Skip to content

Research H WES Theorem and Object Model

blueokanna edited this page Oct 18, 2026 · 1 revision

H-WES: From Recovery Lower Bounds to Expirable Authenticated Objects

Status: inactive research specification. It fixes the problem, model, theorems and implementation interface; it is neither a novelty proof nor a cryptographic audit report.

1. Correcting an over-strong impossibility statement

The following alone do not imply impossibility: old versions are not recoverable, a client initially keeps only a historical witness, validators do not persist a key -> latest authenticated index, and recovery is noninteractive or constant-round. A counterexample sends the entire archive in one round. The verifier recomputes the append-only commitment and scans every version of the key. It keeps no latest index and rejects stale versions, at Theta(N) communication, prover I/O and verifier work.

The provable goal must therefore include succinct recovery. The model below treats the archive as a black-box authenticated store and requires recovery to open a sublinear number of archive cells.

2. State and archive model

For security parameter lambda, key space K and value space V, an object incarnation is

I = (cid, class, k, v, h_val, P_owner, h_create, h_lease),

where versions for one key strictly increase. At height t,

Hist_t = (r_0, ..., r_(N_t-1)),
A_t = CommitArchive(Hist_t).

The archive exposes only authenticated positional opening and append checks:

Open(A_t, i) -> (r_i, pi_i),
VerifyOpen(A_t, i, r_i, pi_i) -> {0,1},
VerifyAppend(A_s, A_t, pi_(s,t)) -> {0,1},  s <= t.

It has no native key query, range non-membership or maximum-version operation. Define Head_t(k) as the maximum-version record for k; consensus rejects duplicate versions before append.

An authenticated latest oracle is any (D_t, ProveLatest, VerifyLatest) that authenticates Head_t(k).(version,status). D_t may be an SMT, Verkle or vector commitment root, a SNARK statement or another digest. A short freshness proof is semantically equivalent to a latest dictionary even when no BTreeMap exists.

3. Lower-bound theorems

Theorem H-WES-N0: secure recovery implies authenticated head recovery

Let recovery Pi=(ProveRec,VerifyRec) be complete for the current expired incarnation, latest-only sound against old or non-expired incarnations, and use public verifier state S_t. Taking D_t=S_t, running ProveRec without the final mutation and running the read-only phase of VerifyRec constructs an authenticated predicate

IsRecoverableHead_t(k,v)=1 iff Head_t(k).(version,status)=(v,Expired).

Completeness and latest-only soundness transfer directly. The theorem need not construct a full latest dictionary, but it shows that secure recovery without any latest-equivalent authentication is semantically contradictory. □

Theorem H-WES-N1: suffix-query lower bound for membership-only archives

Treat A_t as an opaque oracle handle usable only by Open/VerifyOpen. A candidate at position i has suffix length m=N_t-i-1. Without a latest oracle, a zero-error verifier deciding head status for every legal history must open all m suffix cells in the worst case.

If an accepting execution leaves position j unopened, compare histories H0, where no later record has key k, and H1, identical except that cell j is a newer valid version of k. All observed openings are identical, yet the candidate is latest only in H0; one of completeness or soundness fails. □

For randomized verification with perfect completeness and false-accept probability at most epsilon for each single-successor history,

E[Q] = sum_j Pr[query j] >= (1-epsilon)m.

With two-sided error at most epsilon, the bound is E[Q]>=(1-2epsilon)m. This is a cell-probe bound for a black-box unordered membership-only archive, not an unconditional bound for every commitment or SNARK. A latest dictionary, maintained witness/index, key-ordered archive, linear-work archive-scan SNARK or trusted online oracle falls outside its premises.

Corollary H-WES-N2: three costs cannot all disappear

Secure recovery must pay for at least one of authenticated latest state, ongoing witness/index maintenance, or linear suffix data/prover work. H-WES chooses fixed-size consensus roots, external latest/history bodies, and on-demand O(log K)+O(log N) proofs without continual owner witness updates.

4. Expirable Authenticated Object Model (EAOM)

Lifecycle attaches to incarnation (k,v):

Live(k,v) -> Expired(k,v) -> Recovered(k,v; successor=v+1)
                           \-> Consumed(k,v)

Recovered and Consumed are terminal for that incarnation. Recovery atomically closes (k,v) and creates Live(k,v+1). Global authenticated state is

S_t = (R_live,t, R_latest,t, R_null,t, R_hist,t, R_DA,t),
R_state = H_ds("HYPHEN_STATE_V1",
               R_live || R_latest || R_null || R_hist || R_DA).

The roots commit live bodies, per-key latest status, permanent nullifiers, append-only lifecycle history and recovery-data availability.

Expire(k,v,t) requires Head_t(k)=(v,Live) and t>=lease_end(k,v). It removes the live entry, appends an Expired record and changes the latest head; the atomic block-state commit is the linearization point.

Recovery authorizes

AuthRec = H_ds("HYPHEN_WES_AUTHORIZATION_V1",
  Recover || cid || k || v || H(expired_record) || t || new_lease || R_state,t).

It proves history inclusion, current (k,v,Expired) status and matching archive index, owner authorization, new_lease>t>=old_lease_end, and successor body availability. Success creates Live(k,v+1) and appends

E_rec=(cid,class,k,v,Recover,t,H(expired),H(live_successor)).

History append, successor insertion and latest update commit or roll back together. Consume uses a canonical lease sentinel, appends E_con=(cid,class,k,v,Consume,t,H(expired),0^256) and leaves no outgoing transition from Consumed.

5. Security games

Latest-only recoverability lets an attacker see all old bodies, membership proofs, authorizations and DA proofs, then wins only if recovery accepts while Head_t(k)!=(v,Expired). Security reduces to history binding, latest-dictionary soundness and deterministic cross-field consistency.

No-resurrection first commits (k,v) as Consumed. Replay can succeed only by forging the current latest proof or commitment binding. A recovered incarnation also cannot recover again because the head is already Live(k,v+1) or later.

History accountability reduces a forged terminal event inclusion to MMR/hash binding. Recovery events bind the successor hash. This proves what closed and what successor was committed, not permanent availability of history bodies.

These results require collision resistance; MMR and latest-dictionary binding/soundness; owner-policy EUF-CMA or knowledge soundness; atomic, deterministic transition; and one canonical pre-state root. Recoverability also requires a provider retaining the body and fresh proof. A commitment cannot recover data deleted by everyone.

6. Time, storage and witness tradeoffs

Construction Validator digest External index Owner updates Recovery proof/data Prover work
Append archive + scan O(1) O(N) body O(1) Theta(m) Theta(m)
Dynamic client accumulator O(1) O(N) related/global updates scheme-specific scheme-specific
Versioned authenticated dictionary O(1) root O(K) O(1) on-demand proof O(log K) O(log K)
Archive-scan SNARK O(1) O(N) body O(1) constant/polylog Omega(m) scan
H-WES EAOM reference five roots O(K+N) O(1) O(log K+log N)+body O(log K+log N) query

The last row is not storage-free. It moves bodies, tree nodes and proof serving to replaceable providers while validators authenticate roots. The persistent SMT is 256 levels, namespace-separated and stores only non-default nodes. Updates are fixed-depth for 256-bit keys, or Theta(lambda) when key length is the parameter. Long-term space and write-amplification benchmarks remain open.

6.1 SMT opening binding

Define domain-separated leaf, empty and node hashes:

L_ns(k,v)=H_ds("HYPHEN_SMT_LEAF_V1",ns||k||v),
E_ns=H_ds("HYPHEN_SMT_EMPTY_LEAF_V1",ns),
N_ns,d(l,r)=H_ds("HYPHEN_SMT_NODE_V1",ns||d||l||r).

For fixed (ns,k,R), two accepted openings to different values, including absence, imply either a leaf-domain collision or the first parent-level node collision along the fixed path. Thus collision resistance gives same-root opening binding. □

The implementation computes an in-memory overlay, then one sled transaction checks that the old root is unchanged and writes all nodes plus the new root. A failed transaction writes nothing; a CAS conflict returns ConcurrentUpdate; success flushes before returning. Only the complete old or new state is persistently observable.

6.2 Chunk proofs and the DA certificate boundary

Object and chunk hashes bind object, position, count and length:

O=H_ds("HYPHEN_PROOF_BLOB_V1",len(B)||B),
C_i=H_ds("HYPHEN_PROOF_BLOB_CHUNK_V1",O||i||m||len(B_i)||B_i).

Merkle binding and final reassembly verification prevent cross-object, cross-position and truncated-chunk replay absent a hash collision or second preimage. An availability certificate binds chain, epoch, committee, retention and object metadata and requires q=2f+1 distinct signatures. With at most f malicious seats, at least f+1 honest signers fully held the blob when signing. This proves retrievable-at-signing-time, not continued retention. The certificate cannot recover a body later deleted by every signer.

7. Covered object classes

The model can represent leased UTXOs with permanent nullifiers, recoverable account authority, revocable credential incarnations, DA objects requiring a target-root certificate, and contract KV with an explicit code authorization relation. Shielded notes remain out of profile until a zero-knowledge circuit binds hidden owner/input/nullifier relations to EAOM transitions.

8. Implementation mapping and open obligations

expiring_state.rs implements canonical 159-byte records, a deterministic expiry queue, history MMR and latest proofs, action/height/lease/pre-state-bound authorization, atomic successor and terminal receipts, no-resurrection tests, and a roots-plus-witness pure verifier API. Supporting modules implement a persistent namespaced SMT with atomic root CAS, bounded content-addressed blob storage with chunk proofs, bounded typed P2P proof/chunk transport, and full-blob-before-signing 2f+1 availability certificates.

The research profile now persists all five EAOM roots and applies public creation plus bounded expiry in the same atomic block/reorg transition as the current block root. Real owner policy, provider challenge/repair, shielded recovery/consume relations, benchmarks and independent audits remain incomplete. The mechanism must not be activated or called production-ready.

9. Prior-art boundary

Comparison is required against Utreexo/UTXO accumulators, stateless-client witness updates, Ethereum state expiry, versioned authenticated dictionaries, revocable-credential accumulators and historical KV authentication. Starting references are Utreexo (IACR ePrint 2019/611), Ethereum statelessness/state expiry, and Practical Revocable Anonymous Credentials (FC 2012). Until systematic literature/patent searches and independent peer review, H-WES may claim only this model, black-box lower bound and candidate tradeoff, not first resolution of state expiry or first recoverable state.


H-WES:从恢复下界到可过期认证对象

状态:研究规范,未激活到任何 Hyphen 网络。本文固定问题、模型、定理和实现 接口;它不是新颖性证明,也不是密码学审计报告。

1. 先修正一个过强的不可能性陈述

仅有以下四条不足以推出不可能性:

  1. 旧版本不可恢复;
  2. 客户端最初只保存历史 witness;
  3. 验证者不持久维护 key -> latest 认证索引;
  4. 恢复协议为无交互或常数轮。

反例是单轮发送完整 archive。恢复者从不可信 archive provider 下载全部记录, 把记录和认证数据一次发送给验证者;验证者重算 append-only commitment,线性扫描 同一 key 的所有版本,再判断候选是否最新。它不保存 latest index,协议只有一轮, 而且可以拒绝旧版本。代价是 Theta(N) 通信、prover I/O 和 verifier 工作。

因此可证明的命题必须明确加入 H-WES 真正需要的目标:succinct recovery。 下文使用黑盒认证 archive 模型,并要求恢复只打开次线性数量的 archive cell。

2. 状态与 archive 模型

令安全参数为 lambda,key 空间为 K,值空间为 V。对象 incarnation 为

I = (cid, class, k, v, h_val, P_owner, h_create, h_lease),

其中 cid 是链身份,k in K,v in N 是版本,h_val=H(value), P_owner 是授权策略 commitment。对固定 key,版本严格递增。

在高度 t,历史为有序记录数组

Hist_t = (r_0, ..., r_(N_t-1)),
A_t = CommitArchive(Hist_t).

archive 只暴露:

Open(A_t, i) -> (r_i, pi_i),
VerifyOpen(A_t, i, r_i, pi_i) -> {0,1},
VerifyAppend(A_s, A_t, pi_(s,t)) -> {0,1},  s <= t.

它不原生暴露按 key 查询、范围内不存在性或最大版本。MMR 是该接口的一个实例: 它认证位置和 append-only 历史,但不赋予叶内容按 key 的顺序。

定义当前头函数:

Head_t(k) = arg max_v { r in Hist_t : r.key=k and r.version=v },

若同一版本可出现多个记录,历史无效;共识转换必须在 append 前拒绝该情况。

2.1 “等价 latest 信息”不是特定数据结构

认证 latest oracle 是任意 (D_t, ProveLatest, VerifyLatest),满足:

VerifyLatest(D_t, k, v, status, pi)=1
    iff Head_t(k).(version,status)=(v,status),

其中错误概率可忽略。D_t 可以是 SMT root、Verkle root、vector commitment、 SNARK statement 或其他摘要;所以“没有 BTreeMap”不等于“没有 latest 等价信息”。 只要一个短证明能认证 freshness,它在安全意义上就是 latest dictionary。

3. 两个下界定理

定理 H-WES-N0:安全恢复蕴含可恢复 head 的认证能力

设恢复协议 Pi=(ProveRec,VerifyRec) 满足:

  • 完备性:当前处于 Expired 的最新 incarnation 可产生接受 transcript;
  • latest-only 健全性:任何旧 incarnation 或非 Expired incarnation 的接受概率 至多为 negl(lambda);
  • verifier 的公开状态为 S_t。

则 Pi 可黑盒构造认证谓词

IsRecoverableHead_t(k,v)=1 iff Head_t(k).(version,status)=(v,Expired).

构造。 令公开摘要 D_t=S_t。证明算法运行 ProveRec,但不提交最终状态; 验证算法运行 VerifyRec 的只读阶段。对最新 Expired 记录,恢复完备性给出谓词 完备性;对旧版本、Live、Recovered 或 Consumed,latest-only 健全性给出 拒绝。故恢复 transcript 本身就是上述谓词的认证证明。□

这个定理不构造可查询任意状态的完整 latest dictionary;它只认证恢复所需的 (v,Expired) head 谓词。即便如此,“安全恢复但不存在任何与 latest 等价的恢复 认证信息”仍在语义上矛盾。该信息不一定由 validator 保存完整索引,但必须以某种 commitment 和证明关系存在。

定理 H-WES-N1:membership-only archive 的后缀查询下界

考虑 archive-generic verifier:A_t 是只能作为 Open/VerifyOpen 参数使用的 opaque oracle handle;verifier 不能读取、哈希、比较或根据 handle 的表示分支, 只能通过成功打开 cell 获知历史内容。这个限制排除了把额外 latest 信息偷偷编码 进 commitment 表示的构造。候选记录位于 位置 i,其后有 m=N_t-i-1 个 cell。若 verifier 不使用 latest oracle,且对 所有合法历史零错误判断候选是否为 Head_t(k),则最坏情况必须认证打开全部 m 个后缀 cell。

证明。 假设某次接受执行只打开后缀位置集合 Q,且 |Q|<m。取未打开位置 j notin Q。构造两个合法历史:

H0: 所有 j>i 的记录 key 均不等于 k;
H1: 与 H0 相同,但位置 j 是 key=k、version>v 的合法记录。

在该 oracle 接口中,以相同随机带运行 verifier 时,两个世界对已打开位置返回 完全相同的 (index,record);handle 本身不可观察,未打开位置也不能影响执行。 候选在 H0 是 latest,在 H1 不是 latest。 verifier 对两者输出相同,因此完备性或健全性至少一个失败,矛盾。故最坏情况 |Q|=m。□

对随机 verifier,先要求在无后继历史 H0 上具有完美完备性,并对每个只在位置 j 放置唯一新版本的历史 H1_j 具有至多 epsilon 的漏检概率。若某次执行未查询 j,它在 H0 与 H1_j 的 transcript 相同;完美完备性迫使它接受,因此该事件 计入 H1_j 的漏检。故对每个 j 都有 Pr[query j]>=1-epsilon。对 indicator 求和:

E[Q] = sum_j Pr[query j] >= (1-epsilon)m.

因此任何最坏查询上限 q 也满足 q>=E[Q]>=(1-epsilon)m。若允许双边错误,即 H0 的拒绝概率和每个 H1_j 的接受概率都至多 epsilon,相同 coupling 只给出 Pr[query j]>=1-2epsilon,从而 E[Q]>=(1-2epsilon)m(该式在 epsilon<1/2 时有意义)。这是黑盒、无序、membership-only archive 的 cell-probe 下界,不是对所有 commitment 或所有 SNARK 的无条件密码学下界。 以下方法都避开了定理前提:

  • 维护按 key 的认证 latest dictionary;
  • 让客户端或 witness service 持续更新 freshness witness;
  • 给 archive 增加按 key 排序的认证索引;
  • 用 SNARK 证明完整后缀扫描,保持 verifier 简短但 prover 仍做 Omega(m) 工作;
  • 信任在线 freshness oracle。

推论 H-WES-N2:三项成本不能同时消失

在上述模型中,安全恢复至少承担下列一项:

authenticated latest state;
ongoing witness/index maintenance;
linear suffix data/prover work.

H-WES 的目标不是违反该下界,而是选择一个清晰的新权衡:validator 共识状态只 承诺 fixed-size roots;latest map 和历史 body 可由分离存储层维护;恢复者按需 取得 O(log K)+O(log N) 新鲜证明;长期离线 owner 不持续更新 witness。

4. Expirable Authenticated Object Model(EAOM)

4.1 incarnation 与 key head

生命周期属于 incarnation (k,v),不是模糊的“账户”。状态为:

Live(k,v) -> Expired(k,v) -> Recovered(k,v; successor=v+1)
                           \-> Consumed(k,v)

Recovered(k,v) 和 Consumed(k,v) 都是终态。恢复会在同一个原子转换中关闭 (k,v) 并创建 Live(k,v+1);因此不会出现可被其他交易观察到的中间状态。

全局认证状态为:

S_t = (R_live,t, R_latest,t, R_null,t, R_hist,t, R_DA,t).
  • R_live:当前 live body hash;
  • R_latest:每个 key 的 (version,status,archive_index,value_hash,head_hash);
  • R_null:不可删除的消费/nullifier 集;
  • R_hist:append-only 生命周期历史 MMR;
  • R_DA:恢复后数据可用性证书根。

组合根采用域分离:

R_state = H_ds("HYPHEN_STATE_V1",
               R_live || R_latest || R_null || R_hist || R_DA).

4.2 转换与线性化点

Expire(k,v,t) 的前置条件:

Head_t(k)=(v,Live), t>=lease_end(k,v).

它删除 live entry,向历史追加 Expired record,并把 latest head 改成 (v,Expired,index,...)。区块原子状态提交是线性化点。

恢复授权 statement 为:

AuthRec = H_ds("HYPHEN_WES_AUTHORIZATION_V1",
  Recover || cid || k || v || H(expired_record) || t || new_lease || R_state,t).

Recover(k,v,t,new_lease) 验证:

  1. expired record 在 R_hist,t 中;
  2. R_latest,t 证明 (k,v,Expired) 是当前 head,并绑定同一 archive index;
  3. owner policy 对 AuthRec 授权;
  4. new_lease>t>=old_lease_end;
  5. DA policy 证明 successor body 对 R_DA,t 可用。

成功后构造 Live(k,v+1),向历史追加:

E_rec=(cid,class,k,v,Recover,t,H(expired),H(live_successor)),

并原子更新 live/latest。历史 append、successor 插入和 latest 更新共同构成唯一 线性化点;任一步失败必须整体回滚。

消费授权将 Recover 换成 Consume,new_lease 使用规范哨兵值。成功后追加

E_con=(cid,class,k,v,Consume,t,H(expired),0^256)

并把 latest head 设为 (v,Consumed,...)。没有从 Consumed(k,v) 出发的转换。

5. 安全游戏

5.1 Latest-only recoverability

挑战者运行任意合法历史得到 S_t,允许攻击者取得所有历史 body、membership proof、旧授权和旧 DA proof。攻击者输出 (k,v,pi_rec)。若验证接受但

Head_t(k) != (v,Expired),

则攻击者获胜。H-WES 要求胜率为 negl(lambda)。

归约分三种:伪造 history membership 破坏 MMR binding;伪造 latest membership 破坏 authenticated dictionary soundness;让 witness 中版本/状态/index/hash 不一致则被确定性检查拒绝。

5.2 No resurrection

挑战者先让 (k,v) 线性化到 Consumed,随后攻击者可重放任意旧材料并输出恢复 请求。由于 current latest status 为 Consumed 而恢复关系要求 Expired,接受 需要伪造 latest proof 或破坏其 commitment binding。因此成功概率可忽略。

Recovered(k,v) 也不能再次恢复:current head 已是 Live(k,v+1) 或其后继状态。

5.3 History accountability

给定 current R_hist,攻击者若输出未真实追加的终态事件和有效 inclusion proof, 即构成 MMR binding/底层哈希抗碰撞攻击。恢复事件还承诺 successor hash,所以 审计者可检查“哪个 incarnation 被关闭、何时关闭、是否产生后继”。这不证明 历史 body 永久可用;availability 是独立假设。

5.4 可组合结论及边界

在以下假设同时成立时,EAOM 满足前三个游戏:

H collision resistant;
MMR and latest dictionary binding/sound;
owner policy EUF-CMA or knowledge sound;
state transition atomic and deterministic;
consensus exposes one canonical pre-state root.

可恢复性另外需要至少一个 provider 保留 body 和新鲜证明,并满足 DA policy。 commitment 不能从信息论上恢复已被所有人删除的 body。

6. 时间、存储与 witness 权衡

令 N 为历史记录数,K 为不同 key 数,m 为候选后的历史长度。

构造 validator 共识摘要 外部索引存储 owner 持续更新 恢复 proof/数据 prover 工作
仅 append archive + 扫描 O(1) O(N) body O(1) Theta(m) Theta(m)
客户端动态 accumulator O(1) O(N) 每次相关/全局更新 方案相关 方案相关
版本化认证字典 O(1) root O(K) O(1),按需取新 proof O(log K) O(log K)
archive-scan SNARK O(1) O(N) body O(1) 常数或 polylog Omega(m) scan
H-WES EAOM reference 5 个固定根 O(K+N) O(1) O(log K+log N)+body O(log K+log N) 查询

最后一行不是“无存储”:它把 body、latest tree nodes 和 proof serving 移到可替换 provider;validator 仍必须认证 roots,数据可用性和 provider 去中心化仍需协议。 Rust reference 现有一个 256 层、命名空间隔离、仅保存非默认节点的持久 SMT。 单 key 更新访问 256 层,因此在固定 256-bit key 模型中是固定深度工作;若把 key 长度记为 lambda,则更新/证明为 Theta(lambda)。这关闭了“只有 BTreeMap 重建树”的实现缺口,但尚未把 EAOM 的五个根接入当前区块原子状态转换,也没有 长期空间/写放大 benchmark。

6.1 SMT opening binding

定义:

L_ns(k,v)=H_ds("HYPHEN_SMT_LEAF_V1",ns||k||v),
E_ns=H_ds("HYPHEN_SMT_EMPTY_LEAF_V1",ns),
N_ns,d(l,r)=H_ds("HYPHEN_SMT_NODE_V1",ns||d||l||r).

非成员证明把叶值取为 E_ns;成员证明取 L_ns(k,v),两者都按 k 的 256 个 bit 和 256 个 sibling 唯一重算根。

定理 SMT-S1(同根 opening binding)。 假设上述域分离哈希抗碰撞。对固定 (ns,k,R),PPT 攻击者不能以不可忽略概率给出两个均通过的 proof,分别打开为 v != v'(包括一个为 absent)。

证明。 两条验证路径的初始叶哈希不同;否则立即得到 leaf/empty domain 的 碰撞。两条路径最终根相同。沿路径自叶向根取第一个“当前哈希不同、计算出的父哈希 相同”的层。固定 key 使左右位置相同,proof 给出的 sibling 与当前哈希形成两个 不同的有序 child pair,却在相同 namespace、depth、node domain 下产生相同父哈希, 构成 N_ns,d 的碰撞,矛盾。□

原子性。 实现先读取期望根 R,在内存中计算全部 leaf/node overlay,再在一个 sled transaction 内检查当前根仍为 R 并写入全部 overlay 与 R'。事务失败不提交 任何 key;CAS 冲突返回 ConcurrentUpdate;事务成功后 API 在返回前 flush。因此 可观察的持久线性化结果只有旧状态或完整新状态,不存在只更新部分祖先的根。

6.2 Chunk proof 与 DA 证书边界

blob 的对象哈希和 chunk leaf 分别为:

O=H_ds("HYPHEN_PROOF_BLOB_V1",len(B)||B),
C_i=H_ds("HYPHEN_PROOF_BLOB_CHUNK_V1",O||i||m||len(B_i)||B_i).

Merkle proof 对 C_i 的 binding 归约到 Merkle node/leaf hash 抗碰撞;由于 O,i,m,len 都进入 leaf,不能把另一个对象、位置或截断 chunk 重放到该 opening。重组器按 i=0..m-1 取齐后再次验证 O,所以通过的完整重组等于被 metadata 承诺的 blob, 除非找到上述哈希碰撞/二次原像。

availability certificate 包含链、epoch、committee、retention、对象 metadata,并要求 q=2f+1 个不同 seat 签名。诚实 signer 在签名前从本地 store 完整读取并核对 blob。 因此在 n=3f+1、恶意 seat 至多 f 时,证书至少包含 q-f=f+1 个签名时确实持有 blob 的诚实 seat。该结论仅是 retrievable-at-signing-time;证书本身不能证明 retention 窗口内仍有副本。若所有 signer 之后删除数据,commitment 不可能恢复 body。

7. 可覆盖的对象类别

模型只要求稳定 key、单调 version、owner policy 和可认证 body hash,因此可实例化:

  • 租赁 UTXO:Consumed 同时插入永久 nullifier;
  • 可恢复账户权限:owner policy 可为阈值/延迟恢复策略;
  • 可撤销凭证:credential incarnation 过期后恢复或终止,revocation 仍需公开语义;
  • 数据可用对象:恢复前必须给出目标 DA root 的证书;
  • 合约 KV:必须额外定义合约代码对恢复/消费的授权关系。

shielded note 暂不在实现 profile:若 owner、真实输入和 nullifier 关系需要隐藏, 必须有零知识电路把 EAOM transition 与现有隐私交易关系绑定。

8. 实现对应与未关闭义务

crates/hyphen-state/src/expiring_state.rs 当前实现:

  • 159-byte record 规范编码;
  • deterministic expiry queue;
  • history MMR 与 latest membership proof;
  • 绑定 action/height/lease/pre-state 的 authorization digest;
  • 原子语义的 recovery successor 与 terminal history receipt;
  • consumed terminal 和负向 no-resurrection 测试;
  • 纯 verifier API,只需 roots 与 witness。

配套模块现在还实现:

  • persistent_smt.rs:持久命名空间 SMT、成员/非成员 proof、原子 root CAS 与 flush;
  • proof_store.rs:有大小界的 content-addressed blob、chunk Merkle proof 和重组校验;
  • P2P sync:typed SMT proof、blob metadata/chunk 请求与响应上限;
  • availability.rs:完整 blob 验证后签名、2f+1 distinct-seat certificate、上下文与 retention 检查。

研究 profile 已将 expiring_state.rs 的五根、公开创建和有界过期与当前 block state root 放入同一个原子 block/reorg transition。真实 owner policy、provider challenge/repair、shielded recovery/consume relation、benchmark 和独立审计仍未完成, 所以不能激活到生产网络或称为生产协议。

9. Prior-art 边界

必须逐项比较 Utreexo/UTXO accumulators、stateless-client witness 更新、Ethereum state expiry、版本化 authenticated dictionaries、可撤销凭证 accumulators 和 历史 KV authentication。当前可核验入口包括:

在系统检索、专利检索和独立同行评审完成前,H-WES 只能声称“给出了上述模型、 黑盒下界和一个候选权衡”,不能声称首次解决状态过期或首次实现可恢复状态。

Clone this wiki locally