---
name: flow-model
description: Ben's flow-prediction model for new tokens, formalised 2 Sep 2026 — the "will money keep piling in" checklist: watched metric, metric ownership, attention velocity, thin liquidity, reflexivity grades, and analog-benchmark pricing (the "Meteora of X" mechanism). Read on every screen alongside [[price-driver-model]]; this file is the flow layer, that file is the structure layer. Convexity slot reserved, Ben is investigating it.
sources: [cowork]
aliases: [money-piling-in, attention-flow-model, meteora-of-x, analog-benchmark]
---

# The flow model — will money keep piling in?

- [stated] Ben, 2 Sep 2026: likes the current rating model as a good start; will add convexity later (he is investigating it). His framing: the market is not objective, it is attention-driven; flow is key; thin liquidity moves price on low invested dollars; reflexivity keeps it going; the watched-metric/ownership/attention-velocity set looks like a good model for money coming in.
- [stated] Ben's analog-benchmark thesis: "this is the Meteora of Robinhood Chain" is how people actually think — match a token to a leader on another chain, take that leader's mcap as the target, assume RH grows toward/overtakes Solana.

## The flow equation
Money piles in while: (viewers of the story) x (conversion to buyers) x (ticket) > (holders in profit) x (sell triggers). Each model component sits on one term.

### INFLOW PREDICTORS, causal order
1. **Watched metric accelerating on a SHARED dashboard.** The DefiLlama screenshot is the distribution channel; board RANK is what people see, not the absolute number (o1 passing up v3 mattered as a rank event). A metric that cannot be screenshotted converts nobody.
2. **Token owns the metric** — the buy button attached to the dashboard. UP yes (mechanical routing), O/PROLOGUE/post-hoc DEBT no. Without ownership, attention converts to platform usage, not token buying.
3. **Analog benchmark with a wide gap** — the target that fits in one tweet ("32x to AERO"). Converts attention into conviction and holding.
4. **Attention velocity from a small base**: breadth (author count) over size, right-people canaries (the inversebrah follow), live arguments = early. Ben's tested curve: sub-50k accounts +141% at 3 months, 200k+ accounts -39% at 1 week.
5. **Access in one motion**: token on the chain where the audience's wallets are, on the boards they use. o1's failure: story on RH, token on Base.

### AMPLIFIERS
6. **Thin liquidity**: same dollars, bigger candle — and the candle is itself an attention event (trending/gainers tabs), so thinness is part of the attention loop, not just execution. Measure mcap/liquidity.
7. **Supply silence**: nothing mechanical arriving. Emission ratio ~0, no vesting drip, no escrow release. Measure supply-arrival $/day vs revenue $/day.

### REFLEXIVITY GRADES — the "keeps going up" question
- **A-loop**: third-party usage feeds the metric (PONS: strangers' launches). Survives a price dip; the metric keeps printing.
- **B-loop**: price partially feeds the metric (UP: ~38% of fees from its own UP/CASHCAT pools). Dip dampens the metric.
- **C-loop**: the metric IS its own price action (DEBT: toll on own volume; SLVR partly). Reverses instantly; the trapdoor.
Grade determines what a pullback does. CONVEXITY SLOT: C-loops are short convexity, A-loops long — leave for Ben's convexity work.

### OUTFLOW TRIGGERS
First down-print on the watched dashboard (trapdoor evidence, all same week: PDEX -98%/day, VAULT -45%, HOOD10 -23%, CASHCAT/INDEX/STONKBROKER lagging everything) · rank loss on the shared board · analog gap closed under ~3x (the tweet stops working) · 50k+ account saturation · float cost basis far below spot.

## The analog benchmark, first principles
Why it predicts FLOWS even though it is analytically naive: it compresses due diligence to one pattern-match, borrows the analog's verified success, and emits a numeric target that fits in a tweet; the gap itself is the fuel. It does NOT predict outcomes: the analog's own multiple compresses as the chain grows (AERO's ratio-to-TVL compressed 4.5x while Base TVL grew 7.6x; BRETT/DEGEN lost 85% in dollars over the same window). **You are trading the crowd's belief in the mapping, not the mapping's truth — time-box the trade.**

### The live analog shelf, 2 Sep 2026 (mcaps)
PUMP $1,698M · JUP $703M · AERO $462M · PENDLE $318M · JTO $221M · RAY $220M · KMNO $128M · MET $105M · ORCA $75M.
Current RH mappings: PONS->PUMP (5.6x gap) · UP->AERO (32x on circ, the mapping lbexplorer already posted = CLAIMED) · DELTA->MET (6.1x) · INDEX->JTO (8.8x) · perp slot->HYPE (unfilled; the HYPE lesson: analog must match on the metric that pays it, RH perp fees were 98x off at equal TVL).
Decision rule: gap >=10x = fuel · 3-10x = mid · <3x = spent. Then check X for whether the mapping is already circulating (claimed) or unclaimed; unclaimed + inflow predictors green = earliest entry.

### GAP-SHOPPING — the forward call this model makes
Narrative shops for the widest remaining chain gap. Measured 2 Sep: RH DEX volume $1.67B/day is ALREADY 74% of Solana's $2.25B — the "RH volume catches Solana" pitch is nearly spent. RH TVL $743M is 13% of Solana's $5.69B — **TVL is the gap the narrative moves to next.** Expect the next analog rotation to hit TVL-linked slots: money markets and liquidity managers (KMNO $128M, MET $105M analogs; the DELTA/Snuggle slot), IF chain TVL keeps growing. Fee-linked analogs (PONS/UP) are the current rotation.

## BACKTEST RESULTS, 2 Sep 2026 — Ben asked to test everything against the top-100 list. Universe: 99 RH tokens (76 ecosystem + 23 stock-token controls), 35 days of daily prices per token, 13 tokens with DefiLlama daily revenue (207 usable token-days), X CA-mention rates for top 32.

### WHAT SURVIVED
1. **Chain beta is nearly the whole game.** Ecosystem median +80.9% in 7 days (IQR +46 to +223) vs stock-token controls -1.5% (IQR -5 to +3). The controls prove the spread is chain flow, not USD noise.
2. **Metric acceleration buys TAIL, not median — DOWNGRADED on the 2 Sep verification rerun.** P(>+100% in 3d): ACCEL 17% vs DECEL 7%, but two-proportion z=1.89, **p=0.059, not significant**, and all 11 ACCEL tail events come from just 4 tokens (LUTE, PONS, PROLOGUE, UP) with autocorrelated consecutive days, so the effective n is a handful of episodes. Leave-one-token-out keeps the gap positive (+6 to +15pp) — directionally consistent, statistically underpowered. Treat as a tilt, not a fact.
3. Within-category winner-take-all (leaders extended in all four category pairs).

### WHAT FAILED
4. **Metric acceleration does NOT predict forward returns at the median.** Pooled Spearman accel vs fwd3 +0.09, vs fwd7 +0.03 (n=207/191); within-day cross-section mean rho -0.01, positive 15/32 days. **DECEL fwd7 median was +54.1%** — decelerating-revenue tokens still rallied hard in the up-tape. CORRECTION to my own "deceleration-kill 6-for-6": that was same-week co-movement, not a leading signal.
5. **The metric is coincident-to-LAGGING, not leading.** Lead-lag on dlog(rev) vs dlog(price): biggest correlations at lag 0 (UP +0.76, SLVR +0.77, CASHCAT +0.59) and lag -1, i.e. **price leads revenue by ~a day** (PONS -1:+0.38, CASHCAT -1:+0.38, SLVR -1:+0.35). The dashboard confirms and recruits AFTER the move; it does not trigger it. Exit signals must come from price and chain aggregate, not from waiting for a down revenue print.
6. **No within-ecosystem factor was significant this window** (n~21 needs |rho|>=0.43): log mcap -0.36, thinness mc/liq -0.26 (WRONG SIGN for the thin-liquidity amplifier), turnover +0.16, fdv/mc +0.23 (wrong sign for the float thesis), drawdown -0.16, attention rate -0.22 (level has no power, again), max-follower +0.33 (against the sub-50k thesis, insignificant). Within-chain picking this week was mostly noise; beta + tails are the game.

### VERIFICATION RERUN, 2 Sep 2026 (scipy p-values, artifact exclusions, floor sensitivity, clustering checks)
- Beta CONFIRMED, now significant: eco median +76.9% vs stock control -1.1%, Mann-Whitney p=6e-5. Today-selected-universe caveat stands (dead tokens shrink out, median inflated); the contrast survives because both groups share the selection.
- Pooled accel null CONFIRMED, floor-robust: rho +0.095 p=0.175 ($50 floor), +0.072 p=0.324 ($1k floor).
- Cross-section: still nothing significant (best p=0.095). Attention LEVEL moved to -0.37 p=0.095 (borderline negative), maxfol +0.37 p=0.095 — contradictory, both insignificant, noise.
- **NEW, the best result of the whole exercise: within-token heterogeneity.** Accel->fwd3 per token, n=32 days each: **PONS +0.43 p=0.01** (its own revenue acceleration predicted its own forward returns), **CASHCAT -0.38 p=0.03** (its letscash revenue spikes marked local TOPS), UP -0.03, INDEX +0.05, SLVR +0.05, STONKBROKER +0.22 ns. Multiple-testing caveat: 6 tests, 2 hits.
**THE PRINCIPLE: a watched metric predicts a token's future only where that metric IS the token's narrative.** PONS is priced as a launchpad, so its revenue signal works on it. CASHCAT is priced as a meme with a launchpad attached, so the same signal inverts. UP couples same-day (lag-0 0.76) with no lead. Before using metric acceleration on any token, establish what its crowd prices it on.
- Consequences: PONS's valid buy trigger is its NEXT acceleration print (flat now, so PONS = beta hold, not signal buy). UP = attention/momentum trade, its own metric has no lead on itself. Never buy CASHCAT revenue spikes. QUOTRONS pick rests only on "leaders extend", no tested factor.

### STILL UNTESTED
Attention ACCELERATION (needs mention time-series, still open), board-rank-change events, the analog-gap rule. One week / one chain / up-tape caveat applies to all of the above; re-run in a flat tape before trusting any of it.

### DATA TRAPS THIS RUN
Young-pool OHLCV fabricates returns (BONER +41,614%, MOO +4,350%, CLANKER: all mc-null artifacts, excluded). LUTE price-return vs mcap-change contradict because its supply changed; never mix the two. X CA-queries undercount handle-first projects (UP 1.9/hr by CA vs ~16/hr by handle).
New names surfaced by attention rate, UNSCREENED: BONER (5.4/hr), UBIK (5.0/hr, 49k maxfol), RAM ($15.7M, 170k maxfol), MANCER, SHRUB, RIPE.

### PICKS DELIVERED 2 Sep off this test
Core PONS (own the validated factor, beta, in the cleanest liquid structure) + tail slot UP (the one large accelerating watched metric, 2.4x tail odds) + optional QUOTRONS (payout-tier winner). Avoid: decelerating payout tier, SLVR (emission ratio still >1), LUTE (data integrity + team multisigs), AI only as a momentum rental with no falsifier. Kill switch for the whole basket: two consecutive down days in chain DEX volume (api.llama.fi/overview/dexs/robinhood change_1d), because everything in it is one beta.

## APPLICATION: DELTA / TWO / PAIR / BOW / MAST ranked with the validated framework, 2 Sep 2026
Shared handicap: NONE of the five has a DefiLlama adapter, so none owns a watched metric on the shared board — all are outside the screenshot economy where attention converts. All five therefore rank below PONS/UP.
1. **TWO $5.05M** (0x2A4a33A2163D005d8E7f1D9aC08d14c98db288d5): only accelerator (mcap $2.4M -> $5.05M in ~3 days), only one routing anything to holders (bounded yield-skim to TWO staking), 100% float, 139k-follower account touched it. Risks: adoption was $13k TVL at the 30 Aug screen (unverified since), -31% off ATH, new main pool (young-bar artifacts). Momentum + mechanism lottery.
2. **DELTA $17.8M**: best product of the five (~$1.5M managed, real users) and the named local claimant for the coming TVL-gap analog rotation (MET $105M = 6x gap). But docs mention the token ZERO times, Snuggle out-earns it ~10x with no token (Snuggle 30d rev $228,557 and growing), and it is rolling over (-35% r3, -29% on the day). An option on the narrative rotating to TVL categories, not an economics buy.
3. **BOW $6.95M**: strongest price structure (-5% off ATH, only green of the five, +8.7%), 2.98 mentions/hr. But lending on RH earns $73/30d chain-wide, the token has no stated economics, and the longbow four-site sprawl stands. Attention shell in a dead category; momentum only.
4. **PAIR $4.12M**: product works, contracts clean, but the launchpad still has no users beyond itself (was 91% self-volume), the burn is circular PAIR-denominated fee inventory, no contractual claim on fees, -56% off ATH. Upgrade trigger from the 31 Aug screen has not fired.
5. **MAST $0.77M**: falsifier fired, see [[mast-bond]]. Zero.

## VERIFICATION PASS 3 on the ranked buy list, 2 Sep 2026 04:30 UTC. Four corrections.
1. **PONS signal survives but is 3-day only.** Beta-adjusted (same-day median of other revenue tokens subtracted): rho +0.40 p=0.025. Split-half consistent (+0.30/+0.45). **fwd7: rho +0.02 p=0.93 — the predictive power is entirely a 3-day effect.** Trade the trigger on a 3-day horizon, never as a hold signal.
2. **CASHCAT inverse signal STRENGTHENS beta-adjusted: rho -0.45 p=0.009.** The never-buy-its-revenue-spikes rule upgraded from suggestive to solid.
3. **HOOKR's claimed burn FAILS on-chain verification: total supply is exactly 1,000,000,000 with no dead-address accumulation in the top holders.** The "capped once-per-block buyback burn" from the 30 Aug sector map is not visible on chain. DEMOTED from buy list to watchlist pending mechanism verification. Same for **DTF: supply 1B intact, no burn/buyback destination found in top holders** — "app fees buy back DTF" unproven, DEMOTED.
4. **QUOTRONS mechanism CONFIRMED by supply: genesis 4,444, current total supply 1,667 = 62.5% of supply permanently retired by hardwires.** The strongest verified hard-sink on the chain. PROMOTED to the #3 slot.
Also: **UP's metric has stopped accelerating — rolling 24h $144,342 vs 1 Sep's $146,986, flat at the ATH**, the same shape PONS printed four days before its plateau. UP stays #2 on structure and attention but the "only accelerating metric" line is retired; next down-print is the exit tell (remembering price will move first). UP supply still exactly 503,027,596.80, holders 8,632. PONS trigger not armed: rev $868k/$864k/$838k, drifting down. Kill switch green: chain DEX +12.0% 1d.
FINAL ORDER after pass 3: 1 PONS (3-day trigger discipline) · 2 UP (flat-at-ATH caveat) · 3 QUOTRONS (verified sink, thin exit $514k) · 4 TWO (routing real, adoption unverified, +39% 1h) · watchlist HOOKR, DTF (mechanisms unproven on chain), DELTA (rotation option), AI (no falsifier) · avoid list unchanged and CASHCAT rule hardened.