APL Protocol
Frame-Bound Observation Claims.
A claim without its frame is a rumor. ATL proves a record exists and hasn’t been rewritten. APL proves you’re reading it under the same assumptions it was written. Content is inseparable from context — cryptographically.
The Problem
Cryptographic integrity is solved. Semantic integrity is not. And in domains where meaning matters — science, AI evaluation, forensics, cross-org data exchange — that gap is where silent errors live.
“The measurement was 98.6°F.” — Oral? Axillary? Rectal?
“The model scored 94% accuracy.” — On which eval set, which prompt version, which temperature?
“Uptime was 99.9%.” — Measured how, aggregated over what window, excluding what?
“The two readings match.” — Were they taken under comparable conditions, or are you comparing apples to oranges?
Every record carries an implicit frame — the conditions, instruments, assumptions, and scope under which it was produced. Strip the frame and the claim is still verifiable, but no longer meaningful. Worse: it can be silently compared to claims from different frames, producing conclusions that are cryptographically sound and semantically nonsense.
The Solution
APL makes the frame a first-class, content-addressed citizen. The core insight: bind every claim to its frame, refuse implicit comparison.
Frame-Bound Claims
Every observation carries a frame_ref — a content hash of an explicit frame declaring what was observed, how, and under which assumptions. No resolvable frame, no claim: the verifier returns apl-invalid.
Content-Addressed Identity
Frames, bridges, and transformations are identified by SHA-256 of their JCS-canonicalized bytes. URLs and aliases are discovery hints; the hash is truth. Rename anything, break nothing.
No Implicit Comparison
Claims under different frames are incomparable by default. Cross-frame comparison requires an explicit, content-addressed Bridge that declares scope, assumptions, and losses. Comparing apples to oranges becomes a protocol error, not a judgment call.
Transformation Discipline
Transformations (aggregation, normalization, derivation) must declare both what they preserve and what they lose. Empty losses means “nothing lost” — never “unknown.” Silent semantic drift is made structurally visible.
Deterministic Verification
A conforming verifier returns exactly one of two outcomes: apl-valid or apl-invalid. APL does not adjudicate truth — it adjudicates frame binding. Truth is still your problem; ambiguity is not.
Who Needs This
AI Evaluation & Accountability
Which model version, prompt, dataset, temperature, and seed produced this score? Benchmark results without frames are marketing. APL makes cherry-picking structurally detectable and reproduction mechanically checkable.
Scientific & Research Integrity
Findings bound to methodology, sample, environmental conditions, and explicit non-claims. Reproducibility failures rarely come from fraud — they come from lost context. APL prevents the loss.
Measurements & Telemetry
Sensor type, calibration date, aggregation method, operating envelope — all pinned to the reading. A “99% uptime” claim becomes incomparable to another “99% uptime” claim until a Bridge says otherwise.
Legal & Forensics
Chain-of-custody, investigative assumptions, and evidentiary scope become structurally inseparable from the conclusions drawn. Opposing counsel can challenge the frame, not just the data.
Cross-Organization Data Exchange
When data crosses trust boundaries — between vendors, agencies, jurisdictions — implicit assumptions don’t travel. Bridges make comparability contracts explicit, auditable, and machine-checkable.
Technical Foundation
Open Standard
APL is designed as protocol infrastructure, not a product. Deterministic, substrate-agnostic, minimal surface area.
Cryptographic Primitives
- SHA-256 (FIPS 180-4) for content identity
- JCS / RFC 8785 for canonical JSON serialization
- No novel cryptography — only disciplined use of standards
Reference Profile: APL-on-ATL
APL is substrate-agnostic, but the reference profile embeds metadata.apl inside ATL Receipts. Semantic binding (APL) rides on top of cryptographic history (ATL) without duplication: one layer for “what happened,” one for “what it means.”
Reference Implementations
- APL-Core: Minimal frame binding and verifier logic
- APL/AI-Eval: Specialized profile for AI evaluation contexts
- Test vectors for conformance testing
Learn More
Full protocol specification, data structures, bridges, and integration guides: