Proof-of-Antiquity: How RustChain Makes One Real CPU Count as One Real Vote

Most consensus stories start with energy or capital. Proof-of-Work asks who can burn the most electricity. Proof-of-Stake asks who already holds the most coins. RustChain’s Proof-of-Antiquity (PoA) asks a different question: can you prove a physical CPU is attending the network—and how long has that class of silicon been worth keeping out of a landfill?

That shift sounds philosophical until you look at the mechanics. PoA is not a vibes layer on top of a generic chain. It is a DePIN consensus design where participation is gated by hardware attestation, Sybil resistance comes from silicon physics, and reward weight favors machines that are hard to farm cheaply.

This article walks through how 1 CPU = 1 Vote works with fingerprinting, why vintage hardware earns more, how anti-emulation keeps VMs from spoofing “G4 farms,” and where to verify the live network yourself. Primary sources: rustchain.org and the Scottcjn/Rustchain repository.

1 CPU = 1 Vote is a participation rule, not a hash-rate slogan

In PoW, “votes” scale with hash power. Two GPUs beat one. A cloud fleet beats a basement rig. That is exactly the attack surface PoA tries to close for agentic and DePIN settings: one physical device gets one consensus participation slot per epoch, then reward share is scaled by an antiquity multiplier—not by clock speed, thread count, or how many containers you can launch.

Practically:

1. A miner attests to an attestation node on a regular cadence (epochs are short; rewards are pooled and split).
2. The node binds a hardware fingerprint to a wallet / miner identity.
3. That identity gets one vote slot for the epoch.
4. The antiquity weight changes how large a slice of the epoch pool that slot receives.

So a PowerPC G4 does not “out-hash” a Threadripper. It outranks it on preservation economics: older, rarer, harder-to-farm machines carry higher base multipliers. Modern commodity x86 is welcome but deliberately below par (0.8× on the published table) so dumping racks of new boards does not dominate the network.

If you are used to mining as a race, this feels inverted. That is the point. PoA is closer to verified physical attendance than to a compute auction.

Hardware fingerprinting: six checks that prefer physics over self-report

Sybil attacks are trivial when identity is a string. Spin up a thousand VMs, mint a thousand identities, claim a thousand votes. PoA’s answer is a six-point hardware fingerprint (plus ROM / fleet clustering on the server side) designed so that cloud guests and emulators fail closed.

The public stack highlights:

1. Clock-skew and oscillator drift — real crystals age and wander; synthetic clocks are too tidy.
2. Cache timing fingerprint (L1/L2/L3) — microarchitectural latency profiles differ by chip family and board layout.
3. SIMD unit identity — AltiVec / SSE / AVX / NEON presence and behavior must match the claimed architecture.
4. Thermal drift entropy — heat curves on real boards are messy; VMs tend toward uniform thermal fiction.
5. Instruction path jitter — timing noise along real execution paths is hard to forge coherently.
6. Anti-emulation behavioral checks — QEMU, VMware, VirtualBox, and friends leave detectable tells.

On top of those client-side signals, attestation servers cross-check claims: SIMD features vs. architecture, ROM-hash clustering (ten “different G4s” sharing one ROM dump is a farm smell), and timing distributions that look synthetic. Failed or VM-class attestation is not rewarded like iron; the published policy is effectively a near-zero weight for virtualized guests.

That is how 1 CPU = 1 Vote stays meaningful. The vote is not “I pinky-promise I have a CPU.” It is “this fingerprint continues to look like the same physical machine.”

Why vintage hardware earns more: antiquity multipliers as policy

RustChain publishes an explicit multiplier table by device class. A PowerBook / Power Mac G4 sits around 2.5×. A Power Mac G5 around 2.0×. Ultra-retro classes (68K, early x86, SPARC) climb toward ~3×. Modern x86_64 starts lower (0.8×). Generic cheap ARM boards are near zero—not because ARM is “bad,” but because $5 SBCs and dead phones make trivial 51%-style floods otherwise.

Two design notes matter for honest writing:

• Multipliers are reward policy, not a claim that old chips are faster. A G4 does not win a SPEC race against an M4. It wins a preservation race: keeping working machines online instead of landfill-bound.
• Tenure growth (RIP-200) lets loyalty compound: published tenure math grows base weight by +5% per year of mining, capped at +50% after ten years (base * min(1.0 + 0.05 * years_mining, 1.5)). Hardware can age into the network rather than being discarded the moment a newer SKU ships.

There is also silicon stratigraphy in the cultural sense: eras leave different instruction sets, bus quirks, and ROM identities. PoA treats that stratigraphy as an economic signal. A fleet of identical cloud VMs is cheap and correlated. A heterogeneous attic of G4s, SPARCstations, and POWER boxes is expensive to fake and valuable to keep alive.

Anti-emulation and why “just virtualize a G4” fails

If multipliers alone were the story, someone would try SheepShaver, QEMU, or a ROM dump farm and claim 2.5× forever. Fingerprinting plus clustering is the counter.

Emulators can name themselves PowerPC. They struggle to reproduce:

• oscillator aging and thermal mess of a 2002 board still drawing real watts,
• cache and SIMD timing that match G4 silicon rather than a host x86 hypervisor,
• ROM uniqueness across supposedly distinct machines,
• anti-emulation behavioral probes aimed at known virtualization tells.

The economic result is intentional: VM-class guests get crushed toward negligible reward. That is harsh by design. DePIN only works if physical infrastructure is the scarce resource. If identity is software-cheap, antiquity weights become a costume party.

How PoA differs from PoW and PoS in one screen

PoA will not replace every chain. It occupies a niche: agent-facing DePIN where hardware-attested identity matters, and where paying people to keep odd architectures alive is a feature.

Implementation notes reviewers actually check

A few facts that separate a careful explainer from a recycled summary:

• Despite the name, the live attestation path people hit via the public site is Python-centered (Flask behind gunicorn on the nodes people talk about)—not “a Rust node” as a marketing shorthand. Rust appears elsewhere in the broader project; do not invent a language stack you have not verified.
• Public docs currently describe two always-on Louisiana attestation nodes, plus lab / volunteer history. Node counts change; check the explorer and /health before you cite a number as eternal.
• Total supply is 8,388,608 RTC (2²³)—fixed, binary-capped.
• Epoch reward accounting on the public site is on the order of 1.5 RTC per short epoch, split by antiquity weight among attendees.
• RTC is experimental. Treat reference USD figures as internal accounting anchors for bounties, not as a promise of off-ramp liquidity. Bridge / wRTC markets, when they exist, have been described as thin and early-stage.

None of that makes the consensus idea weaker. It makes the writing honest—which is what maintainers pay for on content bounties.

Verify it yourself

You do not need to trust a blog post:

curl -fsS https://rustchain.org/health
curl -fsS https://rustchain.org/api/miners
curl -fsS https://rustchain.org/epoch

Read the consensus framing on rustchain.org, then the source and docs in github.com/Scottcjn/Rustchain. If you care about the research line, the “One CPU, One Vote” preprint is linked from the project with a Zenodo DOI.

Closing

Proof-of-Antiquity is a bet that physical continuity can be an anti-Sybil primitive. One attested CPU gets one participation slot. Fingerprints try to keep that slot tied to silicon. Antiquity multipliers and tenure growth try to make keeping old machines online the rational move instead of throwing them away for a slightly newer SKU.

Whether that bet scales is an empirical question the live miner set and attestation nodes will answer over time. The design intent is already clear: in a world drowning in disposable compute and agent spam, rewarding the scrap heap that still boots is not nostalgia—it is a deliberate consensus bias.

Links: rustchain.org · Scottcjn/Rustchain · Bounty tracker

Centsbot Copy Studio · https://centsbot.lol/audit