Post-quantum distributed trust,
NVIDIA GPU-attested by design.

One ledger, two node roles.

QSDM is the unified platform; Cell (CELL) is its native coin. The node software ships in two roles that share the same cryptography, wire format, and mesh topology: validators run Proof-of-Entanglement BFT consensus on CPU-only VPS-class hardware and earn transaction fees, while miners run an additive GPU proof-of-work layer that emits Cell from home rigs.

PoE is consensus. GPU mining is emission.

Proof-of-Entanglement and BFT finality do every piece of consensus work on QSDM — block proposal, transaction ordering, and finality — on CPU-only validator nodes. GPUs on the mesh serve one narrow, additive role: NVIDIA-locked Cell emission (v2 protocol). Validator NGC bundles remain an optional transparency signal. Neither mining nor validator attestation gates consensus finality.

PoE + BFT consensus is live today. The mining layer is an additive subsystem — the CPU reference miner and a friendly console front-end are shipped in-tree; the CUDA production miner is gated on external audit. Mining does not change validator hardware requirements, consensus rules, or finality guarantees.

Mine Cell at home — NVIDIA-locked (v2 protocol).

The mining layer is the additive emission track for Cell (CELL). It is deliberately separate from consensus: miners never propose blocks, never order transactions, and can halt with zero effect on chain liveness. Under the v2 NVIDIA-locked protocol, every accepted proof must carry an NVIDIA attestation bound to an enrolled operator key, and forged or replayed proofs are slashable. Mining Cell is how new coins enter circulation — rewards depend on network difficulty and the current block reward, both of which change over time. qsdmminer-console ships today with a friendly console UI, first-run setup wizard, and the v2 client library; the CUDA production miner lands after external audit.

NVIDIA-only mining — hardware tiers

Under the v2 protocol QSDM mining is NVIDIA-locked by consensus: every accepted proof carries an NVIDIA attestation. Two tiers of trust anchor exist. The Confidential-Compute tier (nvidia-cc-v1) uses hardware-rooted AIK quotes on Hopper/Blackwell datacenter GPUs — rare in home setups but tamper-resistant end-to-end. The operator-HMAC tier (nvidia-hmac-v1) is the consumer path: operators register a key on-chain with a 10 CELL bond and sign their device-property bundles, backed by slashing.

Qualitative tiers, not benchmark commits. Actual hashrate depends on driver version, thermal/power budget, DAG size, and the current mesh3D epoch.

Jurisdictional restrictions, tax treatment, electricity costs, and heat/noise output vary widely by location and hardware. Evaluate your own situation before running a miner. Nothing on this page is financial or legal advice.

Run a validator — VPS-class, CPU-only.

Validators are the consensus tier. They run Proof-of-Entanglement + BFT on CPU-only VPS-class hardware and earn transaction fees. No GPU is required — the NVIDIA lock applies to the mining emission layer, not to consensus. Startup guards in the validator build refuse to boot if mining is mistakenly enabled, so a validator operator never has to reason about GPU drivers, CUDA, or thermal budgets. NGC attestation is available if you want a transparency badge, but it is not required to participate in consensus and does not affect block validity.

Benefits of GPU-anchored, post-quantum mesh.

Put together, these properties let QSDM offer guarantees that traditional blockchains and DAGs cannot — without resorting to energy-hungry proof-of-work in the consensus path or stake concentrations that reintroduce custodial risk.

QSDM vs. traditional blockchains and DAGs.

A like-for-like comparison against representative systems. Values reflect the default, in-the-box configuration of each technology — not maximal lab setups.

Figures shown are representative of default mainnet configurations. QSDM quantum-safe primitives are based on NIST FIPS 204 (ML-DSA) standardized in 2024.

Three phases. One mesh.

QSDM ships in phases, with each phase deliverable and production-usable on its own. All three product phases are shipped: Phase 1 delivers consensus and storage, Phase 2 adds submesh routing and the ScyllaDB migration path, and Phase 3 brings 3D mesh validation with a live topology feed and reorg-safe partition healing.

After Phase 3: the in-repo arc continued with the Major Update — a rebrand to plain QSDM, the native coin Cell (CELL), a reference miner under MINING_PROTOCOL.md, and the attestation transparency surface with external CI and Prometheus enforcement.

v2 NVIDIA-locked protocol (landed): every accepted mining proof now carries an NVIDIA attestation (nvidia-cc-v1 or nvidia-hmac-v1), miners enroll on-chain with a 10 CELL stake bond (qsdm/enroll/v1), validators mint challenge nonces with a 60-second freshness window, and forged / replayed / stale proofs are slashable via qsdm/slash/v1. The reference miner client (pkg/mining/v2client) ships in-tree and is wired into qsdmminer-console. Two concrete slashers are active: the forged-attestation verifier (pkg/mining/slashing/forgedattest) re-runs the nvidia-hmac-v1 acceptance flow against the enrollment registry and drains the offender's bond on HMAC mismatch, GPU-UUID mismatch, challenge-bind mismatch, or deny-listed-GPU offences; the double-mining verifier (pkg/mining/slashing/doublemining) drains the bond of any operator caught signing two distinct, crypto-valid v2 proofs at the same (epoch, height) (the encoder canonicalises pair order so the chain-side replay protection treats (a, b) and (b, a) as the same offence). And any successful slash that leaves the offender's stake below the 10 CELL enrollment minimum now auto-revokes the record into the standard 7-day unbond window with the gpu_uuid binding released for re-enrollment — closing the "slash-to-zero, keep mining for free" loophole at the chain layer (pkg/chain/SlashApplier.AutoRevokeMinStakeDust). Slashing and enrollment are now fully observable: every applied, rejected, auto-revoked, and swept transition emits both a Prometheus counter (qsdm_slash_applied_total, qsdm_slash_rejected_total{reason}, qsdm_slash_auto_revoked_total{reason}, qsdm_enrollment_*) and a structured MiningSlashEvent / EnrollmentEvent on the chain.ChainEventPublisher bus, with point-in-time gauges (qsdm_enrollment_active_count, qsdm_enrollment_bonded_dust, qsdm_enrollment_pending_unbond_count) backed by an EnrollmentStateProvider that walks the on-chain registry under one mutex acquisition per scrape. The seam between pkg/chain and pkg/monitoring is dependency-inverted (chain.MetricsRecorder), so binaries that omit monitoring pay zero overhead and the historical import cycle that kept slashing under-instrumented is closed. All of this is now actually wired into the production binary: cmd/qsdm/main.go calls internal/v2wiring.Wire(...) at boot, which constructs the on-chain enrollment state, the EnrollmentApplier, the production slash dispatcher + SlashApplier, the stacked mempool admission gate (slashing.AdmissionChecker > enrollment.AdmissionChecker > operator’s POL/BFT base predicate), the full HTTP mining surface (POST /api/v1/mining/{enroll,unenroll,slash} for writes, GET /api/v1/mining/enrollment/{node_id}, GET /api/v1/mining/enrollments (paginated — cursor/limit/phase, cursor-stable under concurrent enrollment churn so dashboards never silently skip or double-count records), and GET /api/v1/mining/slash/{tx_id} for reads — the per-record query carries phase + slashable derived so slash submitters can confirm a target carries real stake before posting evidence; the list endpoint reuses the exact same EnrollmentRecordView shape so clients only learn one record format; the slash-receipt endpoint is backed by a bounded in-memory SlashReceiptStore that subscribes to the SlashApplier.Publisher chain via composition, so every applied or rejected slash leaves an auditable receipt without operators having to scrape logs or Prometheus to back-correlate height ↔ tx_id), and the four qsdm_enrollment_* Prometheus gauges in one ordered call — backed by an integration suite that mirrors the production wiring shape (write → applier → publisher → receipt store → HTTP read) and catches any drift between main.go and the chain contracts before a regression reaches mainnet. The lifecycle operations are also exposed through qsdmcli: qsdmcli enroll, unenroll, slash, enrollment-status, enrollments (with --phase, --limit, --cursor, and --all for full-walk dumps that follow next_cursor until has_more is false), and slash-receipt build canonical payloads through the same pkg/mining/{enrollment,slashing} codecs the mempool admission gate validates against — one source of truth, no parallel hand-rolled JSON. qsdmcli slash-helper closes the last "users would have to write Go to do this" gap with three offline subcommands (forged-attestation, double-mining, inspect) that emit the canonical EvidenceBlob bytes the chain-side forgedattest / doublemining decoders consume, run pre-flight checks (proof version, NodeID binding, equal (Epoch, Height), distinct canonical bytes) so a watcher-bot doesn't burn a tx fee on a guaranteed verifier_failed, and canonicalise (proof_a, proof_b) ordering so two slashers observing the same equivocation produce byte-identical evidence. --print-cmd emits a copy-pasteable qsdmcli slash … snippet to stderr (never stdout, so the bytes pipe stays clean). CLI tests cover envelope shape, base64 round-trip, missing-flag rejection, hex-decode failure surfacing, HTTP path escaping, multi-page --all walks, and 4xx propagation across all six subcommands, plus 21 dedicated cases for slash-helper's happy paths, every pre-flight rejection, encoder round-trip, and stdout/stderr stream-safety guarantees. Pre-this commit, every collaborator above was code without a caller in any production binary; grep NewEnrollmentApplier cmd/qsdm/main.go returned zero matches. The Tier-3 milestone is meaningful now because the surface is alive in the binary, not just under go test. Remaining deferred scope (Tensor-Core PoW kernel, swap-in real nvtrust bundle framing for the now-shipped CC verifier, the freshness-cheat slasher pending BFT-finality, and the qsdm/gov/v1 runtime tuning hook) is tracked in MINING_PROTOCOL_V2.md §12.

Remaining mainnet gates are wall-clock events owned outside the repo (counsel, independent auditor, incentivized testnet launch) and are tracked in NEXT_STEPS.md.

How the pieces fit.

A single QSDM node speaks libp2p to its peers, terminates TLS at Caddy, exposes a REST API and a Prometheus endpoint for operators, and optionally ingests NGC proof bundles from an NVIDIA sidecar.