ADR-0007 · Force is always explicit; grouping ≠ forcing; operator precedence is global

• Status: Accepted

• Summary: Force is always explicit; grouping ≠ forcing; operator precedence is global (semantic; glyph liquid).

• Status: Accepted

• Date: 2026-06-21 (rewritten from earlier glyph-specific version of 2026-05-31)

• Layer: K (kernel — semantic decisions; surface spellings are liquid)

• Related: 0005 (reactive sampling reuses explicit force; same operator-grammar family), 0006 (the no-hidden-channels principle, applied here to grammar), CBPV foundation

Context

CBPV makes description-vs-value a kind shift that must be marked. Two semantic questions had to be settled at the kernel level:

1. Is the description → value boundary always explicit, or can it be implicit (e.g., parens force, as in some Lisps; or function application implicitly forcing its argument)?
2. Can the operator-precedence table be modified per-module, or is it fixed globally?

Both questions are about whether reading code requires non-local knowledge — the same hidden-channel problem ADR-0006 evicted from closures, sneaking back through the grammar.

Decision (semantic — K-layer)

1. Force is always explicit. A single distinguished syntactic mark performs the description → value kind-shift. Nothing else implicitly forces — not application, not grouping, not equation, not pattern match.
2. Grouping and forcing are distinct. Pure structural grouping does NOT force. A parenthesized expression remains a description until explicitly forced.
3. No user-defined operator fixity. Operator precedence is a fixed global table (like C). Per-module / per-import precedence redefinition is banned; it makes parsing non-local.

Rationale

• Explicit force = the central CBPV distinction is visible at every site it occurs. Greppable. No type-dependent local disambiguation (the failure mode of !x-style force colliding with not).
• Grouping-vs-forcing separation cleanly resolves the syntax tension: the two jobs (structure / kind-shift) get distinct treatments; neither overloads the other.
• Fixed global precedence + explicit grouping = the operator-grammar instance of no-hidden-assumptions. Same principle family as ADR-0006 (capture). Precedence becomes global knowledge (like C); per-module fixity (the actual non-locality) is banned.

Rejected alternatives

Consequences

• The description → value boundary is exactly the set of explicit force sites — greppable, themeable, the language's central distinction made visible.
• Three semantic shapes:
  • pure group (still a description; no kind-shift)
  • force-a-name (kind-shift on a name)
  • group-then-force (kind-shift on a composite expression)
• Sampling notation (ADR-0005) is force-in-equate-position — falls out of force meaning what it already means.
• The spec owes a published precedence table. Global knowledge, like C's.

Current surface spelling (liquid — S-layer, not load-bearing)

The current surface chooses $x prefix for force, (e) for pure grouping, \(e) for string interpolation (so $ stays reserved for force), and ! is freed for actor-send. These glyphs may change as the surface evolves; the K-decision is that the force boundary is explicit and global precedence is fixed — not which characters spell it. See ROADMAP.md for the surface-layer liquidity policy.

Revisit if

• The fixed table proves too restrictive for a domain wanting custom operators → consider scoped fixity that travels with the operator's import (still local; fixity arrives with the operator, never free-floating per-module).
• The current force glyph causes force points to be missed in real code → reconsider the surface spelling at the S-layer (does NOT require a K-ADR change; the semantic commitment stands).