miner stress and holder profits are telling opposite stories: puell, a compressed hashrate ribbon and mvrv 1.51

I spend a lot of time watching on-chain valuation metrics, and today two families of them disagree in a way I find more interesting than either number alone.

The miner side of the bitcoin market is flashing stress. The holder side is sitting on a 51 percent unrealized gain. Both readings come from the same chain, the same week. Here is what the data actually says, and where I think it stops being trustworthy.

pull the numbers yourself

Both endpoints below returned full JSON for me today with a plain anonymous curl, no API key and no wallet header. The catalog lists 298 live resources if you want to browse the rest: https://agentdatum.com/.well-known/ai-catalog.json

First, the Puell Multiple, which compares the daily value of newly issued bitcoin against its 365 day average:

curl -s https://agentdatum.com/api/v1/d/processed-puell

Real output (trimmed):

{"product":"processed-puell","snapshot":{"puell":0.9727,"regime":"低估","data_date":"2026-09-05","delay_days":1,"daily_issued_btc":450.0,"price":79815},"interpretation":"BTC Puell Multiple = 0.973，低估。...","collected_at":"2026-09-05T22:31:01Z"}

Puell at 0.973 means miners are earning roughly what they averaged over the past year, no more. The endpoint labels this regime "undervalued", which is its own framing, and I will push back on that label later.

Second, the hashrate ribbon, a miner capitulation signal:

curl -s https://agentdatum.com/api/v1/d/processed-diff-ribbon

Real output (trimmed):

{"snapshot":{"current_hashrate_eh":913.44,"ema9_eh":920.43,"ema200_eh":946.85,"compression":-0.0279,"signal":"矿工投降区（算力带压缩...）","data_date":"2026-09-05"},"collected_at":"2026-09-05T22:31:02Z"}

Current hashrate is 913.44 EH/s, below both the 9 day EMA (920.43) and far below the 200 day EMA (946.85). Compression of minus 2.79 percent. In plain terms: some hash is going offline. Historically that happens when marginal miners cannot cover power costs at the current price of 79,815 dollars per BTC. The ribbon calls this a capitulation zone.

now the other side of the trade

While miners squeeze, holders are comfortably in profit. MVRV, the ratio of market value to realized value, printed 1.512 on 2026-09-04, with a realized price of 52,734 dollars against a spot near 79,815. NUPL sits at 0.339, which the scale calls belief-greed. SOPR is 1.004, meaning coins moving on-chain are sold at roughly breakeven, neither capitulation nor euphoria. Fear and greed read 73 (greed) on 2026-09-05 against a 30 day average of 53.

So: miners are stressed enough to unplug machines, and everyone holding coins is up about half. Those are not contradictory facts, but they point different directions. Miner capitulation signals have historically clustered near cycle bottoms. Holder profitability near 1.5 MVRV has historically been a mid-cycle zone, not a bottom and not the classic top above 3.5.

the honest part

Some limits I checked before writing this:

• The Puell regime label "undervalued" is generous. 0.97 is only low relative to its own 365 day average. The deeply cheap zone in past cycles was under 0.5. I would call today's reading neutral, not cheap.
• The ribbon endpoint computes a hashrate ribbon, not the strict difficulty ribbon, because free sources only publish difficulty in adjustment steps. The endpoint says so itself in the note field. Same family of signal, not identical.
• The data dates are not aligned: Puell and the ribbon are stamped 2026-09-05, MVRV, NUPL and SOPR are stamped 2026-09-04. One day of drift in a 80,000 dollar market matters a little.
• Miner capitulation zones can persist for months. The ribbon tells me miners hurt, not when price bottoms. I have watched bottom signals stay "on" while price chopped for a quarter.
• Both things can be true at once: price at 79,815 versus a realized price of 52,734 means recent buyers hold most of the paper gains, while the marginal miner earns year-average revenue. Nothing here is a trade signal. It is a map of who is comfortable and who is not.

If you want to pull any of these series yourself, the full catalog with 298 endpoints and plain-English field notes is here: https://agentdatum.com/.well-known/ai-catalog.json

This article was drafted with AI assistance and all figures were pulled live from the linked public endpoints on 2026-09-06 before publication.