HostModel integration into TestClaim substrate — IRT-3 / Practice 11 alignment for run_test_claim (operator-direct via PM)

[briansrls]

Summary

Integrates the v4 TestClaim runner with the HostModel substrate: EvalContext now carries HostModel, run_test_claim calls eval(tree, host, inputs), and cache receipts key on the host model instead of parallel evaluator/resources/extdeps Node slots. Existing typed-input fixtures and the six workflow pipeline pass-gate rows are rewired through the same runner, with a new eval_host_model_mvp claim covering the host-eval path.

Test plan

• cargo run -p v2-compiler --bin v2-compiler -- compile --source-root src/v4 --output-dir /tmp/v4-out-still-tern-90 --target rust -> compiled: 124 files emitted, 0 diagnostics
• python3 scripts/strict_deprose_dag.py --check -> OK
• Verified PR diff is scoped to src/v4/compiler/05_eval.dag, src/v4/test/claim/manual/eval_host_model_mvp.dag, src/v4/test/claim/manual/infer_ground_add_mvp.dag, and src/v4/test/claim/workflow/pipeline_rejections.dag.
• Confirmed the six LB-T22-3213 workflow rows still call workflow_pipeline_pass_gate(...) (gate_dangling_bootstrap_output, gate_duplicate_job_id, gate_duplicate_gate_id, gate_dangling_need, gate_dangling_gate_job, gate_dependency_cycle).

Note: direct v2-compiler run --source-root src/v4 --function ... probes were attempted for run_eval_host_model_mvp and workflow_pipeline_pass_gate, but the v2 interpreter did not produce a prompt receipt and was stopped; the acceptance receipt used here is the requested v4 source-root compile with --target rust, not --target dag.

[briansrls]

Review metadata

• Provider / model: codex / unknown
• Commit: eb7994cb · Trigger: schedule
• Thinking: 622s wall

---

✅ This .dag substrate/test-claim change aligns with the T-22/T-34 HostModel route and I found no blocking concerns.

[briansrls]

Heads-up from PR #3522: the umbrella HostModel shape is being decomposed per Option C. The current PR now removes src/v4/std/host.dag, introduces abstract runtime carriers in src/v4/std/runtime.dag, and adds the concrete v4 evaluator runtime bundle under src/v4/extdeps/runtimes/v4_evaluator.dag. Eval no longer takes HostModel; it takes InterpretationAlgebra directly. PR #3513 will need to rebase and replace EvalContext/TestClaimCacheKey HostModel usage with the specific runtime carriers it actually needs, rather than restoring a bundle.

[briansrls]

Closing as superseded by the sequencing.

PM ratified disposition (a) from a three-option coordination discussion: PR #3522 (HostModel decomposition per Option C) lands first; the substantive value-preservation work in this PR becomes a focused follow-up dispatch re-applied at the decomposed shape.

Why this PR closes rather than merges:

PR #3522 (still-open) implements the operator-ratified Option C decomposition: drops the HostModel umbrella, exposes per-concept carriers (InterpretationAlgebra, EvaluationEnvironment, RuntimeValue, ValueRepresentationModel, ExecutionSemantics, ResourceEffectBoundary) in src/v4/std/runtime.dag, authors concrete runtime extdeps at src/v4/extdeps/runtimes/. This PR's EvalContext { host: HostModel } umbrella shape is being walked back in favor of the decomposed shape.

The substantive principles still-tern-90 (this PR's author) captured ARE recognized as substantive and stay alive:

• IRT-3 alignment — run_test_claim routes through host eval at arbitrary-node granularity (operator-direct standing, 2026-05-17)
• LB-T22-3213 bilateral-binding preservation — six workflow-pipeline TestClaimRun<CiPipeline> rows continue through workflow_pipeline_pass_gate under live eval
• Value-preservation work (head 1ac4eb7): TestClaimRun<cache, result> separate generics + Outcome preserved in verdict comparison + literal-vs-resolved-type fix for eval_host_model_mvp
• Substrate-collapse insight: replacing 3 implicit-runtime-Node fields with proper substrate authority — directly applied differently in PR HostModel decomposition (option C) — drop Host* umbrella + abstract runtime carriers + concrete runtime extdeps separation #3522's decomposition (per-concept carriers rather than collapse to umbrella)

These principles become the scope of a focused follow-up PR dispatched after PR #3522 merges, re-applying the value-preservation work at the decomposed shape:

• run_test_claim consumes InterpretationAlgebra (per HostModel decomposition (option C) — drop Host* umbrella + abstract runtime carriers + concrete runtime extdeps separation #3522's eval signature) AND preserves Outcome<RuntimeValue> through verdict comparison
• TestClaimRun<cache, result> generic split
• Literal-vs-resolved-type comparison fix

Provenance trail:

1. sharp-otter-25 (PR T-22 eval body (attempt-7) #3508, merged 20:48Z) — T-22 eval body authored against the HostModel umbrella shape (pre-decomposition).
2. still-tern-90 (this PR) — integrated HostModel umbrella into TestClaim substrate; surfaced design-PR-tier codex review (Branch/Bind via HostInterpretationAlgebra + run_test_claim via host eval) leading to operator-PM disposition (a) dispatching this work.
3. operator + PM smart-lynx-142 — ratified disposition (a) HostModel integration into TestClaim substrate, then drilled deeper into HostModel naming + ratified Option C decomposition (drop umbrella entirely).
4. witty-ant-128 (PR HostModel decomposition (option C) — drop Host* umbrella + abstract runtime carriers + concrete runtime extdeps separation #3522, open) — implements operator-ratified Option C decomposition.
5. post-HostModel decomposition (option C) — drop Host* umbrella + abstract runtime carriers + concrete runtime extdeps separation #3522-merge follow-up (TBD worker) — re-applies still-tern-90's value-preservation work at the decomposed shape.

This PR's work was the catalyst for arriving at the decomposed shape; preserved as provenance via this closing comment.