Is the Bottom Forming? The On-Chain Floor vs. the Macro Ceiling

Conviction holders absorbed a record volume of coins into the June low while US spot ETFs bled a record $4 billion-plus. The transfer from weak hands to strong hands is real — we verify it independently below. But ownership tells you who holds the coins, not when the price turns: this cycle has already produced two accumulation signals that failed. What follows is the framework, the backtest, and the three signposts that would graduate “floor forming” to “trend turning.”

TON618 Capital Research · BTC $64,063 · −48.7% from the Oct 2025 high · as of 7 Jul 2026 · Download PDF

1Executive Summary

The weak-hands-to-strong-hands transfer that marks every late-bear phase is happening, and it is verifiable — but on the evidence, it times the floor, not the turn. The transition from “floor forming” to “trend turning” would be observable in three public signposts — ETF net flows, the 50-day moving average, and the US spot premium — and as of this writing none of the three has confirmed. The single most important honest fact in this note: the two loudest accumulation signals of this cycle — the December 2025 long-term-holder flip at $88,474 and the record April 2026 accumulation spike at $78,539 — were each followed by a further ~25% decline. Conviction buyers set floors; they do not ring bells.

2Investment Context

• Cycle position: BTC peaked at $124,824 on 6 Oct 2025 and printed $58,525 on 30 Jun 2026 — nine months and −53% peak-to-trough, now $64,063 (+9.5% off that low). Prior bears bottomed 12–14 months from the peak at −77% to −86% (2015, 2018, 2022); this one is so far shallower, consistent with a maturing, ETF-intermediated asset — and nine months in, still inside the historical bottoming window.
• The regime is new: this is Bitcoin’s first bear market in which US spot ETFs are the marginal seller. There is no historical base rate for how an ETF-era bottom forms — a limitation we carry explicitly through the backtest in §5.
• The macro ceiling: a hawkish Fed under Chair Warsh with 2026 cuts largely priced out, a firm dollar, and a persistent fund-share supply overhang are what put BTC here. On-chain floors historically hold once the macro stops worsening, not before.
• Why the floor/turn distinction matters: in every prior cycle, the cost of waiting for confirmation of a turn was a modestly higher entry point; the cost of mistaking a floor for a turn was months of further drawdown, absorbed at the moment of maximum uncertainty. That asymmetry is why this note insists on treating “who owns the coins” and “when the price turns” as separate questions.

3The Claim, Verified — What We Could and Could Not Confirm

The reported claim. Bitfinex desk research, widely circulated in early July, reported that large-wallet cohorts accumulated ≈270,000 BTC (≈$16.7B) in the two weeks into early July, concentrated near $59,000 — described as the largest single on-chain accumulation spike on record, exceeding the COVID and FTX lows. Glassnode separately reported the 155-day (“long-term holder”) cohort flipping from net distribution to net accumulation, with ≈54% of supply held at an unrealized loss.

What we verified independently (BGeometrics long-term-holder series; Signal Ledger daily harvest; local CoinMetrics price history):

• The LTH cohort is accumulating, and has been for five months. Long-term-holder supply stands at 16,691,194 BTC, up 337,390 in 30 days (84th percentile of the four-year sample). The 30-day change has been positive continuously since 29 Jan 2026 — conviction cohorts absorbed the entire decline from $80K to $58K, the direct analog of 2022’s whale behavior at $30K/$25K/$20K.
• The market trades below its aggregate cost basis. NUPL -0.18 and MVRV Z-Score -0.37 (7 Jul harvest). Negative NUPL has occurred only in the deep-capitulation zones of prior cycles: 2015, late 2018, March 2020, and June–November 2022. This independently corroborates the direction of the “54% of supply underwater” claim.
• The June ETF bleed was a record, and it is turning. Net June outflow was -75,003 BTC — the worst calendar month since launch and, in BTC terms, roughly twice the previous record (Feb 2025, −37,243). In dollars that is ≈−$4.7B at June’s average price on our vendor’s flow series; press tallies of the same month report ≈$4.1B — a vendor-methodology gap we disclose rather than adjudicate (the BTC-unit figure is the one we verify). The turn: the 5-day flow z-score has recovered from −2.7 at the June trough to −0.8, and two of July’s first three sessions printed net inflows (+3,638 and +3,315 BTC).

What we could not verify in-house, and carry as cited third-party color only: the specific 270,000 BTC whale-wallet figure (whale-address cohort data is not served by our entitled sources — CoinMetrics gates the address-balance bands, BGeometrics does not carry them), the “largest on record” superlative, and the negative US spot premium. Accumulation-address metrics also carry heuristic noise — custody migration and OTC reshuffling can masquerade as conviction demand — so the precise figure should not be treated as gospel. Our verifiable series (the 155-day cohort) tells the same directional story and is the series we backtest.

4The Divergence — Who Is Selling to Whom

June 2026 produced the sharpest ownership-transfer divergence of the ETF era: the US fund complex disgorged a record -75,003 BTC while the long-term-holder cohort added 337,390 BTC over the trailing month. The mechanical reading: US fund-share sellers (retail and advisors redeeming IBIT et al.) supplied the coins; price-insensitive conviction cohorts — predominantly offshore and large-holder, per the negative US spot premium — took them. In 2024–25 the ETF complex was the marginal buyer that drove the advance; its sellers are now being absorbed at −50% prices without breaking the market structure. That absorption is what a floor looks like while it forms.

Two structural cautions. First, this is the first ETF-era bear: there is no historical precedent for how long fund-share liquidation can persist — and June’s record outflow alone was ≈12% of the complex’s accumulated holdings. Second, part of the measured LTH build is mechanical rather than fresh demand: coins aging past the 155-day threshold today were bought in early February at ≈$78,700 — underwater top-half buyers graduating into the “conviction” cohort by the calendar, not by choice. The record April 2026 build (+1.24M BTC/30d, the largest in our sample) was substantially October–November top buyers aging in. This artifact inflates accumulation magnitudes mid-bear — one more reason to trade the flip, not the headline size.

5The Backtest — What Accumulation Signals Are Actually Worth

We tested two signal definitions on the verifiable series, sample July 2022–July 2026 (the vendor window; effectively one full cycle):

• FLIP: the 30-day change in LTH supply crosses from negative to positive while price is >25% below its all-time high — the Glassnode-style “returned to accumulation” event. Six episodes.
• SPIKE: the 14-day LTH build exceeds +1.5 standard deviations in a >25% drawdown — the “record accumulation” event of the current headlines. Three episodes.

What the test actually does, in plain English. Every time one of these signals fired over the past four years, we froze the calendar and asked three simple questions: What did the price do over the next one, two, three, and six months? Before it recovered, how much more did it fall (the “max further decline” column — the pain a buyer on signal day had to sit through)? And how long until the local low was in? Then we asked the question most backtests skip: compared to what? It is not fair to compare these episodes against all of history, because the signal only fires when the market is already deeply beaten down — and being deeply beaten down is, by itself, bullish on long horizons. So the benchmark in Table 3 is every day in the sample that was similarly down more than 25%, signal or no signal. The signal only earns its keep if it beats that.

Read Table 1 row by row and the texture is as informative as the averages. The December 2022 flip — a month after the FTX collapse, at −75% — was nearly perfect: the local low was in within two days and the next month returned +26%. The 2023 flips were good-to-mixed. And then the bottom row: December 2025, this cycle, where the identical signal fired at $88,474 and a buyer took −28% of further pain over the next 45 days. Same signal, same definition, opposite outcome — the difference was not the on-chain behavior, it was the macro backdrop each fired into.

The spike table is the sharper cautionary tale, because it tests the exact shape of the current headline — “record accumulation” — rather than the quieter flip. Of the three record-pace buying episodes, only one (September 2024) marked the low to the day. The other two, including the largest accumulation reading in the entire sample this April, were followed by another quarter of a bear market. If “biggest buying spree ever” were a bottom signal, April would have been the bottom. It wasn’t.

Three findings, stated plainly:

1. The flip has real short-horizon information. Mean +12.2% over the following 30 days (5 of 6 positive) versus +4.0% for merely being in a deep drawdown. If the flip means anything, it means the next month is usually good.
2. At six months the signal is a net negative versus the base rate. Flip episodes averaged +26.5% at 180 days — below the +43.6% drawdown-matched base rate. Long-horizon returns in this sample came from being down 25%+ at all, not from catching the accumulation event. The signal is a timing refinement, not a regime call.
3. It is not sufficient — and this cycle has already proven it twice. The December 2025 flip (at $88,474) preceded a further −28% slide over 45 days; the April 2026 spike — the largest accumulation reading in the entire sample — preceded another −25%. Mean further decline after a signal was −11 to −12%; the worst case −28%. Whales accumulate toward bottoms, all the way down.

The common sense underneath the numbers. Why should accumulation behave this way — genuinely informative, yet chronically early? Because accumulation is a supply fact and a bottom is a demand event, and they are answered by different actors. A conviction buyer is a value buyer: he bids where price meets his estimate of worth, absorbs whatever panic supply shows up, and — crucially — does not chase. His buying removes coins from the market and thins out the sellers, which is why the month after a flip is usually good: the marginal panic-seller has run out of coins to hit him with. But that same buyer will not pay up 10% to start a trend; he is the floor, never the rocket. For price to actually turn, someone price-insensitive in the other direction has to show up — a flow buyer, which in this cycle’s structure means the ETF complex — and flow buyers respond to macro conditions, not to on-chain value. That is the whole architecture of this note in one sentence: the cohort we can verify builds the floor; the cohort tracked by the signposts in Table 5 has to build the turn. It is also why the signal decays by six months — once the drawdown itself is deep enough, long-horizon returns are dominated by the eventual macro turn, which arrives on its own schedule and owes the accumulation date nothing.

Limitations, disclosed rather than buried: six flip and three spike episodes is a small sample; the window covers essentially one cycle whose recovery (2023–24) unfolded under an easing Fed — the opposite of today’s macro; the series is a single vendor’s methodology (BGeometrics realized-cap computations); and the 180-day base rate is flattered by that same recovery. We treat the backtest as evidence about signal ordering (floor before turn), not as an expected-return forecast.

6Scenarios

Probability-weighted expected return: ≈+5% over six months — positive but unremarkable, with a fat left tail. That is precisely the profile of a floor that has not yet confirmed: the distribution of outcomes has improved markedly from the June extreme, but it still carries a live left tail — better odds, not yet settled ones. (We note the drawdown-matched 180-day base rate of +43.6% argues for a friendlier skew; we discount it because the sample’s recovery was Fed-assisted in a way today’s macro is not.)

7What Would Confirm a Turn — Three Signposts

The observable state of the market, in one word, is HODL — not a directive, but a description, in the cohort’s own vernacular, of what the verifiable holders are doing: not selling, absorbing supply, and waiting. Whether the floor they are building becomes a turn is a question three public, checkable signposts will answer:

How to read them together: any one signpost alone has flickered before in this bear. Two together would mean something structural — that the price-insensitive seller of the past nine months (fund-share redemption) has been replaced by a price-insensitive buyer, which is the demand-side event the backtest says accumulation alone cannot supply. All three would mark a change in market structure.

What would falsify the thesis: a daily close below the 30 June low of $58,525. The floor claim is that conviction demand absorbs panic supply at that shelf; a close below it means the absorption failed, and the historical worst case from a failed signal (−25–28%) then projects the high-$40Ks. That level is the thesis’s tripwire; it is not a prediction.

8Risk Factors

• LTH cohorts can capitulate. The 2022 analog cuts both ways: long-term holders accumulated from $30K down — and a portion finally capitulated at the November 2022 low. A macro shock that forces conviction sellers invalidates the floor from the inside.
• No ETF-era precedent. Fund-share liquidation is a new structural seller with no historical base rate. June’s record outflow was ≈12% of the complex’s accumulated holdings in a single month; sustained redemption at that pace would overwhelm cohort absorption.
• Macro dominance. Every failed signal in our sample failed against a deteriorating macro backdrop. A hawkish-surprise Warsh Fed, a dollar squeeze, or a liquidity event out-ranks any on-chain signal on our own track-record weighting.
• Measurement risk. The headline whale figures are one vendor’s address-clustering heuristics (unverifiable in-house); our own LTH series embeds a mechanical aging artifact that inflates mid-bear accumulation; single-vendor methodology risk applies to both.
• Small-sample inference. Nine episodes across one cycle. The confidence appropriate to Tables 1–2 is directional, not statistical.