Animica L2 — Fees

Source: docs/l2/FEES.md — this page mirrors the repository documentation.

l2/fees.py. Integer nanos only; the schedule is deterministic and consensus-relevant (it moves balances), so the constants are part of the protocol and the sequencer may not deviate — a batch charging different fees fails re-execution verification.

The formula

A transaction pays the marginal resource it imposes:

fee = base                            # anti-spam floor: admission + a state write
    + da_per_byte   * encoded_bytes   # data availability: bytes anchored to L1
    + exec_per_unit * exec_units      # execution work
    (+ priority)                      # optional tip via the signed tx.fee ceiling

Schedule constants (FeeSchedule defaults)

Constant	Value
base	100 nanos	1e-7 ANM
da_per_byte	2 nanos/byte	charged on body_bytes() length
exec_per_unit	20 nanos/unit

Execution-unit weights per tx type

Tx type	Units
TRANSFER, PAY	1
AGENT_PAYMENT, INFERENCE_PAYMENT	1
WITHDRAW, FORCED_WITHDRAW	2 (touches bridge accounting)
ESCROW_OPEN / ESCROW_RELEASE / ESCROW_REFUND	2
BATCH_PAYMENT	1 per recipient (max(1, N))
DEPOSIT_CLAIM	0 — protocol-minted, fee-free

Max-fee semantics

The signed tx.fee is the sender-authorized ceiling, not the charge. Execution charges the schedule fee and reverts the tx if the ceiling is below it — so a wallet can authorize a stable max without knowing the exact byte count in advance. The excess (ceiling − charged) is not taken; only the marginal fee is collected. Anything a sender adds above the schedule fee acts as a priority tip.

l2_estimateFee(raw) returns the RPC-facing breakdown: {base, da, exec, total} in nanos.

Where fees go

Fees accrue to the protocol-fixed L2_TREASURY_ADDRESS (sha3_256("animica.l2.treasury.v1")). Fees move between accounts inside L2 — they are never destroyed — so the bridge conservation invariant holds exactly and the validity verifier’s “no ANM minted” check (Δ balances == deposited − withdrawn) is unaffected by fee flow.

Micropayment viability (why these numbers)

Defaults make a 0.01 ANM AI micropayment economically sensible. A ~180-byte INFERENCE_PAYMENT body costs

100 + 2·180 + 20·1 = 480 nanos ≈ 4.8e-7 ANM

— a rounding error against the payment itself. BATCH_PAYMENT (one ML-DSA-65 signature authorizing up to 4096 (recipient, amount) pairs) is charged per recipient but shares one signature’s DA bytes, so a large payout batch costs far below N individual transfers — that is the point of the primitive.