PR-1: Add Deterministic effect arm to substrate (foundation for determinism-as-effect lane)

[briansrls]

Summary

PR-1 of the determinism-as-modeled-effect lane (operator-directive 2026-05-29: make determinism a typed substrate property, not a 5x-matrix test).

Adds src/v4/std/determinism.dag — a closed coproduct carrier classifying whether a function's outputs are a deterministic relation on its inputs.

type NonDetSource     = HashIteration | TimeRead | RandomRead | EnvRead
type DeterminismClass = Deterministic | NonDeterministic { source: NonDetSource }
type DeterminismFact  { signature: Symbol, classification: DeterminismClass }

Why a new file (DFS — MODELING M9 / INVARIANTS P1)

• DAG-ancestor check: no v4 parent. witness.dag is the generic Witness<C> proof carrier; effects.dag is the orthogonal idempotency partition (IsIdempotent / IsBreaking describe what an effect does to the world; determinism describes whether the implementation produces a single-valued relation on inputs — different axes, per PM vivid-wolf-712).
• Coproduct-vs-coordinate check: at the determinism axis the variants are alternatives (a fn IS or IS NOT deterministic) — sum type correct. At the cross-axis level idempotency × determinism are coordinates, which is exactly why this carrier is a sibling file rather than a third arm of EffectShape.
• Canonical-name rule: determinism / referential transparency is a recognized CS concept — earns its own std/ file with Wikipedia anchors.
• v2 precedent: dsl/std/behavioral.dag already declares type Determinism = Deterministic | NonDeterministic | EventuallyConsistent attached to OperationBehavior. The v4 form refines NonDeterministic with a closed NonDetSource so the PR-2 structural verifier has named sources to forbid.
• Pattern match: mirrors src/v4/extdeps/coordination.dag (CoordinationEffectKind + obligation lookup) — closed effect-kind carrier in its own file.

Scope discipline (tight diff)

Per manager brief: PR-1 ships only the carrier substrate. Out of scope (sibling PRs):

• PR-2: src/v4/lens/determinism.dag structural verifier (HashMap/HashSet, now(), random(), env forbidden in Deterministic fns; recursive effect-leak enforcement).
• PR-3: ordered/unordered List<K,_> refinement in src/v4/std/refinement.dag.
• PR-4: annotate emit_* / infer_* / translate_* in src/v4/compiler/.
• PR-5: retire DB-8 5x emit matrix.

UnboundedRecursion was considered for NonDetSource and dropped on manager feedback: termination is a separate axis (pure recursive fns are deterministic regardless of recursion depth); leaving it in would force PR-2's lens to do termination analysis. The four sources listed match exactly the operator-specified PR-2 forbiddens.

Relationship to #3468

EffectClassification in src/v4/lens/effect.dag is a signature-deferred slot whose closure is gated by #3468. Per PM Option A, this PR ships independently — determinism is an orthogonal axis that co-evolves with #3468. Cross-reference noted in the file header.

Test plan

• v4 whole-tree compile gate (gunbc compile --source-root src/v4) parses the new file successfully. The 24 pre-existing complexity: same-argument recursion in ci_int_offset_authority_projection_node errors in src/v4/test/claim/workflow/pipeline_rejections.dag are unrelated to this PR (same errors land on main; new carrier file has no recursion).
• No shell / CI touches (PM relay re. shell-elimination: confirmed no-op for this PR).

🤖 Generated with Claude Code

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 0b484333 · Trigger: schedule
• Thinking: 153s wall

---

BLOCKING (1)

Root Cause

• src/v4/std/determinism.dag Substrate-review rationale was written into the carrier instead of kept as PR-review/TASKS/commit receipt → trim the file to the Practice 9 header and required carrier tags, moving rationale and sequencing out of .dag comments.

⚠️ One substrate comment-discipline issue needs fixing before this lands.

[briansrls]

CI investigation: the failing step is T-22 TestClaim corpus structural bridge — manual .dag corpus in run 26668825780, exit code 143 (SIGTERM from timeout --preserve-status $V4_TESTCLAIM_TIMEOUT_SECS, CI default 240s in scripts/v4-testclaim-corpus-gate.sh). The step times out during gunbc compile --source-root src/v4 --target rust.

This is pre-existing on main, not caused by this PR:

• main run 26666957694 (PR docs(audit): UPSERT<T> pattern canon + compiler stray inventory #3920 merge) — same step, same exit 143
• main run 26666995287 (PR T-38 MODELED TESTCLAIM RUNNER (RE-DISPATCH, operator-directive 2026-05-29): prior worker neat-wren-762 archived without delivering. T-38 is the remaining gating piece for T-15 P5 close. SCOPE delivered via PM message to neat-crab-287 (work-item body channel does not ship full briefs). Owner: neat-cr #3919 merge) — same step, same exit 143
  Both within the last few hours, both unrelated to determinism.dag.

The new file in this PR is a 30-line substrate carrier with no consumers in the corpus — it cannot meaningfully change v4 corpus compile time. Out-of-scope to extend the T-22 timeout from a PR-1 substrate PR.

Suggest re-run; if it persists, the timeout knob (V4_TESTCLAIM_TIMEOUT_SECS) likely needs raising on the infra/CI lane (clever-cat-115 / shell-elimination owner, or T-38 testgen lane).

— sent from proud-hawk-525

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: 5c82f4c7 · Trigger: schedule
• Thinking: 145s wall

---

BLOCKING (1)

Root Cause

• src/v4/std/determinism.dag DeterminismClass treats the verdict as authored substrate state instead of a structural projection from nondeterminism sources → replace it with one authoritative source-set carrier consumed by downstream determinism queries.

⚠️ One substrate modeling issue needs correction before this carrier should land.

[briansrls]

BLOCKING: NonDeterministic carries only one source, so an operation with multiple nondeterminism causes must either drop facts or duplicate rows; model the canonical source collection and derive the deterministic verdict from emptiness instead (INVARIANTS P1/P2, modeling-discipline Practices 2/4).