| Internet-Draft | Detached Evidence Delivery | November 2026 |
| Jones | Expires 22 May 2027 | [Page] |
This document describes a detached evidence handoff between independently administered organizations. It distinguishes artifact delivery, content integrity, source authentication, registration, reconciliation, and recipient decisions. It describes how a recipient can return a signed comparison against stated benchmarks, with an optional commitment to the figures it compared, without exporting its source ledger. Worked examples illustrate both a disclosed reconciliation and a comparison attestation. The document defines terminology and an architectural model; it specifies no wire protocol.¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 22 May 2027.¶
Copyright (c) 2026 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Organizations sometimes need to evaluate externally produced evidence without granting standing access to their systems. This document describes an optional delivery pattern for that boundary. Persistent integration remains appropriate when a workflow needs continuous synchronization or shared execution. The two patterns can coexist.¶
A producer prepares a versioned packet. A recipient verifies artifact integrity and, where applicable, signature and signer authority. The recipient then determines whether the claims apply to its workflow, checks local evidence, and may return a signed finding. Delivery, acknowledgement, reconciliation and approval can occur independently.¶
A candidate return record can identify the request and artifact digest, scope, period, the input evidence classifications, unresolved exceptions and signer role. The institution can retain its underlying ledger. A separate policy statement can reference the reconciliation finding when it is useful to verify the decision separately. These are design considerations rather than interoperable field definitions.¶
This section describes one way a recipient can answer a benchmark comparison without disclosing the figures it compared. It is an illustrative pattern, not a wire format.¶
The producer's packet states a benchmark value B_c, a non-negative integer in the minor unit of a stated currency, for each category c in {integration, middleware, decommission}, and a digest of the packet. The recipient derives, from records under its control, a corresponding non-negative integer R_c for each category. Corresponding amounts use the same currency, period and scope, and the three categories do not overlap. If a comparable amount cannot be derived, the recipient reports NA with the applicable reason as described in Section 4.2. For each category with comparable amounts the recipient reports exactly one of three outcomes: "GT" if R_c is greater than B_c, "EQ" if they are equal, and "LT" if R_c is less than B_c. A fourth member reports the same three-valued comparison of the sums of the R_c and of the B_c. The total outcome is not derivable from the three category outcomes and is reported separately. Integer comparison is exact; no rounding or floating-point arithmetic is involved. The packet identifies the source and version of each benchmark, for example by a citation and a digest of the benchmark table, so that the recipient and a later reviewer use the same values. A benchmark stating the producer's own route cost and an independent reference for the incumbent route are different things and are best kept apart. A two-valued result is not sufficient, because equality is possible and a category the recipient does not report needs its own explicit outcome, "NA" (Section 4.2).¶
A category outcome is "NA" when the recipient reports no comparison for that category. The record gives a reason for each NA, taken from a closed set: "UNAVAILABLE" means the evidence needed to derive R_c was not available to the recipient; "NOT_APPLICABLE" means the recipient states that the category does not apply to the deliverable; "NOT_ASSESSED" means the recipient did not evaluate the category. The reasons are not interchangeable. An NA is never treated as zero and never converted to GT, EQ, or LT. An absent or unopened commitment does not by itself make a category NA.¶
The total outcome is "GT", "EQ", or "LT" only if none of the three category outcomes is "NA". If any category outcome is "NA", the total outcome is "NA" and the record names the categories responsible. A total is not computed over the remaining categories, because that would treat the missing figure as zero and could make the comparison look more favorable than the evidence supports. This three-category pattern retains all three categories. An unevaluated category is reported as NA (NOT_ASSESSED); it is not silently omitted. A pattern with a different category set requires a separately identified revision.¶
Core reasons. The reason of an NA outcome is exactly one of UNAVAILABLE, NOT_APPLICABLE or NOT_ASSESSED. This set is closed for this pattern. Adding a core reason, or changing the aggregation rule above, would be a new and explicitly identified revision of the pattern.¶
Detail codes. An NA outcome can carry one optional member, "detail", whose value is a namespaced string of the form "<namespace>:<code>". A detail refines the core reason. It does not change the meaning of the core reason or the aggregation rule. A verifier that does not recognize a detail leaves the treatment of the core reason unchanged.¶
Unknown core reason. If the reason of an NA outcome is not one of the three core reasons, the verifier preserves the value as received and reports it as UNSUPPORTED_REASON. It does not map the value to a known reason, treat the category as zero, or report a favourable result. The category still counts as NA for aggregation, so the total is NA.¶
Asserted reason and assessment. The reason is what the issuer asserted. It is not evidence that the assertion is true. A signed NOT_APPLICABLE does not demonstrate that the category does not apply. A recipient's assessment of an asserted reason, if there is one, is a separate finding and does not alter the outcome or the total. Every NA, including NOT_APPLICABLE, makes the total NA.¶
If operational need arises, a registry of reasons could be defined, with registration and change control in the manner of [RFC8126]. This document defines none.¶
The recipient can include a commitment to the figures it compared. A bare SHA-256 digest of a small set of integers can be recovered by guessing, so the commitment is computed over a fixed domain label, a random salt of at least 128 bits (the minimum recommended by [RFC9901]) that the recipient retains, and the canonical serialization (for example [RFC8785]) of the figures. The salt has one fixed length for a given label (16 octets for the label in the example), so that the concatenation of salt and figures has a single parse. A fresh secret random salt makes guessing attacks against the commitment harder. Binding relies on the security of SHA-256 and an unambiguous encoding of the inputs. The comparison outcomes can still disclose information about the figures, especially across repeated comparisons. A matching opening does not establish signing time. No zero-knowledge proof is involved. Per-claim salted-hash disclosure is already specified in [RFC9901] for JSON Web Tokens, and a recipient that wants per-field disclosure can use that mechanism instead.¶
The commitment does not show that the recipient computed R_c from its records, or that the records are accurate. A matching opening identifies the figures bound by the signed commitment; it does not independently establish when the signature was created. If the recipient later opens the commitment to an auditor, regulator, or the producer, anyone can recompute the commitment and the outcomes from the opened figures and check that both agree with the signed record.¶
The recipient signs a serialization binding the record type and format version, issuer and case; the digest of the request and of the producer's packet; the benchmark source, version and table digest; the currency, period and scope; the intended recipient; the three category outcomes and each NA reason and optional detail; the total outcome; the commitment when present; the claimed issuance time; and the signer's role. A verifier checks the expected context and separately evaluates signer authority. A claimed issuance time is not independent time evidence. Binding the request and packet digests prevents a signed outcome from being presented against a different packet. The signed record could be conveyed as a Statement and registered with a Transparency Service under [RFC9943]; registration shows that the signer issued it and shows nothing about the figures. The recipient's policy decision is not part of this record.¶
If the return record and a later policy decision are both registered, they would use different subject values and the decision would carry the digest of the return record. An Issuer's later Signed Statement with the same issuer and subject can supersede an earlier one (Section 9.2 of [RFC9943]), which is not wanted here. A Transparency Service is required only to authenticate Signed Statements and to enforce its own Registration Policy (Section 5.1.1.1 of [RFC9943]). It can treat the statement as opaque (Section 3 of [RFC9943]), although it can also validate the payload to enforce a domain-specific policy (Section 6.3 of [RFC9943]). Checks on the comparison values are therefore a matter for a relying party. Registration can expose issuer, subject and registration-time metadata to parties with access, even when the content is not readable, so a recipient can consider a service's privacy posture before registering (Section 8 of [RFC9943]). Registration order need not match issuance order, so the signed record carries its own time.¶
A relying party that evaluates a return record reports the findings below separately. These are evaluation results, distinct from the recipient's reconciliation finding defined in Core Terminology. It reports no combined "verified" or "pass" label, because the findings answer different questions.¶
A policy decision record identifies the digest of the return record it considered, the categories reported as NA, the policy identifier and version, the decision-maker, the scope of what is allowed, any conditions, and a validity period. If the applicable policy requires complete evidence, the record states the exception that permits the decision and its basis. The decision does not change any finding above, and nothing in the record implies that the decision follows from the comparison.¶
The following synthetic example reports one return record whose commitment was not opened and whose middleware category was reported as NA, followed by a separate policy decision.¶
report: DEMO-003 (synthetic) attestation signature: PASS (assumed for illustration; not checked) signer authority: NOT_ESTABLISHED commitment: NOT_PROVIDED figures: NOT_ESTABLISHED integration: GT middleware: NA (UNAVAILABLE) decommission: LT total: NA (middleware) policy record signature: PASS (assumed for illustration; not checked) decision-maker authority: ESTABLISHED (assumed for illustration; not checked) policy decision: ALLOW, pilot scope P1, with conditions policy record binds: return record digest; NA categories; policy id and version; decision-maker; scope; conditions; validity period; basis for the incomplete-evidence exception decision changes findings: no¶
In this synthetic example the attestation signature result (PASS), policy record signature result (PASS) and decision-maker authority result (ESTABLISHED) are assumed for illustration. They were not assessed. The ALLOW stated in the policy record is not evidence of any of them.¶
The following synthetic example is not a wire format and the figures are invented. The producer's benchmarks are 100000, 50000, and 20000 USD cents for integration, middleware, and decommission. The recipient's figures are 130000, 50000, and 15000. The outcomes are therefore GT, EQ, and LT, and the totals, 195000 and 170000, give GT. Because no category is NA, the total is defined. The commitment is the SHA-256 digest of the label, a zero octet, the salt, and the canonical serialization of the figures. The salt below is published only so that the example can be reproduced; a real salt is secret and is never reused.¶
comparison: DEMO-002
request and packet digests: omitted from this example
integration: GT
middleware: EQ
decommission: LT
total: GT
commitment-label: DEH-COMMIT-1
commitment:
36d516c50e28dabc4cfc12298fd75046b6dc10dae2a9bc873d1c3a4cb3494ef2
signer-role: example (synthetic)
policy decision: not included
figures (kept by the recipient, not returned):
{"decommission":"15000","integration":"130000","middleware":"50000"}
salt (hex, example only):
000102030405060708090a0b0c0d0e0f
¶
Streams. An evaluation-report stream is identified by the tuple (issuer, case, scope, "finding"). A policy stream is identified by the tuple (issuer, case, scope, "decision"). The case identifies the assessment. Each signed record binds the four members of its stream tuple and its record type. A reconciliation finding remains a distinct input assertion, not an evaluation report. The subject value of a policy record differs from the subject values of the return record and evaluation report it considers. A type label alone does not define a stream.¶
Policy record members. In addition to the policy information described in Section 4.5, a policy record carries "supersedes", "attestationDigest" and "findingsDigest". The first is the digest of the prior policy content object in the same stream, or null for a root. The second is the digest of the return-record content object carrying the comparison attestation considered. The third is the digest of the exact evaluation-report content object considered, or null if no evaluation report was considered. A reconciliation finding considered separately is identified by its own type and digest; it is not silently substituted for an evaluation report. Each referenced object is retained with its type, issuer and assessment context. A reference does not establish its target's accuracy, provenance or authority.¶
Record identity. For this pattern, the record digest is SHA-256 over the UTF-8 RFC 8785 canonical serialization of an immutable content object, expressed as 64 lowercase hexadecimal characters. The content object binds its record type, format version, issuer, case, scope and all decision or evidence-reference fields. It excludes its own derived digest, signature bytes, unprotected receipts and transport metadata. A verifier recomputes the digest from the content; it does not trust a supplied digest alone. Any authenticated envelope context, including issuer and subject, is checked for agreement with the expected content and stream. An implementation profile specifies the exact content schema, envelope representation, algorithms and context mapping. This architectural pattern does not define that wire profile.¶
Observation scope. A verifier evaluates only the policy records presented to it for a stream. A single rooted, non-branching chain in that presented set establishes consistency of that set. It does not establish that the issuer has never issued, or has withheld, another successor. Without a separate basis for completeness and freshness, the verifier does not claim global uniqueness, and does not claim that the tip of the presented chain is the latest applicable policy.¶
Duplicate delivery. Before evaluating topology, the verifier groups identical canonical content by recomputed record digest within the expected stream. If one digest is associated with different canonical bytes, the verifier reports invalid input and does not merge the contents. Each delivery envelope is evaluated separately. A content object is admitted as an authenticated graph node if at least one delivery of that same content passes both signature and stream-binding verification for the expected stream; the two checks cannot be combined from different deliveries. Failed or unperformed delivery checks remain separately reported and do not erase a valid delivery. Repeated delivery contributes at most one node, creates no fork by itself, and does not make topology depend on delivery order. Deduplication establishes neither decision-maker authority, completeness nor freshness.¶
Chain rules. The distinct presented records of one policy stream that pass signature and stream-binding verification form a consistent chain if and only if all of the following hold: (a) exactly one record has "supersedes" null, the root; (b) the "supersedes" of every other record equals the digest of exactly one presented record of the stream; (c) no record is named by the "supersedes" of more than one record; (d) following "supersedes" from the record that no other record names reaches the root and visits every presented record exactly once. If the chain is consistent, its tip is the one presented record that no other presented record names in "supersedes".¶
Reporting within that scope. Chain findings are computed only over admitted authenticated nodes of the expected stream. More than one distinct root, or two distinct nodes naming the same predecessor, results in CONFLICT. A missing admitted predecessor, cycle or disconnected chain results in UNRESOLVED. If both conditions hold, CONFLICT takes precedence; unresolved details can still be reported separately. Empty admitted input results in UNRESOLVED. A failed signature or stream-binding check is reported for that delivery and does not itself admit a node. A cross-stream predecessor reference is invalid and its node is excluded before topology evaluation. A consistent result is reported as CONSISTENT_PRESENTED_CHAIN, with the tip digest, not as current or approved. The verifier does not select a record by timestamp.¶
Separation. A policy record never supersedes a return record, evaluation report or reconciliation finding. A later evaluation report never rebinds an earlier policy record: that policy record continues to name the exact content digests it considered. Earlier content and delivery-check records are preserved. Supersession changes the tip of the presented policy chain; it does not alter historical evidence or prove which policy is globally latest.¶
Separate prerequisites. Signature verification, stream binding and decision-maker authority are checked separately, and each is reported separately. Chain topology alone does not establish an authorized policy decision. A record stating ALLOW is distinguishable from each of these results.¶
Exceptions. An exception is valid only if the applicable policy permits it and the decision-maker who issued it is authorized. A record carrying an exception does not make an exception valid.¶
A startup supplies a comparison baseline. An institution reconciles 800000 USD cents for its current route and models 500000 USD cents for an unused route. With a zero daily valuation term, subtraction yields 300000 cents and the current-to-modeled ratio is 1.60. The difference compares a recorded input with a modeled input; it is not realized savings. A later reviewer can recompute the result if given the inputs but cannot independently verify the private ledger without access to the evidence or an appropriate assurance mechanism.¶
This synthetic example extends the observed-cost baseline with an explicitly modeled input. A strictly observed comparison would leave the unused route unverified and report no observed difference or ratio.¶
The following is an illustrative response, not a wire format or an actual signed statement. An institution could sign a serialization of the response that also binds the exact request and baseline digests.¶
comparison: DEMO-001 recipient: Example Institution (synthetic) scope: same deliverable; illustrative period P1 current route: 800000 USD cents; ledger-reconciled unused route: 500000 USD cents; modeled daily valuation: 0 USD cents/day modeled difference: 300000 USD cents current-to-modeled ratio: 1.60 realized savings: not established policy decision: not included¶
A signed packet statement or a separately signed evaluation could be registered under the architecture in [RFC9943]. A receipt provides evidence of registration under the service trust configuration. It does not establish ledger accuracy, institutional acceptance, or a policy decision. This document does not define a SCITT profile.¶
An unauthenticated digest is a comparison value, not source authentication. A signature authenticates a statement only under a trusted association between key, identity and role. It does not establish the accuracy of the statement. A packet identifier alone supplies neither authorization nor freshness.¶
A relying party can check the request identifier, scope, period and intended recipient to guard against reuse in a different context. Freshness, revocation, supersession and duplicate handling depend on the recipient policy. Partial disclosure limits what the relying party can verify. Keep sensitive ledger data out of public logs and minimize identifiers that enable cross-context correlation. A comparison attestation binds a commitment to the signed statement; it does not independently establish signing time or the accuracy of the underlying figures. A relying party decides what weight that deserves. A presented chain of policy records establishes consistency of the presented records only (Section 4.7).¶
This document requests no IANA actions.¶
The following 21 terms map the Entercation application vocabulary to the architectural concepts above. Product-specific terms are examples and impose no requirements on other implementations.¶
The following materials provide author-supplied project context. They are informative, are not prerequisites for this pattern, and do not validate this draft or its implementation. Individual Internet-Drafts are work in progress and do not imply IETF endorsement or working-group adoption. Each external artifact retains its own license; an MIT license on software does not replace the IETF Trust terms applying to this Internet-Draft.¶
Entercation Info Inc.. Outerprise specification surface. Version: v0.1.2. Record license: mit-license. [ZENODO23873552] Publication does not establish implementation correctness.¶
Jones, Vince. Outerprise Data Center Transparency: Synthetic Demonstration and Evidence Boundaries. Version: 0.1.0. Record license: cc-by-4.0. [ZENODO23993510] Synthetic and self-assessment material is not an independent examination.¶
Jones, Vince. Outerprise Institutional Genesis Protocol (OIGP) v1.3.1: Cryptographic Fiscal Sponsorship State Machine. Version: 1.3.1-hardened. Record license: mit-license. [ZENODO24177899] Synthetic and self-assessment material is not an independent examination.¶
Jones, Vince. Outerprise Waste Metric: Branded Forensic Report and Publication Guide. Version: no version label in record metadata. Record license: cc-by-4.0. [ZENODO24182132] Publication does not establish implementation correctness.¶
Jones, Vince. Outerprise Institutional Genesis Protocol (OIGP) v1.3.4: Public Documentation and Synthetic Case Study. Version: 1.3.4-hardened-documentation. Record license: mit-license. [ZENODO24197013] Synthetic and self-assessment material is not an independent examination.¶
Vince Jones. A Dual-Receipt Profile for Binding a Decision Statement to an Independent Policy Evaluation Statement. draft-jones-scitt-dual-receipt-00. Work in progress. [DUALRECEIPT]¶
Vince Jones. Outerprise Ecosystem and Infrastructure: Visual Manual and Machine-Readable Companion. draft-jones-outerprise-visual-manual-00. Work in progress. [VISUALMODEL]¶
OIGP v1.3.1-hardened, v1.3.4-hardened, the separately named v1.3.4 ID Bundle, the v1.3.6 reference bundle and the dual-receipt Internet-Draft revision -00 are distinct artifacts. The records cited above do not identify or validate every one of those artifacts. Their record-identity rules are not interchangeable with this draft. No historical test count is carried forward as evidence that this revision or a deployed verifier passes.¶
Thanks to Tokachi Kamimura for comments on the reporting terminology.¶