Public specimen · full audit depth

$FET, audited

Published 2026-10-05 · Analysis window 2026-07-07 to 2026-10-05 · Contract 0xaea4…ad85 on Ethereum. Published as a public specimen, not a client engagement. $FET was not selected because the answer was flattering; it was selected because someone mentioned it and the method had not been run on it.

1 · Verdict summary

21/100 · organic

$FET's attention is carried by a real and growing population. Across three weeks it spiked five times and the number of distinct people posting rose at every one, from 421 to 684. Spam fell as a share of posts during the largest spike, to 0.61x the token's own norm, which is the opposite of a purchased wave. No near-identical wording appears across separate accounts at all.

The strongest evidence sits between the spikes rather than at them. Counting only quiet days, attention runs 2.42x where it did two months ago with 1.58x the people. Arrivals are staying.

Two onchain measurements argue the other way and are reported rather than set aside. Of everyone who has ever held $FET, 32.6% still do, and only 45,252 addresses hold a position worth $100 or more. Spike days converted attention to first-time holders at 0.38x the ordinary rate.

MetricValueRead
Attention vs 30-day baseline3.5x on 2026-10-04elevated
Spam share vs the token's own norm0.184 vs 0.302 (0.61x)clean, falling
Repeated wording across accounts0.0%no campaign
Creator concentration (top 3)62.8%concentrated
Largest single account38.1%, 62,801 followersreal account, not a burner
Attention floor, 90 days2.42x, with 1.58x the peoplecompounding
Onchain conversion on spikes0.38x the ordinary ratepeaks did not convert
Holders with $100+45,252 of 165,57427% of the headline count
Holder retention32.6%two thirds have left
Manufactured score21/100organic (under 30)

Scoring update, 2026-10-12. The score's creator-concentration term is now ranked against same-size peers rather than taken as a raw share, matching how every other figure in this report is compared. Concentration falls with size for arithmetic reasons, so a raw share carries part of a token's size into its score. $FET moves from 23 to 21. The verdict is unchanged. The formula is published.

Score computed with the published formula in the open hype-detector project: spam lift 0.0, spam share 3.7, concentration 17.4, sentiment uniformity 0.0.

2 · Five spikes, a bigger crowd each time

A spike on its own says very little. Attention dies fast whatever its source, so what matters is who showed up and whether they stayed.

DateInteractionsPeople
2026-09-211,317,530 (2.8x)421
2026-09-241,044,228 (2.2x)465
2026-09-251,602,991 (3.4x)538
2026-09-261,003,405 (2.1x)591
2026-10-041,675,803 (3.5x)684
Five $FET attention spikes between September 21 and October 4, each larger than 2x its 30-day normal, with the number of people posting rising from 421 to 684 across them

Counting only the quiet days, with every spike excluded: the floor rose from 208,975 interactions and 238 people ninety days ago, to 506,250 and 376 people now. Each wave left more behind it than the last.

3 · Did anyone actually show up?

This is where the audit argues with itself, and the disagreement is reported rather than resolved.

Every address that has ever received $FET was reconstructed from raw Transfer logs across all 15.1 million blocks since the contract deployed in October 2020: 507,163 addresses across 2,185 days. That gives the day each one first ever held the token.

First-time onchain holders per million interactions: 479 on ordinary days against 183 on spike days, a 0.38x conversion rate

Spike days carried 3.6x the daily attention and produced 1.37x the daily new holders, so the extra attention converted at 0.38x the ordinary rate. On this measure the peaks did not bring buyers in proportion to their size.

How much weight this carries. Roughly 99.9% of $FET's volume is on exchanges rather than onchain: $304M reported against $291k in the deepest DEX pool. A buyer on Binance creates no Ethereum transfer and never appears in this measurement. The metric is strong on tokens that trade onchain, which is what it was built against, and weak on an exchange-listed large cap. It is reported with that limitation rather than dropped.

4 · Who carries the conversation

Share of 24h interactionsFollowersAccount
38.1%62,801JavonTM1
12.6%19,900d3centralizer
12.1%12,752cryptoaimantiktok
9.8%10,000globalrashidshorts
9.4%584traderkitsune

Top-3 concentration of 62.8% is genuinely high and is almost the entire score, contributing 18.8 of its 23 points. What stops it reading as manipulation is the composition: established accounts with real followings, on video platforms, not posting the same words as each other. The one low-follower account at 9.4% is worth noting and is not enough to change the picture.

Classified: broad crowd. Not a megaphone, which needs top-3 above 90% with low spam. Not an institutional broadcast, since no exchange or alert feed appears. Not a coordinated botnet, which needs a fresh spam wave above baseline, and spam fell.

5 · The holder base

A headline holder count is a vanity number on its own. 165,000 holders where most hold airdrop dust is a different token from 8,000 where every one is a position, and the count cannot tell them apart. Every holder was enumerated with their balance.

$FET wallets by position size: 1,455 hold $10k or more, 10,203 hold $1k to $10k, 33,594 hold $100 to $1k, and 113,961 hold under $100
Position sizeWalletsShare
$10k or more1,4550.9%
$1k to $10k10,2036.4%
$100 to $1k33,59421.1%
$10 to $10042,17126.5%
$1 to $1031,72219.9%
under $140,06825.2%

45,252 addresses hold $100 or more, 27.3% of the 165,574 headline count. A quarter of all holders hold under a dollar. The figure a project would quote overstates the holder base with a real position by roughly 3.7x. That is not unusual, which is exactly why it is reported this way.

Retention
ever received the token507,165
still hold it165,574
retention32.6%

341,591 addresses have held $FET at some point and no longer hold any. Two thirds of everyone who has ever held it has left.

This is the sharpest counterweight in the report to the social finding. A conversation that is broadening, with a rising attention floor, is happening around a token whose historical holder base has mostly exited. Both are measured and neither cancels the other. The exchange-volume caveat from section 3 applies here too: an address that bought on an exchange and never withdrew appears in neither figure.

6 · Benchmarks

Ranked against a verified population of 130 tokens by market-cap rank, stablecoins and wrapped assets excluded.

Correction, 2026-10-07. These percentiles were originally computed against the whole population, and two of the three signals move with market cap for arithmetic reasons. An account posting about twenty tokens posts about the big ones by definition, so cross-token overlap rises with size. A token with forty people talking will always have its loudest three owning more of the conversation than one with four thousand, so concentration falls with size. A whole-population rank charges a token for its size. The method now ranks within market-cap bands and the figures below are recomputed; the original global ranks were 70th and 65th. The measurement is published.

Correction, 2026-10-08. The population these percentiles rank against was itself checked, and 78 of its 200 topics turned out not to be about their tokens: short tickers collide with ordinary words, so the baseline contained conversation about sand, cake, bats and a Twitch streamer. The population is now 130 verified topics and the figures below are recomputed against it. The cross-token figure changed as well as its rank, because that signal is measured against the population basket and the basket is what was wrong. On a clean basket $FET sits below its peer median rather than above it. The banding from the previous correction still applies.

$FET sits in the middle band at $521M, against 41 peers of comparable size:

Metric$FETPercentile of 41 same-size peersPeer median
Creator concentration60.2%58th51.4%
Repeated wording across accounts0.0%0th0.0%
Cross-token accounts9.8%31st27.0%

Each correction moved the ranks down. Against the whole market the figures read 70th and 65th, against same-size peers 61st and 61st, and against same-size peers on a baseline that is actually about tokens, 58th and 31st.

Concentration still sits above the median for its band, which remains the one figure an informed reader should push back on. It is not accompanied by any of the signals that usually travel with manufactured concentration, and cross-token accounts now read well below the peer median rather than above it.

Supply. 165,574 holders, cross-checked against a second source at 165,145, a 0.3% difference. Top-10 holders control 48.0% of supply across all addresses but 27.7% once contracts are excluded; the two largest are Gnosis Safes at 10.2% and 9.6%, consistent with treasury rather than individual positions.

7 · What this audit cannot determine

Post-level labels measure amplification, not its source. A bot campaign, coordinated advocacy, algorithmic amplification and an unusually tight real community can produce overlapping patterns. Where evidence is consistent with more than one mechanism, this report says so.

Specifically here: an absence of copypasta rules out one technique and does not rule out coordination conducted carefully. Follower counts are not authenticity. The creators endpoint reflects measurable interactions, so a network spreading activity across many small accounts reads as less concentrated than it is.

The onchain conversion figure cannot separate attention that failed to produce buyers from buying that happened somewhere this measurement cannot see. For a token with this much exchange volume the second explanation is at least as likely as the first, and nothing in the data distinguishes them.

This is not a recommendation. Price is already up 54% over the same ninety days, and the research behind this method found organic attention shifts the odds only while price has not moved yet.

8 · Provenance

Social data from LunarCrush. Onchain data from SQD Portal, Blockscout and Ethplorer. Market cross-check against GeckoTerminal. Spike threshold 2x the token's own 30-day median. Near-duplicate detection at Jaccard 0.6 over 3-word shingles, requiring three or more distinct accounts. Every tool used here is described in the open methodology.

A note on sample selection. The other published specimens scored 98/100 and came back manufactured, 22/100 with one published claim that did not hold, 25/100 with most of the token on another chain, and 25/100 on the week's biggest mover, the flattest crowd of 39 peers. All five are shown because a method that only ever finds problems is not measuring anything.

Get this for your token