Six mining algorithms. Three miners. A chain that starves by design.

The chain didn’t decay. It broke in June 2025.

Through May 2025 the chain held 74–97% of its 45-second schedule with every algorithm carrying a 12–18% share. From June 2025 the monthly record turns into a rollercoaster: 48%, 11%, 55%, 3%, 45%, 11%, 83%, 6%. A chain that produces 1,866 blocks per day in March and 66 per day in October cannot carry payments, exchanges, or confidence.

Half the algorithms lost their miners

Each cell below is one algorithm’s share of one month’s blocks, against the 16.7% an even six-way split predicts. Groestl fell to 0.1% in October 2025; argon2d to 6.6% the month before. The merge-mined algorithms (sha256d, scrypt) never flinch — in the worst months they carry 40–45% of everything produced.

And the blocks that do arrive come in bursts: in May 2026 the median gap was 36 seconds in a month that delivered 6% of schedule. A large miner visits, mines a fast run, difficulty ratchets up — observed live at the tip: sha256d difficulty ×16 in one evening — then the visitor leaves, and difficulty can only fall when blocks arrive, which the stranded difficulty itself prevents.

Dead algorithms drag down the live ones

The difficulty algorithm balances the six algorithms against each other: an algorithm that mined more often than one-in-six recently is hardened roughly 12% per extra slot, every block. That arithmetic assumes six populations exist. With three algorithms dead, the three survivors necessarily mine one-in-three — so the design perpetually punishes the only miners it has left, while the empty algorithms sit eased-to-the-floor, unable to help. No limit or retarget tuning can fill an empty slot.

Same miners, four algorithms: problem solved

An exact integer-math port of the deployed difficulty code (per-algo balancing, damping, adjustment limits, median-time windows; event-driven block arrival; three seeds averaged). Throughput is measured against the same one-block-per-45s ideal in every row.

Three readings. The model matches reality: six algorithms with three dead equilibrates at 3.3% — the real chain’s October 2025 produced 3%. Limits alone fix nothing: widening them without changing the set moves 1.0% to 1.3%. The set change is the fix: the same burst-heavy miner population on four live algorithms produces 88.9%; even with an entire algorithm silent, four algorithms beat today’s law thirty-fold.

Everything here is chain-wide (the difficulty algorithm is chain-wide). After a chain-wide rental spike whose hashrate then leaves, widening 12/22 → 20/35 cuts recovery 35.9 → 21.6 h (−40%), roughly doubling the blocks produced while recovering. Wider settings recover faster still (down to 15 h at 30/60) but with diminishing returns, and each step raises what a stall-then-harvest attacker can extract from an algorithm it dominates (6.1× → 10.2×) — so 20/35 is the balance, not the widest. In the current six-algorithm era 12/22 never recovers within two days at all, and replacing the difficulty algorithm outright with LWMA-1 never recovers either (it is built for single-algorithm chains). No limit value fixes a starved algorithm set — compare these rows with section 4's dead-algorithm rows.

Honest caveats: the burst pattern is illustrative (mechanism reproduction, not a hashrate fit); simulated hashrate has no price feedback — real recovery is faster than simulated for every configuration; hashrate is not fungible across algorithms, which is exactly why the retained four were chosen for their existing populations.

The merge-mined algorithms swing. The native algorithms pull them back.

sha256d and scrypt hashrate is enormous, exogenous and elastic: merge-mining pools point Bitcoin- and Litecoin-scale hash at argentum at zero marginal cost — and withdraw it just as freely. No difficulty algorithm can prevent those swings; the chain has to absorb them. The absorber is the per-algorithm balancing term, and its easing rate is proportional to chain pace — that is, to how alive the other algorithms are. Live native algorithms are what pull it back.

Today only a single dying native algorithm remains alive, so a merge-pool exit removes two of the chain's three living algorithms and strands their difficulty with almost nothing left to keep the chain — and therefore the easing — moving. After the upgrade the same exit removes half the algorithms, while two live native algorithms keep the chain pacing and pull the stranded aux algorithms back within two hours:

And the two native picks are not arbitrary. A native algorithm must have maintained miner software today and a hardware audience that does not overlap the other algorithms. The retired four — lyra2re2, groestl, argon2d, yescrypt — split two audiences four ways, and none of the four can hold its algorithm year-round: their miner bases vanish for months at a time (groestl 0.1% share in October 2025, 2.6% in January 2026; argon2d 6.6% in September 2025) — the section-2 fade is what that looks like. The upgrade consolidates each hardware audience into exactly one algorithm:

verushash is CPU-optimized, so the audience is anyone with a computer — the broadest possible participation at nothing to buy to start; equihash 144,5 is the most widely supported memory-hard GPU algorithm, which makes it recruitable and bootstrappable on day one. No algorithm is rental-proof — CPU time rents as cloud compute, GPU hash through the rental markets — which is exactly why containment is structural: difficulty is per algorithm, so a rented burst can take at most one algorithm's quarter of the schedule, and that algorithm's own retarget reprices it within blocks. Both algorithms are implemented and tested.

Consolidation also makes mining pay predictably. Spreading the GPU and CPU communities across four algorithms left each one with too few miners: difficulty whipsawed with every visitor, rewards came in bursts, and no algorithm had enough people to sustain a pool. With one algorithm per hardware class, difficulty is set by people running the same gear you are — rewards arrive steadily, pools work, and any single visitor is small next to everyone else. That is also what the wider ecosystem buys: when every algorithm has real miners, blocks arrive on schedule, and on-schedule blocks are what make payments confirm when people expect. Half the schedule anchored by merge-mined hash that has never left; half carried by the two hardware audiences that actually exist.

The ten longest silences

What ships at the fork height