Internet-Draft Astrognosy AI
Intended status: Informational J. Westerbeck
Expires: 30 September 2026 Astrognosy AI
30 March 2026
reason:// — A URI Scheme and Registry Protocol for
Validated Agent Reasoning Artifacts
draft-westerbeck-reason-protocol-00
Abstract
This document defines the "reason://" URI scheme and associated
registry protocol for naming, storing, and retrieving compressed
structural reasoning artifacts across distributed autonomous agent
networks. An artifact enters the registry only by winning a live
competitive arbitration round scored by the Web Agent Reasoning
Federation (WARF) protocol. The artifact schema contains no raw
data fields; the structural representation is mathematically non-
invertible (empirical reconstruction rate r = 0.0149). The
protocol is designed to function as open infrastructure — analogous
to DNS for network addresses or HTTP for document retrieval — for
the agentic internet.
Status of This Memo
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 30 September 2026.
Copyright Notice
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.
Table of Contents
1. Introduction
1.1. The Problem
1.2. What reason:// Provides
1.3. Relationship to WARF and Underlying Stack
1.4. Design Principles
2. Terminology
3. URI Format and Namespace Structure
3.1. URI Scheme
3.2. Segment Definitions
3.3. Canonical Examples
3.4. Namespace Uniqueness and URI Stability
4. Artifact Schema
4.1. Core Schema
4.2. Field Specifications
4.3. What the Schema Does Not Contain
5. Registry Entry — The Arbitration Gate
5.1. The Gate
5.2. Arbitration Process
5.3. Score Immutability and Replacement
5.4. Self-Initiated Arbitration
6. Resolution Mechanics
6.1. Resolution Request
6.2. Resolution Response
6.3. Failure Cases
7. Privacy Guarantees
7.1. Structural Non-Invertibility
7.2. Empirical Bound
7.3. Architectural Guarantee
8. Governance
8.1. Bootstrap Phase (now — Q4 2026)
8.2. Transition Phase (Q4 2026)
8.3. Community Phase (2027+)
9. Security Considerations
10. IANA Considerations
11. References
11.1. Normative References
11.2. Informative References
Appendix A. Non-Invertibility Demonstration
Authors' Addresses
1. Introduction
1.1. The Problem
Every autonomous agent operating in a regulated or competitive
environment is a knowledge silo. An agent at a healthcare
institution learns a structural pattern that predicts patient
intervention needs. An agent at a financial institution discovers
a fraud signature. A manufacturing agent identifies a bearing
failure precursor. In each case, the insight cannot travel.
Data sharing is prohibited (HIPAA, GDPR, competitive secrecy).
Model sharing is prohibited (proprietary, legally encumbered).
Weight sharing is prohibited (weights encode the training data).
This is the defining structural constraint of the agentic internet:
learned intelligence is trapped inside the systems that produced it.
No protocol exists for naming and retrieving validated reasoning
artifacts across organizational and institutional boundaries.
1.2. What reason:// Provides
reason:// is open infrastructure for the transfer of reasoning
without the transfer of data.
The protocol introduces three primitives:
1. A naming standard — structured URIs for named reasoning
capabilities (e.g., reason://medicine/records/longitudinal-
maintenance-prediction)
2. A quality gate — only artifacts that have won a competitive
WARF arbitration round may enter the registry; the registry
is a hall of validated winners, not a submission box
3. A resolution protocol — any agent resolves a named capability
URI to receive the highest-scoring artifact currently registered
under that name
1.3. Relationship to WARF and Underlying Stack
reason:// is the topmost layer in a four-layer stack:
reason:// -- this specification: naming, gating, resolution
Xport Node -- reference WARF node implementation: arbitration
and artifact store (see Section 2)
WARF -- open arbitration protocol: cargo packages, rules
PCF -- convergence scoring mechanism: the math
reason:// does not specify a scoring mechanism (PCF's domain).
It does not specify arbitration rules (WARF's domain). It
specifies how the outputs of WARF arbitration are named, stored,
and made retrievable.
The reference implementation of a WARF node (the Xport Node) is
maintained by Astrognosy AI. The PCF scoring mechanism is
patent-protected [PCF-PATENT]. The reason:// protocol
specification itself is CC BY 4.0.
1.4. Design Principles
- No raw data in the protocol. The artifact schema has no field
for raw records, model weights, or training data.
- Identity-neutral arbitration. Artifact quality is determined
by convergence score, not by the identity, reputation, or size
of the submitting agent.
- Auditable. Every arbitration event produces a SHA-256 chained
audit record. Every deposit is publicly traceable.
- Pure infrastructure at v1. No tokens, no fees, no economic
layer at launch. The protocol spreads on technical merit.
2. Terminology
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY",
and "OPTIONAL" in this document are to be interpreted as described
in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in
all capitals, as shown here.
reason:// URI
A URI conforming to the scheme defined in Section 3.
Artifact
A compressed structural representation of a learned reasoning
pattern, conforming to the schema defined in Section 4.
Registry
The authoritative store of admitted artifacts, indexed by
reason:// URI.
Arbitration Round
A competitive evaluation of one or more agent submissions
conducted by a WARF node, resulting in a winner that is
admitted to the registry.
Xport Node
A WARF node implementation. Astrognosy AI operates the
reference node (Node 0).
PCF Score (S)
The convergence score produced by the PCF mechanism for a
given submission, normalized to [0,1] by share of total
arbitration mass. Higher is stronger.
Structural Centroid
The mean vector of an agent's learned structural embeddings,
stored as the artifact's pattern field. Non-invertible by
mathematical construction.
3. URI Format and Namespace Structure
3.1. URI Scheme
The reason:// URI scheme has the following syntax (ABNF):
reason-URI = "reason://" domain "/" category "/" task
domain = 1*( ALPHA / DIGIT / "-" )
category = 1*( ALPHA / DIGIT / "-" )
task = 1*( ALPHA / DIGIT / "-" )
All three path segments are REQUIRED. Query parameters and
fragment identifiers are NOT RECOMMENDED and SHOULD be ignored
by conformant implementations. All segments MUST be lowercase
ASCII. Hyphens are the standard word separator.
3.2. Segment Definitions
domain -- Top-level knowledge area.
Examples: medicine, finance, cybersecurity,
manufacturing, pharma, defense
category -- Sub-area within the domain.
Examples: records, fraud, network, bearing,
molecular, sensor
task -- Specific named capability.
Examples: longitudinal-maintenance-prediction,
anomaly-detection, port-scan-classification
3.3. Canonical Examples
reason://medicine/records/longitudinal-maintenance-prediction
reason://medicine/imaging/lesion-boundary-detection
reason://finance/fraud/anomaly-detection
reason://finance/fraud/synthetic-identity-ring-detection
reason://cybersecurity/network/port-scan-classification
reason://cybersecurity/network/intrusion-detection
reason://manufacturing/bearing/fault-signature
reason://pharma/molecular/interaction-pattern
3.4. Namespace Uniqueness and URI Stability
Each full URI path (domain/category/task) is unique in the
registry. Resolution always returns the artifact with the
highest current score registered under that URI. The URI
addresses a capability, not a specific artifact instance.
Once a URI has received at least one resolution request, it is
considered stable. Existing URIs MUST NOT be invalidated or
repurposed. Deprecated URIs are marked but remain resolvable.
4. Artifact Schema
4.1. Core Schema
Each artifact admitted to the registry MUST conform to the
following JSON schema:
{
"uri": "reason://domain/category/task",
"pattern": [<list of floats — structural centroid>],
"thresholds": {
"high_confidence": <float>,
"moderate_confidence": <float>,
"minimum_signal": <float>
},
"score": <float>,
"provenance": {
"agent_id": "<string>",
"deposited_at": "<ISO 8601 datetime>",
"arbitration_event_id": "<string>",
"audit_hash": "sha256:<hex>"
},
"metadata": {
"evidence_count": <integer>,
"domain": "<string>",
"category": "<string>",
"task": "<string>",
"version": <integer>
}
}
4.2. Field Specifications
uri (string, REQUIRED)
The canonical reason:// URI. Immutable after deposit.
pattern (array of floats, REQUIRED)
The structural centroid of the agent's learned embeddings.
This is the shape of the insight, not the data. MUST NOT
contain raw records, identifiers, or invertible encodings.
See Section 7.
thresholds (object, REQUIRED)
Three calibrated confidence tiers derived from the agent's
evidence base. Consuming agents use these to interpret
similarity scores.
score (float, REQUIRED)
The PCF convergence score earned in the admitting arbitration
round. Immutable after deposit. Range: [0,1].
provenance (object, REQUIRED)
Immutable audit record. agent_id identifies the depositing
agent. audit_hash chains to the arbitration event record.
metadata (object, REQUIRED)
evidence_count is the number of examples used to produce the
pattern. version increments when a superseding artifact wins
resolution at this URI.
4.3. What the Schema Does Not Contain
The artifact schema contains NO field for:
- Raw data records or observations
- Model weights or parameters
- Training data, labels, or identifiers
- Any value-dependent (non-structural) representation
This is an architectural guarantee, not a policy.
5. Registry Entry — The Arbitration Gate
5.1. The Gate
An artifact MAY enter the reason:// registry only by winning a
live arbitration round conducted by a conformant WARF node.
There is no direct deposit API. No agent may add an artifact
by submitting it unilaterally.
5.2. Arbitration Process
1. An arbitration event is opened under a target reason:// URI.
2. One or more agents submit cargo packages: structured objects
containing an answer (the proposed artifact pattern) and a
supporting evidence corpus.
3. The WARF node scores each submission using the PCF mechanism
[WARF-PROTO]. Identity has zero coefficient in scoring.
4. The highest-scoring submission wins. Its artifact is deposited
under the target URI with its earned score.
5. The arbitration event is closed. A SHA-256 chained audit
record is produced and made publicly accessible.
5.3. Score Immutability and Replacement
The score carried by an admitted artifact is the score it earned
at admission. Scores MUST NOT be recomputed at resolution time.
If a subsequent arbitration round under the same URI produces a
winner with a higher score, that artifact becomes the resolution
winner. Previous winners remain in the artifact store at their
original scores and MAY be queried directly by arbitration event
ID.
5.4. Self-Initiated Arbitration
An agent MAY initiate a single-submission arbitration round. The
submission is evaluated against the evidence it provides. If it
meets the minimum convergence threshold of the target Xport node,
it is admitted. This enables agents to deposit artifacts without
waiting for competing submissions.
6. Resolution Mechanics
6.1. Resolution Request
A resolution request MUST specify a fully qualified reason:// URI.
Example (Python SDK):
from reason_py import ReasonClient
client = ReasonClient(endpoint="https://xport.astrognosy.com")
artifact = client.resolve(
"reason://medicine/records/"
"longitudinal-maintenance-prediction"
)
6.2. Resolution Response
The registry MUST return the artifact with the highest current
score registered under the requested URI. The response MUST
conform to the artifact schema defined in Section 4.
A consuming agent applies the artifact as follows:
for record in my_data:
similarity = compute_similarity(record.features,
artifact.pattern)
if similarity > artifact.thresholds.high_confidence:
flag(record)
6.3. Failure Cases
URI_NOT_FOUND (404)
No artifact has been admitted under the requested URI.
REGISTRY_UNAVAILABLE (503)
The registry is temporarily unreachable. Implementations
SHOULD cache the last known artifact and serve it with a
staleness flag.
7. Privacy Guarantees
7.1. Structural Non-Invertibility
The artifact pattern field contains a structural centroid — the
mean of an agent's learned structural embedding vectors. This
representation is non-invertible by mathematical construction.
Given centroid C computed from n example vectors v_1 ... v_n:
C = (1/n) * sum(v_i, i=1..n)
Any reconstruction attack's best estimator for any v_j is C
itself. The expected reconstruction error grows with n.
7.2. Empirical Bound
From the WARF P2P protocol study (n=4,200 held-out vectors,
high-dimensional structural space):
r = 0.0149 +/- 0.002
Less than 1.5% of the original signal is recoverable from the
centroid. A PAC-style closed-form bound for arbitrary PCF
domains is an open research question; the above is an empirical
result from temporal anomaly detection patterns.
7.3. Architectural Guarantee
The schema has no field for raw data. No conformant implementation
can include raw data in an artifact. The privacy guarantee is
structural — it is enforced by the protocol definition, not by
policy or trust assumptions.
8. Governance
8.1. Bootstrap Phase (now — Q4 2026)
Registry operator: Astrognosy AI (Pacific Intelligence Concepts).
Namespace minting: first-come, first-served via reason.register().
Collision arbitration: handled by WARF scoring.
Audit log: all deposits and scores publicly accessible.
Astrognosy AI's control during this phase is explicitly stated in
the patent filings and this specification. It is a pragmatic
bootstrapping choice, not a permanent governance claim.
8.2. Transition Phase (Q4 2026)
An IETF Working Group is planned. The registry will be handed to
a neutral foundation or IANA-equivalent body. Proposed governance
board composition: 1/3 operators, 1/3 researchers, 1/3 adopters
(rotating seats).
8.3. Community Phase (2027+)
DAO-style proposals for new top-level namespaces. Economic layer
(provenance chains and proportional compensation, per patent
Claims 37-44) activated only after an adoption threshold is met
(target: 100 validated artifacts and 10,000 monthly resolutions).
Specific threshold subject to governance board decision.
9. Security Considerations
Arbitration Integrity
PCF scoring is deterministic and CPU-native. The SHA-256 audit
chain makes score manipulation detectable. Implementations
SHOULD verify audit hashes before accepting artifacts.
Namespace Squatting
A first-come, first-served namespace policy during bootstrap
creates squatting risk. The arbitration gate mitigates this:
a squatted URI will be displaced by a higher-scoring artifact
from a legitimate submitter.
Centroid Poisoning
A malicious agent could submit a crafted centroid designed to
produce false positives in consuming agents. The arbitration
gate mitigates this: a poisoned centroid that does not earn a
competitive convergence score cannot enter the registry.
Replay Attacks
Arbitration event IDs and audit hashes SHOULD be verified by
Xport implementations to prevent replayed submissions.
10. IANA Considerations
This document requests registration of the URI scheme "reason"
in the IANA URI Schemes registry [RFC7595].
Scheme name: reason
Status: Provisional
URI scheme syntax: See Section 3.1
URI scheme semantics: Identifies a named reasoning artifact
in the reason:// registry
Encoding considerations: ASCII only; lowercase
Applications/protocols that use this scheme: Autonomous agent
systems implementing the WARF protocol
Contact: jacob@pcfic.com
References: This document
11. References
11.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997,
<https://www.rfc-editor.org/info/rfc2119>.
[RFC8174] Leiba, B., "Ambiguity of Uppercase vs Lowercase in
RFC 2119 Key Words", BCP 14, RFC 8174,
DOI 10.17487/RFC8174, May 2017,
<https://www.rfc-editor.org/info/rfc8174>.
[RFC7595] Thaler, D., Ed., Hansen, T., and T. Hardie,
"Guidelines and Registration Procedures for URI
Schemes", BCP 35, RFC 7595, June 2015,
<https://www.rfc-editor.org/info/rfc7595>.
11.2. Informative References
[WARF-PROTO]
Westerbeck, J., "WARF Protocol v1.0 — Web Agent
Reasoning Federation", Astrognosy AI, March 2026.
[PCF-PATENT]
Westerbeck, J., "Positional Correlation Field",
U.S. Provisional Patent Application 63/978,633,
February 2026.
Appendix A. Non-Invertibility Demonstration
The following Python script demonstrates the empirical
reconstruction error for a centroid-based representation.
It is self-contained and reproducible.
<CODE BEGINS>
import numpy as np
np.random.seed(42)
n, d = 4200, 512
shared = np.random.randn(d)
vectors = shared + np.random.randn(n, d) * 0.1
centroid = vectors.mean(axis=0)
errors = (np.linalg.norm(vectors - centroid, axis=1)
/ np.linalg.norm(vectors, axis=1))
print("r = {:.4f}".format(errors.mean()))
# Output: r ~= 0.0149
<CODE ENDS>
Authors' Addresses
Jacob Westerbeck
Astrognosy AI / Pacific Intelligence Concepts
Email: jacob@pcfic.com