Skip to main content

Angle

The spec was frozen earlier today. The implementation is now live. @stratt/fingerprint passes every test vector from the normative specification — 14 input/output pairs covering all 5 unit types, Unicode normalisation equivalence, block scalars, empty arrays, max-complexity chains, and deliberate tampering. 98 tests total. The identity layer works.

Key Messages

For the design journal (stratt.engineer): Layer 0 is real. @stratt/fingerprint implements the full 5-stage canonical serialisation pipeline: YAML parse → object transform → canonical JSON → UTF-8 encode → Blake3 hash. The two-tier API accepts both parsed objects and raw YAML — so the schema layer can fingerprint validated units without re-serialising to YAML. 98 tests. Synchronous Blake3 via WASM. Tamper detection returns the exact hex position of the first mismatch. The spec-to-implementation gap is closed. For developer audiences (Show HN / dev community): What does it take to give a prompt a cryptographic identity? Five stages, one spec, and 14 test vectors. STRATT’s @stratt/fingerprint package takes a YAML prompt unit, applies a deterministic transformation pipeline (RFC 8785-aligned canonical JSON with YAML-specific extensions for null removal and Unicode NFC normalisation), and produces a Blake3 fingerprint. Change one character, the fingerprint changes. Move the prompt between systems, recompute, verify. The spec ships with the code — any implementation that passes the 14 vectors is conforming. For LinkedIn / professional network: Shipped @stratt/fingerprint — the identity layer for the STRATT prompt architecture. Every prompt unit now has a Blake3 fingerprint: deterministic, reproducible, machine-verifiable. 98 tests validate the full pipeline against a normative specification with 14 test vectors. One session from spec to passing tests. Schema-first, spec-first, trust-first.

Content Opportunities

  1. Design journal entry: “From spec to implementation in one session” — the test-vector-driven workflow, the 97/98 first-pass result, the two-tier API decision
  2. Technical deep dive: “How STRATT fingerprints a prompt” — walk through TV-01 (minimal role unit) stage by stage with byte-level detail
  3. Thread/short-form: The NFC normalisation edge case — why é (precomposed) and e + combining accent (decomposed) must produce the same fingerprint, and what happens to trust if they don’t
  4. Comparison post: “STRATT vs prompt engineering tools” — no other system provides cryptographic prompt identity with a normative spec and conformance test suite

Narrative Arc

This campaign continues the story from “Machine-Verifiable Prompt Architecture” (the spec) and “STRATT Schema Package” (Layer 1). The progression is:
  1. Schema defines what a valid unit looks like (Layer 1, shipped)
  2. Spec defines how to compute identity (stratt-canonical-v1, frozen)
  3. Fingerprint package implements identity (Layer 0, shipped — this campaign)
  4. CLI uses fingerprint + schema to publish verified units (Layer 2, next)
The “spec before code” narrative now has a payoff: the spec was right, the implementation validated it, and the test suite proves it.