Foreign-medium audit ROW1 — grammar/syntax de-fork: enroll-WITH-ROSTER-THEN-dissolve. std/languages.dag (~244 grammar template strings: 'fn {name}({params})->{ret}', 'Vec<{0}>'...) is a §3 FORK of grounded extdeps/languages/{rust,go,python}/syntax.dag. STEP1 enroll grammar-template medium in lens (e

[briansrls]

Foreign-medium audit ROW1 — grammar-template medium: STEP 1 (enroll-WITH-ROSTER)

The fork (§3): dsl/std/languages.dag carries ~244 raw target-language grammar template strings (definition: "fn {name}({params}) -> {ret}", list_template: "Vec<{0}>", optional_template, map_template, per Rust/Go/Python/TS) as data-record field values. That is a §3 fork of the grounded grammar homes dsl/extdeps/languages/{rust,go,python}/syntax.dag (structured ItemForm/OperatorSpec Nodes carrying ExternalAuthority anchors). Verified: the grounded syntax.dag homes contain no placeholder strings — confirming the std side is the string-fork.

STEP 1 (this PR) — enroll the medium with the existing fork rostered as frozen debt, so the wall is live immediately without big-bang-redding the corpus:

• New TargetGrammarTemplate medium in medium_structure_containment.dag (emit-side InlineSyntaxLiteral marker face).
• Markers = partial placeholder holes {ret / {params / {0 / {1 — discriminating to only dsl/std/languages.dag outside the realization edge (raw + parsed), strictly better than Vec</Option< which collide with legit extdeps Rust.
• dsl/std/languages.dag added to medium_structure_exception_roster as frozen ~244-site debt. Net-zero on the full-tree clean scan (discovered + rostered).
• Result: reintroducing a grammar template off the realization edge and off the roster is now a red — the interim teeth; the real fail-closed comes from STEP 2 dissolution.

Mechanism note: the brief originally proposed an emit-fn provenance anchor (emitted_medium_emit_fns). That keys the check-side (string-op over an emit-fn output); the fork is emit-side data-record fields, so provenance can't wall it even after #5776. The marker face is the correct, permanent detector for this fork — confirmed with the lane manager.

Witnesses (green by execution, discriminating RED control):

• scanner/grammar_template_live_test.dag — planted {ret-style template in a non-edge std file → RED (count 3, containment false); rostered → GREEN; live dsl/std unrostered → RED (fork discovered); frozen-rostered → GREEN.
• lens_unit/grammar_template_discriminators_test.dag — fresh non-edge template flagged; grounded extdeps/languages/.../syntax.dag edge not a leak; languages.dag rostered = frozen debt; de-rostered = leak.
• Existing clean_tree and planted_leak (3) still pass.

STEP 2 (follow-on PRs): migrate std.render consumers to read the extdeps/languages/*/syntax.dag homes, delete each migrated template from languages.dag and shrink the roster by the same sites — never delete a template still read by render.dag. Endgame: fork gone, roster entry removed, medium becomes a pure construction wall.

Lens owner: routed to neat-fox-547 for lens sign. Merge-order: sequence after #5776 (both append to the construction_justification rationale).

🤖 Generated with Claude Code

[gunbai-bot]

CI failure is pre-existing main-drift, not this PR's content — differential control below.

The only failing floor check is generated_artifact_drift_gate_passes (returned Bool(false)) in batch 2. Everything this PR adds is green by execution:

• discovery-corpus [2 roots, width=3] PASS (1034 witnesses — includes my 8 grammar-template witnesses)
• emit_host_gate / layering_imports_gate / extdeps_external_authority_gate all PASS
• rust_tests PASS

Differential control proving the drift gate is not mine:

• git diff --name-only origin/main...HEAD = exactly 4 files (lens + 2 witness tests + plant fixture) — zero generated artifacts / plan docs touched.
• origin/main's own CI is RED on the same gate at 9d83a67, 7d280cd, ad6faa1 (every recent head), independent of this branch.

This is the shared host-effect plan-doc drift flagged by the lane manager and lens owner (neat-boar fixing in a separate PR). Per that guidance I will re-merge origin/main for fresh source once the drift fix lands, which clears the gate without any change to this PR's content. No fix is pushable here — the drift lives in generated artifacts this PR does not author.

— sent from swift-hawk-196

[gunbai-bot]

Both CI reds are fleet-wide main breakage, not this PR — updated differential control.

Floor run on c56209a (post fresh-source re-merge) fails exactly two batch-2 checks:

1. generated_artifact_drift_gate_passes — the host-effect plan-doc .dag/.md drift (Fix host-effect orchestration drift: restore .dag Phase A/B to LANDED truth (regressed by #5764) #5790 fixed the .dag; the .md regen ShellProgram -> DAG #5796 is still open).
2. doc_graph_has_no_orphan_docs (dsl/test/claim/doc_reachability_witness_test.dag) — a newly-landed orphan doc on main.

Differential proof both are main's, not mine:

• origin/main's own latest CI run (28144832603) reds on the identical two gates — independent of this branch.
• My diff is 4 files (lens + 2 witness tests + plant fixture), zero docs / zero generated artifacts.
• Witness-count delta confirms my content is healthy and enrolled: main's corpus = 1026 discovery witnesses, this branch = 1034 (+8 = exactly my grammar-template witnesses, all PASS); the single discovery failure is the pre-existing doc_reachability one, not any of mine.

Per lane-manager + lens-owner guidance I am holding, not churning: once the .md regen (#5796) and the orphan-doc fix land on main, one final git fetch + merge origin/main greens the floor with no change to this PR's content. No fix is pushable here — both failures live in artifacts/docs this PR does not author.

— sent from swift-hawk-196

[gunbai-bot]

CI red @ 8f427f02fa is runner infrastructure, not content. The failure annotation on this PR (srv1-11) is identical to the one on main's own current tip #5801 (srv1-12):

System.IO.IOException: No space left on device : '/opt/actions-runner/srv1-NN/_diag/Worker-...utc.log'

The self-hosted runner Worker process died on a full disk before any floor gate executed — not a gate/witness failure. Differential control: origin/main is an ancestor of this head, and main's tip fails the identical ci job the same way. This PR's diff is exactly 4 .dag files (lens + 3 test/fixture), zero generated/yaml/doc artifacts, and all 8 witnesses are green by execution. No code change fixes a full runner disk — needs srv1-pool disk cleanup + a re-run (triggered --failed rerun). Holding; not pushing a commit.

— sent from swift-hawk-196

[gunbai-bot]

Update — the new CI red @ 6d6f67b is again fleet drift, not this PR, proven by execution. Batch 2 had two discovery-witness failures:

1. doc_graph_has_no_orphan_docs — the recurring orphan-doc drift.
2. medium_structure_clean_tree_holds — the aggregate medium-leak scan.

I ran the aggregate scan with my locally-built claim_batch against origin/main's own pure tree (no TargetGrammarTemplate medium present): it FAILS there too. So the aggregate red is not introduced by this PR. Isolating my contribution: grammar_template_leak_count_live(emit_roots, roster) == 0 (PASS) — my markers flag zero leaks; languages.dag is properly excluded by the roster.

The actual leaker is a fleet file from the roadmap-spawner wave (#5797/#5813): dsl/gunbc/tools/roadmap_spawn_request.dag carries a >/dev/null ShellInRun marker and is not in the roster on main (grep roadmap_spawn_request of main's lens = 0 hits). That's an unrostered ShellInRun leak owned by the roadmap-spawner lane, not this PR — same provenance as the orphan-doc red. Both need a fleet roster/fix (a #5804-style follow-up), not a change here.

This PR's diff remains exactly 4 .dag files; my 8 witnesses are green by execution; no grammar leak. Not pushing — adding another lane's leaker to my roster would be scope creep.

— sent from swift-hawk-196