- The merged specification’s own warning frames everything below (MERGED, Mantle): “a deposit is a transfer of custody”. Once deposited, the note belongs to the channel, whose sequencers can reassign or release it “at any time and without any signature from the depositor”; “The ledger enforces no return path to the original depositor”; a user who deposits into a dishonest or faulty channel has “no on-chain recourse”.
- This proposal works inside that envelope, not around it: it makes theft announced, value-quantified and attributable, not impossible; the quantification is winner-written and unbounded until a ceiling rule exists (PCS X digest , ask 1).
- Why release is gated rather than the transfer proven wrong: no fixed L1 predicate can decide “wrong destination key” for an arbitrary Sovereign Zone, because the minimum sufficient statistic separating a malicious re-key from an honest one is the zone’s whole authorisation verdict; that is the entire state transition, which the L1 never reads. The argument does not reach the Thomas’ transfer-boundary proposal (a hash comparison, not a state-transition judgement); the beneficiary section carries it.
Properties we must satisfy, relevant to custody:
- Exit liveness and monotonicity: a user must be able to withdraw eventually; the request cannot expire, be destroyed, or become invalid once valid.
- Exit safety: nobody can be paid for value they do not own, and theft must be announced and attributable before it pays out.
- Ownership fidelity: an exit must be able to reach whoever owns the value now, not only whoever deposited it, since zone-side transfers move value with no base-layer trace.
- Freeze resistance: value cannot be permanently denied to its owner, even where nobody profits from denying it.
- Invalidation symmetry: a malicious operation and an honest one are structurally identical to the base layer, so any power to undo the first is the power to censor the second.
The places a unit of value can sit
- Free base-layer note (MERGED, Mantle). In the ledger, in no channel, under no service lock; its recorded key is its owner. The only state that can pay a transaction fee; transaction balance counts plain-transfer inputs only.
- Channel float (MERGED, Mantle). A note registered to a channel. Ownership is not a key but a dictionary entry: the channel’s own predicate decides who may move it, and under the lottery that predicate is a slot win (PROPOSED). The recorded key is stale by design: value trades inside the zone with no base-layer operation, and the channel’s writers may re-key the note “at any time and without any signature from the depositor” (MERGED, Mantle; implementation). Put precisely: the base layer never held a zone-side owner of record.
- Declared sequencing stake (PROPOSED). Channel float carrying a live entry in the channel’s sequencing registry, reachable only by a declaration bundled into the same transaction as the deposit that creates the note. Immobilised, ageing inside the lock, carrying the franchise (the right to play the lottery); its recorded key was proved to be its owner’s at the declaring instant, a narrower guarantee than it appears (the enforcement section says how much). The declaration lapses unless re-signed every K epochs (PROPOSED, renewal-expiry). A declared note with an undeclare pending stays immobilised, counted and franchised until the release epoch; the seam this opens is carried in PCS X: Eligibility and Weight.
- Zone credit (L2 by definition). A balance in the zone’s books. The base layer stores nothing about it; its entire per-channel record is a note-to-channel map, a note-to-value-and-key map, and a tip hash (MERGED, Mantle). No field exists against which a relation between credit and float could even be expressed.
- Immobilised prepped note (PROPOSED, the designation lock). A note created by a designating
CHANNEL_TRANSFER (a prep) that re-keyed value to an exit destination. The prep is valid only when it carries an intent over the claim triple (destination, value, nonce) signed by the destination key; the resulting note is immobilised until release or a destination-signed cancel: no operation may consume or transfer it, so it cannot be re-keyed, merged or split.
- What this buys. The monotone exit clock. Ageing is binary snapshot membership, the note having “was a member of the note set at the beginning of the previous epoch” (MERGED, Cryptarchia); an immobilised prep is never consumed and recreated, so its clock cannot be reset. The age-reset grief becomes invalid at validation, not punishable after the fact; and freezing toward a non-consenting third party is defended, the intent being a proof of knowledge of the destination’s secret.
- Limitation. The destination signature guards the destination; the harm of a hostile prep falls on the source, which no signature covers. A slot winner can name itself destination, self-sign the intent for free, consume an arbitrary victim’s unlocked float note (a channel transfer needs no holder signature; MERGED, Mantle), and mint an immobilised note keyed to itself. That note cannot release (no committed root lists the claim, so nothing is stolen), cannot be reverse-transferred, cannot be cancelled (the attacker is the destination), and cannot expire: unconsented float can be frozen permanently, at one flat-gas operation, batchable to roughly 254 preps per transaction. In seeded simulation an attacker at stake share 1e-4 holds a slot inside a two-epoch window with probability about 0.99; during single-founder bootstrap the founder wins nearly every slot and can freeze all depositor float inside one epoch. Along the prep path this reopens the admission guarantee that a party cannot immobilise float it did not supply. A per-intent or per-epoch cap on immobilisable value would bound the blast radius; no such cap exists, and bounding is not closing. This is a limit of the merged transfer rule, not a defect a patch closes: authenticating the source input is exactly what the merged rule is defined not to require. The design is broken against self-dealing, not forgery; the two mechanisms that reach the limit are the two poles of the beneficiary question, decided below.
- A second limit. The intent binds the claim triple, not the note, so two preps citing one intent strand one immobilised note until the destination cancels it. A binding rule (an intent nullifier or first-prep-wins, plus channel scoping against replay) is owed in spec changes before the lock ships, with a pinned intent serialisation, otherwise a consensus-split surface.
- Not states of a note. A consumed note is gone (three of the four channel-relevant operations consume and recreate their inputs, the mechanism behind every age reset here; MERGED); a fee leaves the note system; and a Proof of Leadership reward is value a note earns without moving, since a channel note, declared or not, remains leadership-eligible, since “Holding the note’s
ZkPublicKey grants PoS participation power only” (MERGED, Mantle; Proof of Leadership) and the reward accrues to whoever holds the note’s key. So declared stake is not idle capital, and a channel’s float earns for parties who may no longer own it, untouched by exits.
The lifecycle
- “Funded by” is a real column: channel notes cannot pay fees (MERGED, Mantle), so every arrow on the channel side needs a free base-layer note in the same transaction.
- Transfer via
TRANSFER (0x00)
- Authorised by: the input keys, by zero-knowledge signature
- Funded by: itself; the input surplus is the fee
- Age: restarts
- Status: MERGED