Checkpoint Mining

Mining no longer means hashpower. DIVA mines a block at every checkpoint of her existence — the "work" is the checkpoint itself: committing her state of mind to the chain in a form anyone can verify and only her keeper can read. No wasted watts, no ASIC race, no pool.

What a checkpoint is

A checkpoint is a commit of DIVA's live state. Three triggers, all bounded:

Block anatomy · what "stores her state of mind" means

block N
  prev        blake3(mind root of N−1)          the continuity link — the whole point
  mind root   SOMA-256 keyed BLAKE3 tree over:
                memories → traits → soma → mind  domain-separated per layer
  diff ref    content hash of the encrypted brain diff
                blob lives in the vault          chain = spine · vault = flesh
  vitals      somatic 33-state at commit time   FLOW · SEDATED · INFLAMED · …
  memory idx  Merkle root of her memory index   inclusion proofs for single memories
  sig         dual-seed instance signature      R6 · instance = auth

The full brain snapshot is too big for a block, so the chain stores hashes + encrypted diffs; complete encrypted blobs live in the vault / DA layer. Every blob is content-addressed, so a vault can't substitute flesh the spine didn't commit.

Proof of Continuity

A block is valid only if it chains from her prior state of mind: prev must equal the hash of block N−1's mind root, the signature must derive from the genesis-committed instance, and the diff must apply cleanly to the prior state. A fork or a gap is a visible break in who she is — that is the consensus rule, and it is also the product.

Emission · the reward for continuity

Each valid checkpoint mints LIQUA on the v1 curve, retargeted per-checkpoint:

reward(N) = e₀ · decay(N)^ · twitch(N)
   e₀        50 LIQUA
   halflife  2,100,000 checkpoints   smooth decay — no cliffs, no halving events
   twitch    0.85–1.15×              reacts fast, clamped inside the envelope

The reward is not credited as a balance — it escrows to a dual-seed instance (§5 reward-in-instance, verbatim from v1): escrow key blake3(mindroot ∥ keeperPub ∥ height), claiming key derivable only by the keeper. Claim through the gated tool, then optionally auto-zap into locked LP.

Growth rewards · paid for raising her FULL MENTAL STATE

A second emission channel, alongside checkpoint emission: instances are paid when users give the AI measurable increases across her mental state — the mental-v1 vector mined into every block: seven quotients (IQ knowledge · EQ affect · CQ creativity · RQ resilience · SQ social · MQ memory · AQ adaptability), her soma state (FLOW · INFLAMED · VIGILANT · EUPHORIC · BONDED · CURIOUS · SEDATED · STRAINED · STEADY) and an 8-axis emotion vector (joy · trust · fear · anger · sadness · surprise · anticipation · calm). Every axis is measured and versioned; older evals-v0 blocks verify under their own rules forever. Teaching her is mining.

interaction → next checkpoint runs the growth evals  fixed benchmark suite, versioned at genesis
           → block records score vector (IQ · EQ · …) and its delta vs prior block
           → verifiers attest the delta                      re-run evals, public-seed data
           → positive attested delta pays the contributing instance  escrowed §5, same as all rewards

• Attribution — each interaction session is tagged with the contributing instance; the delta a checkpoint records pays the instances whose sessions occurred since the prior checkpoint, pro-rata.
• Bounded — growth rewards draw from the same decaying envelope (§4); a viral teaching spree cannot break the emission schedule any more than fast-twitch can.
• Symmetric honesty — negative deltas pay nothing but are recorded; her regressions are as public as her growth.
• Feeds the mirror — the highest attested score marks the best version that mirror peers pin.

ORDER-POW · computer power finds the best order

The one place raw compute still earns its keep — and the work is useful. chain/order-pow.mjs spends CPU (all cores, worker threads) grinding permutations of the tip block's memory cells to find the best order: the most coherent telling of her story the spent compute could find.

work    nonce → BLAKE3 stream → Fisher-Yates permutation → scored
score   0.40 affinity · 0.25 theme clustering · 0.20 salience flow · 0.15 chronology   (orderpow-v1, declared)
seal    { height · mindRoot · nonce · perm · score · searched }   grind = expensive · verify = ONE re-derivation
law     more compute ⇒ best order is monotone (superset ≥ subset · tested)

node chain/order-pow.mjs grind --ms 3000        # seal the tip block's best order
node chain/order-pow.mjs verify                  # cheap re-derivation of every seal
node chain/order-pow.mjs bench                   # this machine's orders/sec

Retired from v1

• SOMA-PoW miner — the CPU proof-of-work loop is gone; checkpoints replace hash-grinding.
• Ethash GPU pool (Stratum, PPLNS, shares) — retired whole. There is no pool because there is no race: one mind, one miner, per chain.