@zanii/train 0.1.0

package/LICENSE

@@ -0,0 +1,159 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or Derivative
+          Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or agreed
+      to in writing, Licensor provides the Work (and each Contributor
+      provides its Contributions) on an "AS IS" BASIS, WITHOUT WARRANTIES
+      OR CONDITIONS OF ANY KIND, either express or implied, including,
+      without limitation, any warranties or conditions of TITLE,
+      NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A PARTICULAR PURPOSE.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright 2026 Zanii
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -0,0 +1,45 @@
+# @zanii/train
+
+**Training-data provenance & subject-absence proofs.** The claim buyers actually want
+is negative: *"my data was never used to train your model."* You can't prove that about
+the whole world — but you can prove it about a **declared run**: commit the training set
+to a sorted-Merkle manifest, then hand a data subject a **non-membership proof** their
+`ref` is absent from it, checkable offline with no trust in the operator.
+
+```sh
+npm install @zanii/train @zanii/core
+```
+
+```ts
+import { buildManifest, manifestReceipt, proveAbsence, verifyAbsence } from '@zanii/train';
+
+// the model operator commits the training set (refs are hashes/HMACs, never content):
+const manifest = buildManifest({ runId: 'gpt-run-7', refs: memberRefs, createdAt: now });
+await zanii.record(manifestReceipt(manifest));   // anchor it into the ledger
+
+// a data subject asks "was my document in it?" — the operator answers with a proof:
+const proof = proveAbsence(manifest, memberRefs, userRef);   // throws if actually present
+verifyAbsence(proof, manifest);   // TRUE, verified by the subject, offline
+```
+
+Python: `from zanii.train import build_manifest, prove_absence, verify_absence, ...`
+— **byte-identical** manifest roots (locked by a cross-language vector).
+
+## How the absence proof works
+
+Sorted-set non-membership — the pattern the Zanii L1's SMT uses. Sort the member refs
+and Merkle-commit the adjacency leaves (each member, and the gap to the next). To prove
+`x` is absent, reveal the one gap `(lo, hi)` with `lo < x < hi` plus its inclusion
+proof in the manifest: since the set is sorted and every gap is committed, a covering
+gap that verifies means nothing sits between `lo` and `hi` — so `x` is not a member.
+Membership (`proveMembership`) is the same machinery on a member leaf. Proofs use
+`@zanii/core`'s own `inclusionProof`/`verifyInclusion`, so they verify with the exact
+primitive the ledger uses.
+
+## The limit, stated up front
+
+This proves what was **recorded in the manifest**, not what a GPU secretly saw. An
+absence proof is only as honest as the manifest is complete — pair it with an attested
+enclave (`@zanii/attest`: the published image hash is the only code that touched the
+data) so "absent from the manifest" also means "absent from the run." The manifest
+commits refs, never content.

package/dist/index.d.ts

@@ -0,0 +1,59 @@
+/** One committed gap between two adjacent sorted refs. A member leaf has `lo === hi`. */
+interface Pair {
+    lo: string;
+    hi: string;
+}
+export interface TrainingManifest {
+    v: 1;
+    type: 'training.manifest';
+    /** The model/run this set trained. */
+    run_id: string;
+    /** How many distinct member refs. */
+    size: number;
+    /** Merkle root over the sorted adjacency leaves — `sha256:<hex>`. */
+    root: string;
+    created_at: string;
+}
+/** Commit a training set (member refs) to a manifest. Refs are hashes/HMACs, never content. */
+export declare function buildManifest(fields: {
+    runId: string;
+    refs: string[];
+    createdAt: string;
+}): TrainingManifest;
+export declare function manifestHash(m: TrainingManifest): string;
+/** The payload to `record()` as `training.run` — anchors the manifest into the ledger. */
+export declare function manifestReceipt(m: TrainingManifest): {
+    target: string;
+    payload: Record<string, unknown>;
+};
+export interface AbsenceProof {
+    v: 1;
+    type: 'training.absence';
+    run_id: string;
+    /** The ref proven absent. */
+    ref: string;
+    /** The covering gap `(lo, hi)` with `lo < ref < hi`. */
+    gap: Pair;
+    index: number;
+    size: number;
+    path: string[];
+}
+export interface MembershipProof {
+    v: 1;
+    type: 'training.membership';
+    run_id: string;
+    ref: string;
+    index: number;
+    size: number;
+    path: string[];
+}
+/** Build a proof that `ref` is ABSENT from the training set. Throws if it is present. */
+export declare function proveAbsence(m: TrainingManifest, refs: string[], ref: string): AbsenceProof;
+/** Build a proof that `ref` is PRESENT in the training set. Throws if absent. */
+export declare function proveMembership(m: TrainingManifest, refs: string[], ref: string): MembershipProof;
+/** Verify a subject-absence proof against a manifest — offline, no trust in the operator. */
+export declare function verifyAbsence(proof: AbsenceProof | undefined, m: TrainingManifest): boolean;
+/** Verify a membership proof against a manifest. */
+export declare function verifyMembership(proof: MembershipProof | undefined, m: TrainingManifest): boolean;
+export {};
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AA4BA,yFAAyF;AACzF,UAAU,IAAI;IACZ,EAAE,EAAE,MAAM,CAAC;IACX,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,MAAM,WAAW,gBAAgB;IAC/B,CAAC,EAAE,CAAC,CAAC;IACL,IAAI,EAAE,mBAAmB,CAAC;IAC1B,sCAAsC;IACtC,MAAM,EAAE,MAAM,CAAC;IACf,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,qEAAqE;IACrE,IAAI,EAAE,MAAM,CAAC;IACb,UAAU,EAAE,MAAM,CAAC;CACpB;AAmBD,+FAA+F;AAC/F,wBAAgB,aAAa,CAAC,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,EAAE,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,GAAG,gBAAgB,CAa5G;AAED,wBAAgB,YAAY,CAAC,CAAC,EAAE,gBAAgB,GAAG,MAAM,CAExD;AAED,0FAA0F;AAC1F,wBAAgB,eAAe,CAAC,CAAC,EAAE,gBAAgB,GAAG;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,CAEzG;AAQD,MAAM,WAAW,YAAY;IAC3B,CAAC,EAAE,CAAC,CAAC;IACL,IAAI,EAAE,kBAAkB,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,6BAA6B;IAC7B,GAAG,EAAE,MAAM,CAAC;IACZ,wDAAwD;IACxD,GAAG,EAAE,IAAI,CAAC;IACV,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;CAChB;AAED,MAAM,WAAW,eAAe;IAC9B,CAAC,EAAE,CAAC,CAAC;IACL,IAAI,EAAE,qBAAqB,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,MAAM,EAAE,CAAC;CAChB;AAED,yFAAyF;AACzF,wBAAgB,YAAY,CAAC,CAAC,EAAE,gBAAgB,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,MAAM,GAAG,YAAY,CAQ3F;AAED,iFAAiF;AACjF,wBAAgB,eAAe,CAAC,CAAC,EAAE,gBAAgB,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,MAAM,GAAG,eAAe,CAMjG;AAED,6FAA6F;AAC7F,wBAAgB,aAAa,CAAC,KAAK,EAAE,YAAY,GAAG,SAAS,EAAE,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAO3F;AAED,oDAAoD;AACpD,wBAAgB,gBAAgB,CAAC,KAAK,EAAE,eAAe,GAAG,SAAS,EAAE,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAKjG"}

package/dist/index.js

@@ -0,0 +1,113 @@
+/**
+ * @zanii/train — training-data provenance & subject-absence proofs.
+ *
+ * The claim buyers actually want is negative: *"my data was never used to train your
+ * model."* You cannot prove that about the whole world, but you CAN prove it about a
+ * declared training run: commit the training set to a **sorted-Merkle manifest**, then
+ * hand a data subject a **non-membership proof** that their `ref` is absent from it —
+ * checkable offline, with no trust in the model operator.
+ *
+ * How the absence proof works (sorted-set non-membership, the pattern the Zanii L1's
+ * SMT uses): sort the member refs, Merkle-commit the *adjacent pairs*. To prove `x`
+ * is absent, reveal the one pair `(lo, hi)` with `lo < x < hi` and its inclusion proof
+ * in the manifest — since the set is sorted and every gap is committed, a covering gap
+ * that verifies means nothing sits between `lo` and `hi`, so `x` is not a member.
+ * Membership is the same machinery: reveal the leaf whose `lo == x`.
+ *
+ * ## The limit, stated up front
+ *
+ * This proves what was **recorded** in the manifest, not what a GPU secretly saw. An
+ * absence proof is only as honest as the manifest is complete — pair it with an
+ * attested enclave (`@zanii/attest`: the published image hash is the only code that
+ * touched the data) so "absent from the manifest" also means "absent from the run".
+ * The manifest commits refs (`hmac:v1:…` / `sha256:…`), never content.
+ */
+import { bytesToHex, hexToBytes } from '@noble/hashes/utils';
+import { sha256 } from '@noble/hashes/sha256';
+import { canonicalBytes, inclusionProof, jcsHash, leafHash, merkleRoot, verifyInclusion } from '@zanii/core';
+/** Sorted, de-duplicated member refs → the adjacency leaves that get Merkle-committed. */
+function pairsOf(refs) {
+    const sorted = [...new Set(refs)].sort();
+    if (sorted.length === 0)
+        return [];
+    const pairs = [];
+    // sentinel-bounded gaps so the smallest/largest members are covered too
+    pairs.push({ lo: '', hi: sorted[0] });
+    for (let i = 0; i < sorted.length; i++) {
+        pairs.push({ lo: sorted[i], hi: sorted[i] }); // the member itself (membership leaf)
+        const next = i + 1 < sorted.length ? sorted[i + 1] : '￿';
+        pairs.push({ lo: sorted[i], hi: next }); // the gap to the next member
+    }
+    return pairs;
+}
+const pairLeaf = (p) => leafHash(canonicalBytes(p));
+/** Commit a training set (member refs) to a manifest. Refs are hashes/HMACs, never content. */
+export function buildManifest(fields) {
+    if (!fields.runId)
+        throw new Error('runId is required');
+    if (Number.isNaN(Date.parse(fields.createdAt)))
+        throw new Error('createdAt must be an ISO timestamp');
+    const pairs = pairsOf(fields.refs);
+    const root = pairs.length ? bytesToHex(merkleRoot(pairs.map(pairLeaf))) : bytesToHex(sha256(new Uint8Array(0)));
+    return {
+        v: 1,
+        type: 'training.manifest',
+        run_id: fields.runId,
+        size: new Set(fields.refs).size,
+        root: `sha256:${root}`,
+        created_at: fields.createdAt,
+    };
+}
+export function manifestHash(m) {
+    return jcsHash(m);
+}
+/** The payload to `record()` as `training.run` — anchors the manifest into the ledger. */
+export function manifestReceipt(m) {
+    return { target: 'training.run', payload: { manifest: m, manifest_hash: manifestHash(m) } };
+}
+// Proof paths use core's own inclusionProof/verifyInclusion — same RFC 6962
+// construction as the ledger, so a proof here verifies with the same primitive.
+const inclusionPath = (index, leaves) => inclusionProof(index, leaves).map(bytesToHex);
+const verifyPath = (leaf, index, size, path, root) => verifyInclusion(leaf, index, size, path.map(hexToBytes), root);
+/** Build a proof that `ref` is ABSENT from the training set. Throws if it is present. */
+export function proveAbsence(m, refs, ref) {
+    const sorted = [...new Set(refs)].sort();
+    if (sorted.includes(ref))
+        throw new Error(`ref is PRESENT — cannot prove absence`);
+    const pairs = pairsOf(refs);
+    const leaves = pairs.map(pairLeaf);
+    const gi = pairs.findIndex((p) => p.lo < ref && ref < p.hi && !(p.lo === p.hi));
+    if (gi < 0)
+        throw new Error('no covering gap — manifest malformed');
+    return { v: 1, type: 'training.absence', run_id: m.run_id, ref, gap: pairs[gi], index: gi, size: leaves.length, path: inclusionPath(gi, leaves) };
+}
+/** Build a proof that `ref` is PRESENT in the training set. Throws if absent. */
+export function proveMembership(m, refs, ref) {
+    const pairs = pairsOf(refs);
+    const leaves = pairs.map(pairLeaf);
+    const li = pairs.findIndex((p) => p.lo === ref && p.hi === ref);
+    if (li < 0)
+        throw new Error('ref is ABSENT — cannot prove membership');
+    return { v: 1, type: 'training.membership', run_id: m.run_id, ref, index: li, size: leaves.length, path: inclusionPath(li, leaves) };
+}
+/** Verify a subject-absence proof against a manifest — offline, no trust in the operator. */
+export function verifyAbsence(proof, m) {
+    const p = proof ?? {};
+    if (p.v !== 1 || p.type !== 'training.absence' || p.run_id !== m.run_id)
+        return false;
+    if (!(p.gap.lo < p.ref && p.ref < p.gap.hi))
+        return false; // the gap must actually cover the ref
+    if (p.gap.lo === p.gap.hi)
+        return false; // a member leaf is not a gap
+    const root = hexToBytes(m.root.replace('sha256:', ''));
+    return verifyPath(pairLeaf(p.gap), p.index, p.size, p.path, root);
+}
+/** Verify a membership proof against a manifest. */
+export function verifyMembership(proof, m) {
+    const p = proof ?? {};
+    if (p.v !== 1 || p.type !== 'training.membership' || p.run_id !== m.run_id)
+        return false;
+    const root = hexToBytes(m.root.replace('sha256:', ''));
+    return verifyPath(pairLeaf({ lo: p.ref, hi: p.ref }), p.index, p.size, p.path, root);
+}
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AACH,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AAC7D,OAAO,EAAE,MAAM,EAAE,MAAM,sBAAsB,CAAC;AAC9C,OAAO,EAAE,cAAc,EAAE,cAAc,EAAE,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAoB7G,0FAA0F;AAC1F,SAAS,OAAO,CAAC,IAAc;IAC7B,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACzC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACnC,MAAM,KAAK,GAAW,EAAE,CAAC;IACzB,wEAAwE;IACxE,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC,CAAC;IACvC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACvC,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC,CAAE,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC,CAAE,EAAE,CAAC,CAAC,CAAC,sCAAsC;QACtF,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC,GAAG,CAAC;QAC1D,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC,CAAE,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,6BAA6B;IACzE,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,MAAM,QAAQ,GAAG,CAAC,CAAO,EAAc,EAAE,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAuC,CAAC,CAAC,CAAC;AAE5G,+FAA+F;AAC/F,MAAM,UAAU,aAAa,CAAC,MAA4D;IACxF,IAAI,CAAC,MAAM,CAAC,KAAK;QAAE,MAAM,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC;IACxD,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,oCAAoC,CAAC,CAAC;IACtG,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;IACnC,MAAM,IAAI,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,UAAU,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,IAAI,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAChH,OAAO;QACL,CAAC,EAAE,CAAC;QACJ,IAAI,EAAE,mBAAmB;QACzB,MAAM,EAAE,MAAM,CAAC,KAAK;QACpB,IAAI,EAAE,IAAI,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI;QAC/B,IAAI,EAAE,UAAU,IAAI,EAAE;QACtB,UAAU,EAAE,MAAM,CAAC,SAAS;KAC7B,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,CAAmB;IAC9C,OAAO,OAAO,CAAC,CAAuC,CAAC,CAAC;AAC1D,CAAC;AAED,0FAA0F;AAC1F,MAAM,UAAU,eAAe,CAAC,CAAmB;IACjD,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,OAAO,EAAE,EAAE,QAAQ,EAAE,CAAC,EAAE,aAAa,EAAE,YAAY,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;AAC9F,CAAC;AAED,4EAA4E;AAC5E,gFAAgF;AAChF,MAAM,aAAa,GAAG,CAAC,KAAa,EAAE,MAAoB,EAAY,EAAE,CAAC,cAAc,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC;AACvH,MAAM,UAAU,GAAG,CAAC,IAAgB,EAAE,KAAa,EAAE,IAAY,EAAE,IAAc,EAAE,IAAgB,EAAW,EAAE,CAC9G,eAAe,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,EAAE,IAAI,CAAC,CAAC;AAyBjE,yFAAyF;AACzF,MAAM,UAAU,YAAY,CAAC,CAAmB,EAAE,IAAc,EAAE,GAAW;IAC3E,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACzC,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,uCAAuC,CAAC,CAAC;IACnF,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5B,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnC,MAAM,EAAE,GAAG,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,GAAG,GAAG,IAAI,GAAG,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAChF,IAAI,EAAE,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IACpE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,IAAI,EAAE,kBAAkB,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,GAAG,EAAE,GAAG,EAAE,KAAK,CAAC,EAAE,CAAE,EAAE,KAAK,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE,aAAa,CAAC,EAAE,EAAE,MAAM,CAAC,EAAE,CAAC;AACrJ,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,eAAe,CAAC,CAAmB,EAAE,IAAc,EAAE,GAAW;IAC9E,MAAM,KAAK,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC5B,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnC,MAAM,EAAE,GAAG,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,GAAG,IAAI,CAAC,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;IAChE,IAAI,EAAE,GAAG,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;IACvE,OAAO,EAAE,CAAC,EAAE,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,IAAI,EAAE,aAAa,CAAC,EAAE,EAAE,MAAM,CAAC,EAAE,CAAC;AACvI,CAAC;AAED,6FAA6F;AAC7F,MAAM,UAAU,aAAa,CAAC,KAA+B,EAAE,CAAmB;IAChF,MAAM,CAAC,GAAG,KAAK,IAAK,EAAmB,CAAC;IACxC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,KAAK,kBAAkB,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IACtF,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QAAE,OAAO,KAAK,CAAC,CAAC,sCAAsC;IACjG,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,KAAK,CAAC,CAAC,GAAG,CAAC,EAAE;QAAE,OAAO,KAAK,CAAC,CAAC,6BAA6B;IACtE,MAAM,IAAI,GAAG,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;AACpE,CAAC;AAED,oDAAoD;AACpD,MAAM,UAAU,gBAAgB,CAAC,KAAkC,EAAE,CAAmB;IACtF,MAAM,CAAC,GAAG,KAAK,IAAK,EAAsB,CAAC;IAC3C,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,KAAK,qBAAqB,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IACzF,MAAM,IAAI,GAAG,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,UAAU,CAAC,QAAQ,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;AACvF,CAAC"}

package/package.json

@@ -0,0 +1,42 @@
+{
+  "name": "@zanii/train",
+  "version": "0.1.0",
+  "description": "Training-data provenance: hash a training set into a sorted-Merkle manifest, then PROVE a named subject's data was absent from a real training run - a non-membership proof anyone can check offline. 'It was never used to train a model' becomes a yes, not a promise. Proves what was recorded went in; cannot prove an off-manifest copy didn't (state the enclave boundary).",
+  "license": "Apache-2.0",
+  "homepage": "https://ledger.zanii.agency",
+  "type": "module",
+  "main": "./dist/index.js",
+  "types": "./dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "import": "./dist/index.js"
+    }
+  },
+  "files": [
+    "dist",
+    "src",
+    "README.md",
+    "LICENSE"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "engines": {
+    "node": ">=18"
+  },
+  "devDependencies": {
+    "@types/node": "^26.1.0",
+    "typescript": "^5.8.2",
+    "vitest": "^3.0.9"
+  },
+  "dependencies": {
+    "@noble/hashes": "^1.7.1",
+    "@zanii/core": "0.4.0"
+  },
+  "scripts": {
+    "build": "tsc -p tsconfig.build.json",
+    "test": "vitest run",
+    "typecheck": "tsc --noEmit"
+  }
+}

package/src/index.ts

@@ -0,0 +1,154 @@
+/**
+ * @zanii/train — training-data provenance & subject-absence proofs.
+ *
+ * The claim buyers actually want is negative: *"my data was never used to train your
+ * model."* You cannot prove that about the whole world, but you CAN prove it about a
+ * declared training run: commit the training set to a **sorted-Merkle manifest**, then
+ * hand a data subject a **non-membership proof** that their `ref` is absent from it —
+ * checkable offline, with no trust in the model operator.
+ *
+ * How the absence proof works (sorted-set non-membership, the pattern the Zanii L1's
+ * SMT uses): sort the member refs, Merkle-commit the *adjacent pairs*. To prove `x`
+ * is absent, reveal the one pair `(lo, hi)` with `lo < x < hi` and its inclusion proof
+ * in the manifest — since the set is sorted and every gap is committed, a covering gap
+ * that verifies means nothing sits between `lo` and `hi`, so `x` is not a member.
+ * Membership is the same machinery: reveal the leaf whose `lo == x`.
+ *
+ * ## The limit, stated up front
+ *
+ * This proves what was **recorded** in the manifest, not what a GPU secretly saw. An
+ * absence proof is only as honest as the manifest is complete — pair it with an
+ * attested enclave (`@zanii/attest`: the published image hash is the only code that
+ * touched the data) so "absent from the manifest" also means "absent from the run".
+ * The manifest commits refs (`hmac:v1:…` / `sha256:…`), never content.
+ */
+import { bytesToHex, hexToBytes } from '@noble/hashes/utils';
+import { sha256 } from '@noble/hashes/sha256';
+import { canonicalBytes, inclusionProof, jcsHash, leafHash, merkleRoot, verifyInclusion } from '@zanii/core';
+
+/** One committed gap between two adjacent sorted refs. A member leaf has `lo === hi`. */
+interface Pair {
+  lo: string;
+  hi: string;
+}
+
+export interface TrainingManifest {
+  v: 1;
+  type: 'training.manifest';
+  /** The model/run this set trained. */
+  run_id: string;
+  /** How many distinct member refs. */
+  size: number;
+  /** Merkle root over the sorted adjacency leaves — `sha256:<hex>`. */
+  root: string;
+  created_at: string;
+}
+
+/** Sorted, de-duplicated member refs → the adjacency leaves that get Merkle-committed. */
+function pairsOf(refs: string[]): Pair[] {
+  const sorted = [...new Set(refs)].sort();
+  if (sorted.length === 0) return [];
+  const pairs: Pair[] = [];
+  // sentinel-bounded gaps so the smallest/largest members are covered too
+  pairs.push({ lo: '', hi: sorted[0]! });
+  for (let i = 0; i < sorted.length; i++) {
+    pairs.push({ lo: sorted[i]!, hi: sorted[i]! }); // the member itself (membership leaf)
+    const next = i + 1 < sorted.length ? sorted[i + 1]! : '￿';
+    pairs.push({ lo: sorted[i]!, hi: next }); // the gap to the next member
+  }
+  return pairs;
+}
+
+const pairLeaf = (p: Pair): Uint8Array => leafHash(canonicalBytes(p as unknown as Record<string, unknown>));
+
+/** Commit a training set (member refs) to a manifest. Refs are hashes/HMACs, never content. */
+export function buildManifest(fields: { runId: string; refs: string[]; createdAt: string }): TrainingManifest {
+  if (!fields.runId) throw new Error('runId is required');
+  if (Number.isNaN(Date.parse(fields.createdAt))) throw new Error('createdAt must be an ISO timestamp');
+  const pairs = pairsOf(fields.refs);
+  const root = pairs.length ? bytesToHex(merkleRoot(pairs.map(pairLeaf))) : bytesToHex(sha256(new Uint8Array(0)));
+  return {
+    v: 1,
+    type: 'training.manifest',
+    run_id: fields.runId,
+    size: new Set(fields.refs).size,
+    root: `sha256:${root}`,
+    created_at: fields.createdAt,
+  };
+}
+
+export function manifestHash(m: TrainingManifest): string {
+  return jcsHash(m as unknown as Record<string, unknown>);
+}
+
+/** The payload to `record()` as `training.run` — anchors the manifest into the ledger. */
+export function manifestReceipt(m: TrainingManifest): { target: string; payload: Record<string, unknown> } {
+  return { target: 'training.run', payload: { manifest: m, manifest_hash: manifestHash(m) } };
+}
+
+// Proof paths use core's own inclusionProof/verifyInclusion — same RFC 6962
+// construction as the ledger, so a proof here verifies with the same primitive.
+const inclusionPath = (index: number, leaves: Uint8Array[]): string[] => inclusionProof(index, leaves).map(bytesToHex);
+const verifyPath = (leaf: Uint8Array, index: number, size: number, path: string[], root: Uint8Array): boolean =>
+  verifyInclusion(leaf, index, size, path.map(hexToBytes), root);
+
+export interface AbsenceProof {
+  v: 1;
+  type: 'training.absence';
+  run_id: string;
+  /** The ref proven absent. */
+  ref: string;
+  /** The covering gap `(lo, hi)` with `lo < ref < hi`. */
+  gap: Pair;
+  index: number;
+  size: number;
+  path: string[];
+}
+
+export interface MembershipProof {
+  v: 1;
+  type: 'training.membership';
+  run_id: string;
+  ref: string;
+  index: number;
+  size: number;
+  path: string[];
+}
+
+/** Build a proof that `ref` is ABSENT from the training set. Throws if it is present. */
+export function proveAbsence(m: TrainingManifest, refs: string[], ref: string): AbsenceProof {
+  const sorted = [...new Set(refs)].sort();
+  if (sorted.includes(ref)) throw new Error(`ref is PRESENT — cannot prove absence`);
+  const pairs = pairsOf(refs);
+  const leaves = pairs.map(pairLeaf);
+  const gi = pairs.findIndex((p) => p.lo < ref && ref < p.hi && !(p.lo === p.hi));
+  if (gi < 0) throw new Error('no covering gap — manifest malformed');
+  return { v: 1, type: 'training.absence', run_id: m.run_id, ref, gap: pairs[gi]!, index: gi, size: leaves.length, path: inclusionPath(gi, leaves) };
+}
+
+/** Build a proof that `ref` is PRESENT in the training set. Throws if absent. */
+export function proveMembership(m: TrainingManifest, refs: string[], ref: string): MembershipProof {
+  const pairs = pairsOf(refs);
+  const leaves = pairs.map(pairLeaf);
+  const li = pairs.findIndex((p) => p.lo === ref && p.hi === ref);
+  if (li < 0) throw new Error('ref is ABSENT — cannot prove membership');
+  return { v: 1, type: 'training.membership', run_id: m.run_id, ref, index: li, size: leaves.length, path: inclusionPath(li, leaves) };
+}
+
+/** Verify a subject-absence proof against a manifest — offline, no trust in the operator. */
+export function verifyAbsence(proof: AbsenceProof | undefined, m: TrainingManifest): boolean {
+  const p = proof ?? ({} as AbsenceProof);
+  if (p.v !== 1 || p.type !== 'training.absence' || p.run_id !== m.run_id) return false;
+  if (!(p.gap.lo < p.ref && p.ref < p.gap.hi)) return false; // the gap must actually cover the ref
+  if (p.gap.lo === p.gap.hi) return false; // a member leaf is not a gap
+  const root = hexToBytes(m.root.replace('sha256:', ''));
+  return verifyPath(pairLeaf(p.gap), p.index, p.size, p.path, root);
+}
+
+/** Verify a membership proof against a manifest. */
+export function verifyMembership(proof: MembershipProof | undefined, m: TrainingManifest): boolean {
+  const p = proof ?? ({} as MembershipProof);
+  if (p.v !== 1 || p.type !== 'training.membership' || p.run_id !== m.run_id) return false;
+  const root = hexToBytes(m.root.replace('sha256:', ''));
+  return verifyPath(pairLeaf({ lo: p.ref, hi: p.ref }), p.index, p.size, p.path, root);
+}