured. pumpfundata's April 2026 cohort had 6,379 graduates. 1,513 (≈24%) bonded in under 60 seconds and 735 in the same block they were created ("the creator wallet filled the entire bonding curve alone"). Bot graduates had a median of 6 unique buyers, against 312 for real ones. (pumpfundata, 2026-07-16)
Academic work stops at graduation. The CMU/EPFL "Meme Coin Factories" paper covers 15.2M coins: 1.02% graduate, 17.7% of graduates are copycats, creator clusters made 62.99% of coins, and wash-traded coins graduate at 2.0% vs 0.90% (arXiv, 2026-09-09). Kamat measured a median time-to-graduation of 1.66 minutes for observed graduations (arXiv v3, 2026-08-17). Neither measures what happens after migration.
The graduation rate itself swings widely. It was 0.26% on a 7-day average in June (The Block, 2026-06-16) and 6.7% on a single day after BOOST (The Block, 2026-07-29). There were about 1,100 graduations a day in mid-September (Memecoinist citing MemeFees, primary source unverified). Our memeradar digest for the 24h to 2026-10-09 14:19Z counted 1,050 graduates out of about 54,853 launches (1.9%). 317 of them were instant self-bonds (ATH within 2 min of creation, then −50%+).
Traders demonstrably trade this column, and pay for it. Axiom Pulse has a "Migrated" column plus an in-house migration sniper (Axiom docs, migration actions). BullX Neo Vision has "Graduated: the most recently migrated tokens" (BullX docs). Photon Memescope's Graduated column showed tokens "5, 8 and 11 minutes old" (Photon guide). Fees are about 1% per trade: Axiom 0.75–0.95% net, Photon 1%, GMGN 1% (CoinCodeCap, 2026-07-21).
Snipers watch migrations. After BOOST, one sniper wallet appeared in 13 of 14 traced curve completions, firing 0–6 s after pool creation (Onchain Divers, 2026-07-24). Bundlers are sold openly at 0.3 SOL per launch or 1% of volume (The Coin Republic, 2026-09-25).

Market now

Live numbers from Gradbook's own backend and the memeradar digest, 2026-10-09 (UTC):

Graduations: 1,050 graduates in the 24h to 14:19Z, against about 54.9K launches (1.9% bond rate). Organic graduates were bonding at 17–55 an hour through the day.
Today's class (Gradbook backend, 15:06Z, partial day): 620 enrolled. 47% bonded within 2 minutes of launch (294), and most of those were "bonded at birth" in the creation transaction. 197 (32%) were off-curve: they "graduated" on a non-standard curve at caps of a few dollars to a few hundred dollars, so we don't grade them. Of the graded ones, about 10% were at or above their graduation cap at that moment, and 14% were still standing one hour after graduating (n=177). Yesterday's late graduates (Oct 8, n=65 with both checkpoints): 11% standing at +1h, 6% at +6h.
Painted spikes and paper caps: 35 of today's graduates printed a raw high 8×+ above anything they held for 15 minutes, and 21 trade at a paper cap right now. Several "dividend fund" clones showed caps of $14M–$462M on thin pools.
Venue context: PumpSwap did $11.33B over the last 30 days, about 15% of Solana DEX volume, and the pump.fun bonding curve did $3.87B (DefiLlama, accessed 2026-10-09). Solana spot DEX volume was $197B in Q3 2026 (Crypto Briefing, 2026-10-02).
Fees: pump.fun's fee page (updated 2026-10-08) lists PumpSwap fees tiered by market cap, with the creator fee at 0.95% in the 420–1,470 SOL band, which is right where graduates trade (pump.fun docs). pump.fun's protocol revenue was $40.0M over 30 days on DefiLlama (API, accessed 2026-10-09).

Competitors and the gap

Axiom Pulse (about 1% per trade, 0.75–0.95% net): live New / Final Stretch / Migrated columns, filters and a migration sniper. Gap: no outcome history, no base rates, no cohort view.
GMGN Trenches (1% per trade): New / Almost bonded / Migrated columns. Gap: same, a live snapshot only.
Photon Memescope and BullX Neo Vision (1% per trade): a Graduated column. Gap: same.
comparememecap survival tracker (free): % of graduates from 24h / 7d / 30d ago with ≥$5K liquidity. Gap: showed "Collecting data…" when checked; no +1h/+6h, no self-bond split, no per-coin page; liquidity-only definition.
Scorp Labs blog (free): a one-off survival table (+5m / +30m / +6h / +24h). Gap: a static post, not a live product.
pumpfundata ($20–$50 per credit pack): raw data API plus a one-off April analysis. Gap: a data vendor, not a product.
Dune dashboards (free): migration counts and graduation rate. Gap: counts only, no post-migration outcomes.

The gap: nobody combines (1) a live, browsable record of every graduate with (2) a fair grade against a fixed reference (the graduation cap), (3) protection from painted and paper caps, (4) speed-bonded vs organic splits, and (5) base rates that come from the same graded data. Gradbook is free and has no trading fee, so it complements the terminals: it's the tab you open before you click buy in one.

The product

Views (one page, hash routes):

Today (#/): live class header (graduates so far, still standing), KPIs (share bonded ≤2 min, honor roll count, median held peak), a status distribution bar, standing rates at +1h/+6h/+24h, superlatives, and the yearbook grid of portraits with status stamps. It has filters (Graded, Everyone, Standing, Slipping, Dropped out, Bonded ≤2 min, Honor roll, Paper caps, Off-curve) and sorts (newest, best now, held peak, volume). It refreshes every 60 s, highlights new enrollments, and shows a "since your last visit" diff.
Yearbook page (#/coin/<mint>): portrait, status stamp, yearbook quote, and a report card (graduated cap, +1h, +6h, +24h, held peak, now, each as a multiple). It also shows a market-cap chart since launch (log scale, 1-min candles for young coins) with the graduation line, the graduation-cap line and the held-peak line. Below are the record (time to bond, raw high, liquidity, volume, last trade, deployer), odds for a graduate like this, and the deployer's report card (coins launched, how many graduated, recent ones linked). Actions: save to locker, share card PNG, copy link. Works for any pump.fun graduate, even one not yet enrolled; it is graded client-side with the same engine.
Classes (#/classes, #/class/<day>): one row per UTC day with graduates, % bonded ≤2 min, standing at +6h and +24h, honor roll and valedictorian. Each row opens that day's full yearbook.
Odds desk (#/odds): base-rate table by checkpoint (+1h or +6h) and multiple bucket, with outcome at +6h or +24h, split all / organic / speed-bonded, with n per cell. Deep-linkable with the user's coin highlighted.
Locker (#/locker): saved graduates in localStorage, with "was Standing, now Dropped out" and % change since the user last looked.

How it works:

Enrollment: a worker job every 60 s reads pump.fun's completed-coin listing (newest two pages plus one older page from a cursor that sweeps the whole ~22h listing every ~20 min). DexScreener supplies the PumpSwap pair creation time (= graduation time), pool, liquidity and live cap.
Graduation cap: on the standard curve the frozen virtual reserves are 115.005 SOL / 279.9M tokens, so every standard graduate completes at 410.9 SOL of market cap. We price that at the current SOL price. Coins whose reserves don't match (micro curves, non-SOL pairs) are marked Off-curve and not graded, because a wrong number is worse than none.
Grading: pump.fun 5-minute candles give the cap at +1h/+6h/+24h (last close at or before each time) and the held peak: the highest cap the coin held across three consecutive 5-minute closes, with at least 2% of that cap traded in the window. Re-grades run just after each checkpoint. Candle calls are paced across the hour within the worker's px budget (920/h self-cap), and the backlog is split newest-first and oldest-first.
Paper caps: the current cap is "paper" if it is 25×+ the pool's liquidity (above $250K), or 10×+ anything the coin held with volume. Paper caps never count as standing and are excluded from superlatives and the honor roll.
Live caps: DexScreener (30 mints per call) every 5 minutes for live coins and hourly for dead ones. Because DexScreener lags on new pairs, its value is only trusted once the pair is ≥45 min old and the read is newer than our last candle grade. Otherwise the latest pump.fun trade wins.
Backend: SQLite table grads (one JSON record per graduate, indexed by UTC day, graduation time and due time). Jobs: enroll (60 s), grade (60 s), live (300 s). Routes: /class?day=, /classes, /coin?mint=, /odds?at=&to=, /status. Rows are kept 30 days. No personal data is stored.

Calibration (real inputs, 2026-10-09):

Graduation cap: standard curves completed at 410.9 SOL in every sampled coin (virtual reserves 115,005,3xx,xxx lamports / 279.9M tokens). At SOL $110 that's $45.1K. Candle paths of organic graduates cross ~$40–45K at the pool creation time (BER: $41–42K within 20 minutes of its 07:10 pool).
Painted spikes: SARP (5sAY…pump) was created and bonded in the same second. Its first 5-min candle closed at a $35.0M cap, the next at $35.5M, then $2K. With the 15-min/2%-volume rule its held peak drops out, and it's flagged "Most creative accounting". FOMO (CPkp…pump) went $441K → $534K → $652K → $2K in 20 minutes; we show the level it actually held with volume, not the $700K wick.
Paper caps: D O T showed a $74M cap on a $761K pool (98.7×), GOIF $36M on $525K (68×), QI $14M on $325K (42×) and RHEA $5.6M on $207K (27×). Organic runners sat at 12–18× (SpaceXSI 17.5×, NASA 17.8×, FOMO G1Hk… 15.2×), so the threshold is 25×. A first version used 0.5% volume for held peaks and let a $71M "fund" clone become valedictorian. At 2%, those clones can't qualify, because their whole day of volume ($0.3–0.7M) is under 2% of the claimed peak.
Stale caps: Fomoween (kw9e…pump) showed $53K on DexScreener 10 minutes after graduating, while pump.fun 1-minute candles showed it at $57K at minute 2, $11K at minute 3 and $3K at minute 8. Hence the freshness rule above.
Hand-checked rows (engine output against raw pump.fun candles, 15:05Z):
- FI (4At7…pump): created 01:28:24, pool at 01:31:29 (bond 184 s). The last candle before +1h (02:30) closed at a $91K cap and the one before +6h (07:30) at $120K; the engine says $91K and $120K. The top 5-min closes were $219K (03:00) and $208K (04:25). The engine's held peak is $182K (04:20–04:30, with ≥2% traded), so status is Standing at about $93K now.
- the day after (137N…kXsD): bonded in 26 min. Candles closed at $47K/$46K at graduation and $9K at +1h; the engine gives +1h $9K (0.21×), so status is Dropped out.
- BNKR (5btm…pump, bonded at birth): candle closes were $54K, $92K, then $3K. The engine gives a held peak of $3K and a raw high of $92K, so status is Dropped out.
- All three match the candles exactly.
Early base rate (103 graduates with both +1h and +6h, first hour of backend data): of 88 graduates under 0.4× at +1h, 2% were back at or above their cap at +6h. All 63 speed-bonded ones in that bucket stayed under, against 8% of the 25 organic ones. The table gets denser every hour.

Limits:

pump.fun's listing reaches back ~22h, so the first class is partial and +24h odds fill in after a day of running.
A coin that bonds more than ~22h after creation can be missed by the sweep.
Off-curve graduates are listed but not graded.
SOL price for the graduation cap is the current price, not the price at graduation time (a few % drift over 24h).
Held peak is a conservative "held" measure, not the ATH.
Candles cover the newest ~83h of trading per coin.

Token design

$GRADS is planned, not launched. It would launch on pump.fun by a human operator, and the mint is set only then.

Creator fees fund the data bill, said plainly: candle and market-data calls, the backend, and a longer archive (90-day classes, a full-history backfill through an indexer).
Holder tier (read-only balance check, never a signature): alerts when a locker coin changes status, the full odds desk (more buckets, by bond speed and deployer history), CSV export of a class.
Gradbook grades its own token: $GRADS appears in its own class like any other graduate, with no special treatment. If it drops out, the yearbook says so.
No staking, no yield, no "airdrop to holders", no promised returns.

Launch plan

Daily "Class of " post from the share card at 00:05 UTC: graduates, standing %, bonded at birth %, valedictorian and most creative accounting. The format is the hook.
Reply to "just migrated" calls with the coin's yearbook link and its odds row.
Pitch the class report to memecoin data accounts and newsletters, since it's original data (speed-bonded vs organic survival is not published anywhere we found).
KOL and creator angle: "my coin made the honor roll" cards.

Roadmap

Next 2 weeks: per-deployer yearbook (every graduate a wallet has produced, with its survival rate); locker alerts; odds by bond speed in finer buckets (same block / <60 s / <2 min / organic).
Next month: 90-day archive, weekly "reunion" report (where last week's class is now), and Custom Pairs graduates (USDC/stock-quoted) with their own reference caps.
Later: graduates from other launchpads (Raydium LaunchLab, Meteora DBC) with per-curve reference caps.

Risks and kill criteria

API changes: pump.fun's listing or candle endpoint changes shape or blocks us. Mitigation: grade from DexScreener snapshots where possible. Kill/pivot if candles disappear for more than a week.
Thin data early: odds with small n can mislead. Every cell shows n, and the copy says base rates are not predictions.
Gamed outcomes: wash volume can still fake a held peak at 2%. We publish the rules and keep tightening them with calibration.
Demand: if fewer than ~300 weekly users return to the class view after launch week, or the daily card gets no engagement, fold it into a broader trenches tool.

Sources

All accessed 2026-10-09.