This section states the relation that is proved for every Action. It fixes the bundle flags that enter the relation as public inputs, states the Action statement over the objects constructed in §2 to §8, pairs each condition with the requirement it serves and with the check outside the proof that completes it, and states the proof system together with the two assumptions on the Action proof that “Security” (§12) uses.
Each bundle carries one flags byte, present exactly when the bundle has at least one Action: for the Orchard-pool bundle of a version or version transaction, and for the Ironwood-pool bundle of a version transaction (ZIP 229, “Transaction Format”). Bits are numbered from the least significant, bit . In both bytes bit is and bit is . The consensus rules of ZIP 229, “Consensus Rules”, fix the remaining bits:
in , bit is , and bits to are reserved and must be ;
in , bits to are reserved and must be , in version and version transactions alike.
ZIP 229, “Transaction Format”, lists at
bit of in its format table; this contradicts
the consensus rules of the same ZIP, which reserve that bit, and the volume
follows the consensus rules. The two texts differ in the role of
the bit, not in its value in a valid transaction: ZIP 258, “Consensus rules
from NU6.3 activation”, requires every Orchard-pool Action to be created
with
Let a bundle have flags byte
and
the value
the value
the value
The Action statement takes as its eighth primary input (same section). No normative text fixes the position of this input in the encoded instance, and none is asserted here.
ZIP 229 calls
The flags are properties of the bundle. Encoded once in its flags byte, they are the same for every Action of the bundle (protocol specification, §“Action Descriptions”, note on the encoding of these components once per pool). The verifier reads them from the bundle and supplies , and as public inputs of each Action; they are not computed inside the proof. They enter the relation only through conditions A8, A9 and A10 of Definition 9.2.
The Action statement is a relation in the sense of the Crypto Guide’s Definition “Language, relation, witness” (§“Languages, relations, and witnesses”), with the primary input as instance and the auxiliary input as witness, each a tuple of the typed objects below, encoded as bit strings by the encodings of §2.1. Every condition is an equation over objects constructed in §2 to §8.
In this definition abbreviates the Pallas group
, with identity , and
The relation is the set of pairs of a primary input
and an auxiliary input
that have the following types and satisfy conditions A1 to A10 below.
Primary input:
Auxiliary input:
Let
the -bit messages of Definition 4.4, with the diversified bases witnessed. The conditions, all equations in or in , are:
Old note commitment integrity:
New note commitment integrity:
Merkle path validity:
Value commitment integrity:
Nullifier integrity:
Spend authority:
Diversified address integrity: with
either
Enable spend flag:
Enable output flag:
Cross-address restriction: the four equations
(Protocol specification, §“Action Statement (Orchard)”, for the typed inputs and conditions A1 to A9; §“Action Descriptions” for the eighth primary input and the meaning of that A10 formalises.)
The anchor is the bundle’s anchor (“Anchors”,
§5.2). The rule
Points of the primary input enter the proof by their coordinates. The specification’s encoding note fixes the first seven primary inputs as the nine elements
of , in this order (protocol specification, §“Action
Statement (Orchard)”, note on the encoding of primary inputs); the
convention
Conditions and the checks that complete them. Each condition is paired below with the requirement of “Requirements on a shielded payment” (§1.4) that it serves and, where one exists, with the check outside the proof that completes it; Table 6 collects the pairing and Figure 3 draws it. All ten conditions constrain one auxiliary input, so the note whose extracted commitment A3 proves to be a leaf is the note that A1 opens and whose nullifier A5 derives, under the key that A7 ties to its address; Lemma 12.5 (§12.2) relies on this.
Conditions A1 and A2 bind the consumed and the created note to the commitments and under , the Sinsemilla commitment with domain z.cash:Orchard-NoteCommit (Definition 4.4), whose binding is Proposition 4.5; they serve R1. Condition A2 fixes the created note’s to the nullifier that the same Action publishes for the note it consumes, the chaining rule of “Nullifier chaining” (§6.3). Its consequences for the uniqueness of and for the Faerie Gold attack are proved in “Double-spend resistance” (Corollary 12.7).
Condition A3 serves R2. Since
The two relaxations are the circuit’s: it does not check that the encoding
of each layer’s input is canonical, and its path check may pass when a hash
on the path, the root included, is , which occurs only when
returns (protocol specification,
§“Merkle Path Validity”, notes on Orchard; §“Action Statement
(Orchard)”, notes on the Merkle path validity check). Remark
“-weakened conditions” below treats the case. The bits
, absent from the specification’s auxiliary input, record the
encodings that the circuit witnesses. Since is odd and has bit
length (§2.1),
Condition A4 serves R4. It is Definition 8.2: the integer
Condition A5 serves R3. It is Definition 6.3; the multiplier
of is the integer in
Conditions A6 and A7 serve R5. Condition A6 is the re-randomisation of
“Randomised validating keys” (§7.2), with
Conditions A8 and A9 have the product form of A3. The values are below
Condition A10 also has the product form of A3: each of its equations is a
public field element times a difference of witnessed values. The four points
lie in , where the affine coordinates determine the point. For
Condition A10 constrains each Action separately: when
For the Orchard pool, whose flag encoding forces
The expanded receiver , the meaning of
that A10 formalises, and the eight-input primary
input with
The statement does not constrain the sign of the witnessed point . Condition A7 reads only , which is the same for and (Lemma 2.4), so the even- normalisation of key generation (“The spending key and the spend-side secrets”, §3.1) is not enforced by the proof (protocol specification, §“Action Statement (Orchard)”, note on the sign of the spend validating key). Theorem 12.10, in “Authorisation of spends”, treats both signs.
Conditions A1, A2, A3 and A7 are the specification’s -weakened forms:
each holds when the Sinsemilla instance it evaluates, , the
hash of or , returns on the witnessed
input, for A3 on some input of the fold, because the circuit evaluates
Sinsemilla with incomplete addition and cannot exclude its exceptional
cases (protocol specification, §“Action Statement (Orchard)”, note on
outputs and note on the Merkle path validity check; §“Merkle Path
Validity”, notes on Orchard). The relation is therefore the
disjunction, not the plain conjunction. The weakening costs a negligible
term. A witness that satisfies A1 or A2 only through contains a
-bit message, or , on which
No condition involves . The point
The trapdoors and and the elements and are free witnesses: no condition checks that and were derived from as in “The note seed” (§4.3). The sender computes that derivation; the recipient checks it, in “Trial decryption and note acceptance” (§10.4), and consensus checks it only for coinbase outputs (“Chain state and pool rules”, §11.4). Hence, in the deployed protocol, the hashed adds no binding: against an adversary able to compute discrete logarithms among the bases of , is not binding, and such an adversary can open one commitment to two distinct notes (ZIP 2005, Rationale, “Attacks against binding of note commitments”). The volume claims nothing further in that setting.
The statement reads none of the Action’s note-encryption fields, which “Note encryption” (§10) constructs; the specification records that the absence of a check on the ephemeral key is intentional (§“Action Statement (Orchard)”, note).
The statement takes the key components as
witnesses and checks them only through A5, A6 and A7. It therefore applies
equally to key components whose and come from the
alternative source of ZIP 2005 (“Changes to the Protocol Specification”,
§4.2.3 “Orchard Key Components”) cited in “The spending key and the
spend-side secrets” (§3.1). The key-derived results of the
volume, namely Proposition 3.18,
Assumption 3.13, Propositions 3.17,
6.10 and 10.6, and the hypotheses on
honest key generation of Theorems 12.10
and 12.13, remain restricted to keys with
Table 6 is the requirement map of the volume. For each requirement of “Requirements on a shielded payment” (§1.4) it lists the soundness content, with the conditions, the checks outside the proof and the results that discharge it, and the hiding content, with the results that discharge it.
| Requirement | Soundness | Hiding |
|---|---|---|
| R1, hidden representation of value | A1 and A2 bind the consumed and the created note to their commitments (Proposition 4.5) | commitment hiding (Propositions 4.5 and 4.8); value-commitment hiding (§8.1); indistinguishability of real and dummy Actions (Theorem 12.13) |
| R2, membership without identification | A3 with the anchor rule of “Anchors” (Proposition 5.9) | zero knowledge of the proof (Assumption 9.15) with a locally computed authentication path (Lemma 5.16), composed in Theorem 12.13 |
| R3, no second consumption | A5 with the nullifier-set rule (Theorem 12.6) | nullifier unlinkability (Proposition 6.10) |
| R4, conservation | A4 with the binding signature (Theorem 12.4) | value-commitment hiding (Theorem 12.13) |
| R5, spend authority | A6 and A7 with the spend-authorisation signature (Theorem 12.10) and bundle binding (Proposition 11.10) | none: R5 has a soundness part only |
| Flags | A8 to A10 serve no requirement of their own: they tie the hidden values and receivers to the public flags of §9.1 | — |
The requirement map pairs each of conditions A1 to A7 with a requirement that it serves. Each of the following conditions is also needed: removed from the relation, it admits witnesses that an accepted proof would then attest, and with them an attack.
Without A1, the note fields of the witness are unrelated to the commitment whose extracted value A3 proves to be a leaf. A prover pairs any leaf, the note of another party included, with an address of its own, any value and fresh , and so consumes one leaf many times, under distinct nullifiers and at values of its choice, against R2, R3 and R4.
Without A3, an Action consumes a note of non-zero value that was never committed, one that the prover invents at its own address with any value: value is created, against R2 and with it R4.
Without A5, the published nullifier is unrelated to the consumed note, so a prover consumes one committed note many times, publishing a fresh nullifier each time, which the nullifier-set rule does not detect: against R3.
Without A6, the key need not derive from the witnessed , so a holder of the full viewing key of a note’s key, which yields no (Proposition 3.18(a)), signs under a key of its own and consumes the note. Without A7, the witnessed key components need not be those of the note’s address, so any party that knows the note, such as its sender, consumes it with key components of its own, under a nullifier deriving key of its choice; step (2) of the proof of Lemma 12.5 then fails as well. Both are theft, against R5.
Without the range
both in
Without the chaining
No party chooses a weaker relation: consensus fixes the verifying key in force (§9.3). A flaw in the circuit that encodes the relation can have the effect of a removed condition; Remark “Outflow bound” (§11.4) names a recorded one.
The Action statement of Definition 9.2 is a relation and prescribes no computation. The post-NU6.3 Action circuit is one fixed PLONKish circuit (Halo 2 Guide, §“PLONKish arithmetisation”) whose constraints are designed to encode the typing and conditions A1 to A10; that every satisfying assignment yields a witness of is part of Assumption 9.12 below. The same circuit serves every Action. Its instance is the encoding of the eight primary inputs of Definition 9.2, bound to the proof as in the Halo 2 Guide, §“Instance columns and binding the public statement”; this instance is the only Orchard-specific part of the proof system constructed in this volume. Proofs are Halo 2 proofs (Halo 2 Guide, §“The Halo 2 proof system” and §“The complete Halo 2 protocol”; protocol specification, §“Zero-Knowledge Proving System”; ZIP 229, Abstract): a PLONKish polynomial interactive oracle proof compiled with the inner-product polynomial commitment and made non-interactive by the Fiat–Shamir transform, with no trusted setup.
The circuit’s native field is , the base field of the
Pallas group, so the Pallas point operations of A1 to A10 are arithmetic in
the native field. The proof’s polynomial commitments lie in the Vesta group
, the group of points of the curve
The circuit has rows and a constraint-degree bound of across its gates and its permutation and lookup identities (Halo 2 Guide, §“From statement to circuit: arithmetisation in practice”). Both are cited constants; in this volume they enter only the Schwartz–Zippel term of the first supporting analysis below.
Every Action of either pool, whatever it consumes or creates, dummy sides included, is an instance of the same relation and circuit. The proofs of a bundle’s Actions are aggregated into one proof, encoded once per bundle, which the verifier checks against the primary inputs of all the bundle’s Actions and the verifying key; each Action carries its own spend-authorisation signature (protocol specification, §“Action Transfers and their Descriptions” and §“Action Descriptions”, note on proof aggregation; ZIP 229, “Transaction Format”). The verification procedure is stated in “Verification of a transaction” (§11.6).
Both pools use the post-NU6.3 circuit and its proving and verifying keys; the pools are distinguished by their note commitment trees, nullifier sets, value pool balances and component position in the transaction, not by circuits (ZIP 229, Abstract, and Rationale, “Reuse of the Orchard protocol with minimal changes”). Consensus selects the Action verifying key by network upgrade (protocol specification, §“Action Descriptions”, consensus rules). The key in force for every Action, in either pool and in version and version transactions alike, is the post-NU6.3 verifying key, named by upgrade and not by pool (ZIP 258, “Consensus rules from NU6.3 activation”). The selection by upgrade bears on soundness: the circuit whose verifying key consensus selected before NU6.2 had a soundness flaw in its implementation, not in the Action statement, which could have allowed the creation of value and the theft of funds; the corrected circuit has a different verifying key, and a proof made for one of the two circuits verifies only under its key (protocol specification, §“Action Statement (Orchard)”, vulnerability disclosure; ZIP 257, Motivation). A flaw of this kind in the post-NU6.3 circuit would falsify Assumption 9.12 below, which includes that every satisfying assignment of that circuit yields a witness of the relation; Remark “Outflow bound” (§11.4) states what the chain value pool balances bound in that case.
For every classical probabilistic polynomial-time algorithm, given a generator of and for uniform in , the probability of outputting is negligible (Crypto Guide, §“Standing assumptions”, Remark “Standing assumptions for the sequel”, item 1, for the Vesta curve). Concretely, the best known classical attack costs about group operations (Crypto Guide, §“Instantiation on elliptic curves; the Pasta curves”, Proposition “The Pasta curves against the criteria”); the level is cited, not recomputed. The volume uses this assumption only in the extraction analysis of the inner-product commitment that supports Assumption 9.12 (the second supporting analysis below).
Let the hash of the Fiat–Shamir transform of the proof system be modelled as
a random oracle. For every efficient adversary that outputs primary
inputs
| is accepted for under the post-NU6.3 verifying key | ||
is negligible. This is the knowledge-soundness clause of the Crypto Guide’s Definition “SNARK” (§“SNARKs: succinct non-interactive arguments of knowledge”) for the relation of Definition 9.2. It includes that every satisfying assignment of the post-NU6.3 circuit yields a witness of , for which the specification requires canonical decompositions of the scalar of A5 and of the scalar in (protocol specification, §“Action Statement (Orchard)”, note on canonical scalar decompositions).
Assumption 9.12 is an assumption, not a proved result. Its status is designed but unspecified: bounds exist for the analysed protocol, and no theorem carries them to the deployed Fiat–Shamir-compiled protocol (Halo 2 Guide, §“The algebraic group model”). The volume assumes the property, proves no bound and asserts no loss factor. The two supporting analyses that follow are cited, not re-proved, and not combined into a bound.
A non-zero gap polynomial over , fixed by the prover’s
commitments before the challenge is drawn, vanishes at a uniform challenge
with probability at most (Halo 2 Guide,
§“Why checking one random point is a proof”). At the Orchard parameters,
rows and constraint-degree bound , the constraint expression has
degree below
The inner-product polynomial commitment, whose commitments lie in , has an interactive opening protocol that is knowledge-sound under the discrete-logarithm assumption in that group, Assumption 9.11, with an expected-polynomial-time tree extractor (Halo 2 Guide, §“The inner-product polynomial commitment”, Theorem “Knowledge soundness of the opening protocol”). The Fiat–Shamir-compiled argument inherits the interactive extraction bound only up to a loss in the adversary’s random-oracle query count (§“From the interactive extractor to the non-interactive argument”), and the tighter analyses in the algebraic group model are proved for the analysed protocol, not for the deployed transcript (§“The algebraic group model”). No loss factor is asserted.
Let the hash of the Fiat–Shamir transform be modelled as a programmable
random oracle. There is an efficient simulator that, given only the primary
inputs
Assumption 9.15 is an assumption, not a proved result. The Halo 2 Guide, §“Zero knowledge: hiding the witness in Halo 2”, records honest-verifier zero knowledge of the interactive protocol and zero knowledge of its Fiat–Shamir compilation in the programmable random-oracle model, subject to its composition and Fiat–Shamir hypotheses, without reproving it. Assumptions 9.12 and 9.15 are the two assumptions on the Action proof that “Security” (§12) composes with the propositions of the preceding sections.