Wire Frontier Binding Syntax

Date: 2026-05-07 Scope: syntax pressure exposed by the C build example, source includes, makeEach, trailing where, node-local CorePure residue, and egress projection. Branch / PR: wire-haskell-linearity-parity / PR 168. Layers examined: implementation, Wire reference docs, architecture chapter 05, ADRs 0031/0039/0048/0049/0050/0051, and the C build example. Layers skipped: Lean theory and external literature; this memo records syntax design space before theorem or literature alignment. Method: architecture classification plus research synthesis through the seven archetype lenses, with emphasis on Poiesis.

Executive Summary

The C build example exposed an authoring pressure, but the first-pass note overstated the missing surface. ADR 0031 already rejected the pre-output let ... in block and chose trailing where <record-expr>; for node-local authority-free scope. The honest open question is narrower: whether trailing where remains ergonomic for source-driven graphs, and how to name the distinct operation that maps local record fields into output ports.

Architecture classification: Existing ADR Concern + Scope Too Big. ADR 0031 constrains binding placement; ADR 0039 supplies the node boundary normal form; ADR 0049 covers * as a topology adapter; ADR 0050 covers CorePure residue; ADR 0051 covers source includes. The surviving novel slice is egress projection: typed, injective field-to-port exposure inside a node, distinct from topology fan-out.

Archetype Synthesis

Key Findings

[P1] Node-local planning already has a canonical surface: trailing where.

Category: ADR Constraint Status: Corrected Confidence: High Surfaced by: Episteme, Themis, Sophia Primary evidence: docs/ADRs/0031-wire-binding-forms-and-where-clauses.md:293-355, docs/ADRs/0039-wire-node-boundary-transform-normal-form.md:33-68, docs/Reference/Wire/grammar.md:618-631 Cross-reference: ADR 0031 ↔ ADR 0039 ↔ grammar

The first-pass finding treated node-local planning as a missing surface and proposed a pre-output let block between <- and ->. ADR 0031 already considered and removed that exact placement. Its canonical rewrite is trailing where <record-expr>;, with where let ... in { ... }; for sequential shared intermediates. ADR 0039 then assigns that local record to the node boundary normal form rather than to graph topology.

Why it matters: The C build smell is not proof that Wire lacks local scope. It is evidence that the existing where position should be exercised on large examples before any supersession of ADR 0031 is justified.

Next step: Keep the C-build plan inside plan_build using trailing where let ... in { ... };. Only revisit placement if large examples still read poorly after using the existing surface.

[P2] The real binding question is placement ergonomics, not semantics.

Category: Existing ADR Concern Status: Reframed Confidence: High Surfaced by: Logos, Kritikos, Techne Primary evidence: docs/ADRs/0031-wire-binding-forms-and-where-clauses.md:293-355, docs/ADRs/0050-wire-corepure-output-residue.md:41-69, src/Cortex/Wire/Parser.hs:1170-1188 Cross-reference: ADR 0031 ↔ ADR 0050 ↔ parser

ADR 0050 says static CorePure and delayed input-dependent CorePure are the same authority-free language separated only by dependency timing. ADR 0031 says that node-local CorePure sharing is expressed by a trailing where-record, not by pre-output syntax. That combination is coherent: inputs are available to the where-expression, and its record fields are opened into the node body scope used by output equations.

The remaining concern is ergonomic and visual. Trailing where puts the declarations after the output equations that use them. That may be alien for large source-driven graphs, but it is an accepted surface, not an absence.

Why it matters: A future ADR should not be “add node-local let.” If the evidence eventually warrants a change, it should be “supersede ADR 0031’s placement decision while preserving its three-surface binding model.”

Next step: Rewritten examples should compare three forms: top-level plan record, trailing where, and any proposed pre-output form. The burden is now on the proposed placement change.

[P3] Record-to-port projection is not graph fan-out, but it needs a linear rule.

Category: Terminology Gap Status: Inferred Confidence: High Surfaced by: Themis, Kritikos, Poiesis Primary evidence: docs/Architecture/05-wire-language.md:169-187, docs/ADRs/0049-wire-fan-phantom-adapter.md:34-72, docs/ADRs/0051-wire-source-includes-and-item-generation.md:58-66, examples/wire/c-build/c-build.wire:102-198 Cross-reference: architecture ↔ ADR 0049 ↔ example

Wire rejects implicit edge fan-out and ADR 0049 provides * for topology-level record↔ports adapters. But inside a node, taking a local record value and assigning distinct fields to distinct output ports is not topology fan-out. It is egress projection: the node’s egress adapter maps local values into the declared output frontier. The language lacks a name for that distinction, so the source must currently spell every projection:

-> linkAppSpec: CommandSpec = linkAppCommandSpec;
-> smokeSpec: CommandSpec = smokeCommandSpec;
where let
  plan = { ... };
in
{ linkAppCommandSpec = plan.linkAppCommandSpec; smokeCommandSpec = plan.smokeCommandSpec; };

The dangerous version would be a spread form that silently maps one field to multiple outputs or duplicates a whole value across ports. The admissible version is an injective field-to-port mapping: each output port consumes exactly one expression, and each projected record field is used at most once unless the expression is deliberately copied inside CorePure.

Why it matters: Without the term “egress projection”, reviewers can misclassify useful syntax as forbidden fan-out. With the term, the proof obligation is clear: projection preserves output port linearity because it allocates distinct record fields to distinct output resources.

Next step: Add “egress projection” to the ADR draft vocabulary and require a compile-time duplicate-field-use rejection for any spread shorthand.

[P4] Source-generated graphs need a phase-indexed binding story, not more macros.

Category: Design Tension Status: Observed Confidence: High Surfaced by: Logos, Themis, Techne Primary evidence: src/Cortex/Wire/Include.hs:9-12, src/Cortex/Wire/Parser.hs:1364-1371, docs/ADRs/0051-wire-source-includes-and-item-generation.md:34-56 Cross-reference: implementation ↔ ADR 0051

The current implementation has two phase facts:

• source includes expand before parsing and embed ordinary literals;
• makeEach only sees preceding closed static item lists during structural expansion.

Those facts are good, but the syntax can make them feel arbitrary. A future binding design should make phase visible by placement and dependency, not by adding unrelated macros. Top-level let, form-local let, source includes, and node-local trailing where should all be explainable as one phase-indexed binding ladder.

Why it matters: More ad hoc macros would make the language powerful but unteachable. A phase ladder lets authors understand why include_dir may shape graph topology while runtime CorePure cannot read files.

Next step: Add a reference-doc section that lists binding phases explicitly: source elaboration, graph expansion, node-frontier CorePure, executor runtime. ADRs 0031/0050/0051 already decide the policy; the missing artifact is one consolidated explanation.

[P5] The novelty is not syntax sugar; it is a language where frontier declaration and data construction interleave.

Category: Novel Idea Status: Speculative Confidence: Medium Surfaced by: Poiesis, Sophia Primary evidence: current example and docs cited above; no external literature checked Cross-reference: implementation ↔ architecture ↔ future paper framing

Most languages separate “compute a value” from “declare an interface”. Wire is pushing toward a different object: a node is a boundary transformer whose input frontier, local CorePure plan, output frontier, and graph composition all live in one source artifact. That makes syntax unfamiliar: ordinary block structure fights with frontier order. But the payoff is also unusual: the graph’s resource frontier can remain visible while deterministic data construction is still local enough to read.

Why it matters: This is paper-worthy if handled carefully. The ergonomic contribution is not “we added local lets”; it is “we found a syntax where boundary resources and deterministic local construction are co-designed rather than layered after the fact.”

Next step: Park as paper-language material. Validate first through a prototype and large example rewrite; do not claim novelty externally until compared against dataflow DSLs, Nix-style source elaboration, build systems, and string-diagram languages.

Poiesis Option Set

Missed Abstractions

Evidence Gaps

Recommended Actions

Act now

• Use trailing where let ... in { ... }; for the C-build plan_build node so PR 168 tests ADR 0031’s current answer.

Design next

• Draft ADR: Wire Egress Projection. Owner: architecture. Cost: M.
• Patch the Wire reference with a single binding-phase ladder: source elaboration, graph expansion, node-frontier CorePure, executor runtime. Owner: docs. Cost: S.
• Revisit where placement only after where has been tried on at least two large examples. Owner: architecture. Cost: M.

Write up

• Add a paper-note paragraph: “egress projection maps deterministic local data into linear output resources.” Owner: research / paper track. Cost: S.

Parked

• Contract-driven output derivation. Revisit only after egress projection has a typed carrier.
• Pre-output let syntax. Revisit only as an explicit supersession of ADR 0031’s placement decision.
• Record-spread output syntax. Revisit only if projection vocabulary still leaves large nodes too verbose after a large-example rewrite.
• External prior-art comparison. Revisit before public novelty claims.

Claim-to-Evidence Matrix