Buyer Addresses Rebounded 174% While Sales Fell 15%, and the Base Gap Closed to 5.3%

The week ending Sep 19, 2026 recorded $37.54M in organic NFT sales — a second consecutive decline — while buyer addresses rose 174.04% to 114,977, one week after collapsing 84.67%. The two comparison bases that sat 42% apart a week ago closed to 5.3%.

This report decomposes last week's base gap to a single chain, scores last week's calls against the new capture, and sets out what the address series can and cannot settle about whether the buyer metric broke or the market did.

1. Background: Both Columns Reversed at Once

The week ending Sep 19, 2026 reversed both columns of this series simultaneously. CryptoSlam's seven-day dashboard recorded $37.54M in organic sales — down 15.28% against the ~$44.31M prior week cited inside that capture, and down 19.75% against the $46.78M in our own Sep 12 capture. It is the second consecutive weekly decline. In the same window, buyer addresses rose 174.04% to 114,977 and seller addresses 151.72% to 108,037, one week after collapsing 84.67%. Transactions fell 9.08% to 808,432. Fewer dollars, more participants — the exact inversion of last week, when dollars rose and participants vanished.

The comparison bases closed as well. Last week the prior-week figure inside the capture sat 42% below the level our own capture had recorded seven days earlier, and I wrote that no percentage that week was quotable without its base attached. This week the two bases are 5.3% apart, and the address columns reconcile to within three addresses: the capture's own +174.04% implies a prior buyer count of 41,956, against the 41,959 we recorded. Value and transaction counts still do not reconcile, but the market's divergences are narrowing, not widening.

My position up front, because last week's report made two claims this capture can test. I held the cause of the participation collapse loosely and its levels firmly, and I called the market excluding BNB Chain — flat at $42.76M for two weeks — "the single most checkable claim in the series". The test has now been run and the floor broke: ex-BNB printed $34.96M, 18.2% below our base and 13.2% below the capture's own. What did not break is the address series, which is now internally consistent across three consecutive captures. And the 42% gap I called unreconcilable has a single identifiable cause: 93% of it was one chain's one-off week.

2. Event Timeline

1. Sep 12 — our previous capture records $46.78M in organic sales, 917,549 transactions, 41,959 buyer addresses and 43,247 sellers, with BNB Chain reverted to $4.02M after its 1,042% week and Ethereum back in the chain lead.
2. Sep 12–18 — ETH rises 1.73%, from $2,514.85 to $2,558.32 — a mild week for the unit of account, against the 6.94% rally of the previous window.
3. Sep 18 — NeoDrop's briefing, data cutoff 09:00 ET, records four of five blue-chip floors lower in ETH terms and BAYC the sole gainer at +0.91% — the second consecutive week in which a single blue-chip floor rose against a falling cohort, after Doodles was the sole gainer the prior week.
4. Sep 19 — the capture records $37.54M in organic sales on 808,432 transactions, with 114,977 buyer and 108,037 seller addresses. Its cited prior week is ~$44.31M, only 5.3% below our Sep 12 capture — the closest the two frames have come.
5. Sep 19 — Ethereum leads at $15.32M (40.8% of organic volume); Base is the only chain of the six to grow; BNB Chain falls a second consecutive week to $2.58M, about 92% below the level its own 1,042% week reached.
6. Sep 19 — Alchemix V3 Transmuter ranks third among all collections at $1.83M across eight transactions involving four buyers and three sellers — an average of $228,750 per trade.

3. Data Support

3.1 The 42% base gap was one chain

Last week's report treated the two comparison frames as unreconcilable and named that unreconcilability a finding. Decomposing them settles the question: the gap was almost entirely BNB Chain's $32.75M spike, which our Sep 5 capture recorded and the Sep 12 capture's own implied prior week did not.

BNB Chain alone accounts for 93.1% of the $31.74M total discrepancy. Excluding it, the two frames agreed within 5.1% from the start — the 42% headline gap was a single chain's small-base event, not a structural difference in how the two dashboards count. That changes the audit conclusion from "unreconcilable" to "reconciled, with one identified exception".

This week the same decomposition comes out flat, which is why the bases converged:

Both frames now place BNB Chain within 1% of each other and the residual difference sits in the ex-BNB market at about 5–6%, symmetric across the segments and consistent with a modest window offset rather than a coverage difference. The source still publishes no span behind its comparison, so this remains inference — but it is inference that now survives a test it failed last week.

3.2 The address series across three captures

The cleanest way to read the participation column is to hold the frame fixed: the three figures below all come from our own captures, so no comparison-base problem applies.

Two features stand out. First, transactions varied within a 41% band across the three weeks — 650,332 to 917,549 — while buyer addresses varied by a factor of 6.5, from 41,959 to 273,655. Almost all of the volatility in this market's published participation numbers lives in the address column, not in the trade count. Second, value per transaction is the only series that behaves like a trend: $116.16, $50.98, $46.44 — a monotone decline of 60% across three weeks. Everything address-based moved violently and non-monotonically.

A note on what 114,977 is not. It is a rebound from 41,959, but it is still 58% below the 273,655 buyers recorded on Sep 5. The participant base has stopped contracting; it has not returned. And the two ratios that spiked last week — 21.87 transactions per buyer address and $1,115 per buyer address — fell back to 7.03 and $327, neither of which is the Sep 5 norm of 2.38 and $276.

3.3 Last week's calls, scored

Four claims from the Sep 14 edition were checkable against this capture. One landed, one failed outright, and two came back mixed.

The failed call deserves plain language. I called the ex-BNB level the most directly testable claim in the series and endorsed it as a working hypothesis; the correct description now is not "supported" but "flat for two weeks, then down 18.2%" — a plateau before a break, which is what a floor looks like right before it stops being one. The mixed call is genuinely mixed: the address rebound reproduces the signature I associated with whale concentration rather than with a metric change, but concentration at the top was not what lifted the count, since the count rose on all six chains.

3.4 The ex-BNB level, on both bases

The break is not an artefact of which base is used.

And against the capture's own implied prior week, where BNB Chain's prior level is back-derived from the reported −36.26%:

Both frames agree that the ex-BNB market fell this week — by 18.2% on our base and 13.2% on the capture's. The 5-point difference between them lives entirely in the two BNB priors, one of which is derived rather than published. What is no longer in question is the direction: the figure I flagged as the series' most stable quantity was the one that moved.

3.5 Blue chips: one gainer, four decliners

Against a 1.73% ETH rise — a far quieter window than last week's 6.94% — four of five tracked floors still fell in ETH terms, and for the second consecutive week exactly one blue-chip floor rose against a falling cohort: BAYC, after Doodles was the sole gainer the prior week.

BAYC added 0.91% as reported and 2.24% between our two dated snapshots, to 6.85 ETH. CryptoPunks was next-most resilient at −0.74% reported (−0.34% on the snapshot basis); Doodles was weakest at −7.02%, with Pudgy Penguins at −6.29% and Azuki at −4.62%. Direction of turnover now diverges inside the cohort: Doodles traded 52.96% more volume into its weakest floor, while CryptoPunks traded 62.15% less into its mildest decline. Rising turnover against a falling floor reads as sellers meeting bids; falling turnover against a firm floor is an absence of trade rather than a sign of strength.

3.6 The wash overlay, and what carried the volume

CryptoSlam published wash figures for the same three chains as last week, so the transparency gap has now persisted two weeks.

Polygon's ratio snapped back to 2.55× from 2.22×, so three of the four weeks tracked sit at or above 2.5× and the structural label holds. Base's figure returned after a one-week gap at 2.22× ($4.80M against $2.16M) — its widest ratio yet, and above organic in every week it has been published. Ethereum's wash share fell to 2.9% ($443,802 on $15.32M) from 4.7%. Bitcoin, BNB Chain and Solana remain blank; unknown is still not zero, and after two consecutive weeks that absence is a coverage question rather than a one-off omission.

The concentration signature is visible again, one rank lower than last week. Alchemix V3 Transmuter took third place among all collections with $1.83M across eight transactions — $228,750 each, from four buyers and three sellers. Above it, Courtyard led a fourth consecutive week at $6.30M across 123,504 transactions (16.8% of global volume, $51 per trade) and Argonauts held $2.74M across 1,204 trades ($2,276 each). A top three containing both the market's broadest tape and an eight-trade collection is the same concentration story as last week's BRC-20 block, in a different chain.

4. Competing Readings

Where I land: the window reading is now the most economical explanation of the base gap, and unlike last week it is a claim that has survived a test. The redefinition reading stays alive only as a one-time step between Sep 5 and Sep 12, which no published data can retest — I continue to hold it open rather than dismiss it, but it no longer explains this week's numbers. The genuine-swing reading is what the data literally says, and I accept it as the description of the series while declining to call it a description of behaviour. And I withdraw the ex-BNB floor as a working hypothesis: the ex-BNB market was flat for two weeks and then fell 18.2%, and the honest label for that is a plateau, not a support level.

5. Implication Projection

Conditionals, not forecasts. Each is checkable against the Sep 26 capture.

1. If the window reading is right: the Sep 26 capture's cited prior week should again land near our Sep 19 capture of $37.54M — within roughly 6%, so about $35M–$40M — and the value-change columns should keep converging. A reversion to a 40%-scale mismatch would falsify it.
2. If the redefinition reading is right: buyer addresses should stay in the same order of magnitude — a third consecutive print between 100,000 and 120,000 would leave the Sep 5 → Sep 12 step as the only candidate discontinuity in the archive, and every address comparison crossing that boundary would need a note from here on.
3. If the ex-BNB decline is a trend rather than a step: the Sep 26 capture prints ex-BNB below $34.96M. A reclaim above $38M would mark this week as the outlier instead, and the three-week plateau would look less like a level than like noise around a declining mean.
4. For blue-chip holders: one gainer is not a rotation back. A second week with any floor rising — or BAYC holding 6.85 ETH through a flat ETH week — would be the first evidence that the cohort has stopped leaking. Note that BAYC's gain came on 48.94% less weekly volume, which is thin confirmation.
5. For wash-trading analysis: if Polygon stays above 2.5× and Base stays above organic, rank both chains on organic volume only. If the three missing chains publish again, last two weeks read as a reporting omission; a third week of absence makes it a coverage change worth documenting separately from the market data.
6. For anyone quoting the headline: "buyer addresses up 174%" is true and misleading on its own — 114,977 is still 58% below the Sep 5 print, and the same week's sales fell 15.28%. The honest sentence carries the level, not just the change.

6. Open Questions

• Why does the address column swing by a factor of 6.5 while the transaction column stays inside a 41% band? Two consistent captures rule out ongoing drift but cannot rule out a single step between Sep 5 and Sep 12 — and that step remains the largest unresolved question in this archive.
• Where exactly did the ex-BNB plateau break? Our base says −18.2%, the capture's implied base −13.2%. The whole difference sits in two BNB priors, one of which is derived from a reported percentage rather than published.
• Is the residual 5–6% ex-BNB gap a window offset or a coverage difference? Three consecutive captures now show each cited prior landing close to the previous capture's current value — the pattern of a ~7-day offset, but the source still publishes no span.
• Who moved $1.83M through eight Alchemix transactions at $228,750 each? Nothing in the published data identifies the counterparties, and eight trades moving more than 1,200 Argonauts trades is a statement about settlement, not participation.
• Did BAYC's +0.91% break the cohort's decline or merely interrupt it? ETH rose 1.73% over the same window and the gain came on 48.94% less volume, so a single ETH-denominated increase is a thin signal to build a rotation call on.
• Will the missing wash figures for Bitcoin, BNB Chain and Solana return? Two weeks of absence turns a reporting omission into a coverage question, and the chains affected are precisely those where wash would change the read.