Status: Accepted Version: 1.0 Date: 2025-12-21 Supersedes: N/A Related ADRs: ADR-003, ADR-014 Related PRDs: N/A
SEA-Forge™ operates on a foundation of formal specifications. However, domain knowledge often originates as unstructured material:
We need a formalized methodology for extracting architecture specifications from unstructured sources while eliminating ambiguity, narrative drift, and accidental complexity.
Adopt the T2S (Transcript-to-Spec) protocol as the canonical extraction pipeline for converting unstructured intent into executable architecture specifications.
T2S (Transcript-to-Spec) is the method of turning messy, chronological “what happened / what we meant” text into a clean, domain-agnostic system specification.
T2S is executed explicitly and sequentially:
Identify the core building blocks from the source:
Keep metaphors as helpful labels, but translate them into implementable mechanics:
Pull out the non-negotiables and restate them as enforceable constraints:
Determine the system’s operational pattern:
Each unit gets a signature:
Turn “make it good” into explicit gates that fail fast:
Capture how the system changes safely over time:
Output the structured specification:
T2S creates a closed loop with the rest of SEA-Forge™:
1
Transcript/Source → T2S → Spec → Implementation → Validation → Evolution → New Transcript
The spec is not an endpoint — it’s a waypoint in a continuous refinement cycle.
Rejected — Introduces interpretation variance; loses traceability to source.
Rejected — Needs human review for invariant identification and governance decisions.
Rejected — Unstructured documents contain hidden complexity that causes downstream drift.
The following prompt can be used to invoke T2S on any source:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
## T2S Invocation
Apply your **T2S (Transcript-to-Spec)** process to the attached source.
Treat the source as raw, unstructured intent (transcript / essay / design notes / conversation).
Perform T2S explicitly and sequentially:
1. **Extract primitives** — core nouns, verbs, roles, artifacts, and conceptual objects
2. **Translate metaphors into mechanisms** — restate metaphorical language as concrete, implementable system behavior
3. **Declare invariants (anti-technical-debt rules)** — identify non-negotiable constraints implied by the source
4. **Identify the execution model** — pipeline, DAG, compiler stages, feedback loops, or state machines
5. **Define contracts** — inputs, outputs, responsibilities per component or stage
6. **Insert quality gates and validation mechanisms** — tests, audits, isomorphism checks, drift detection
7. **Extract governance and evolution model** — how the system changes safely over time
8. **Produce spec-ready building blocks** — structured so they can be assembled into a full architecture spec
Be faithful to the source's intent, but remove ambiguity, narrative drift, and accidental complexity.
Output should be **domain-agnostic**, **implementation-ready**, and **explicitly anti-technical-debt**.
Fast T2S (Extraction Only):
Apply T2S but stop after: primitives extraction, metaphors → mechanisms, invariants. Output only structured bullet points suitable for later spec synthesis.
Full T2S → Architecture Spec:
Apply T2S, then synthesize results into a full architecture spec structured like: Core intent, Non-negotiable principles, System decomposition, Execution model, Contracts, Quality gates, Governance & evolution, Portability.
✅ MVP (core methodology for specification extraction)