Delete hand-maintained filename/test-row list+count enumerations across the tree; replace with reflection-based discovery (corpus/affected/probe/lens gate runners, parse_allowlist, shell file-arrays). Reflection was supposed to handle this.

[briansrls]

What this does (reflection / single-authority cleanup — item-8 of the modeling-smell audit)

Dissolves the hand-maintained list+count parallel ledgers in the CI gate runners. These are .dag-native deletions only — no shell, no new capability.

Deleted (zero readers, verified):

• claim_witness_corpus_claim_run_row_count / shard_a_row_count / shard_b_row_count (corpus runner)
• ci_runner_node_frontier_claim_run_row_count (affected_set runner)
• affected_testgen_claim_run_row_count (affected_testgen runner)
• claim_witness_corpus_claim_run_rows — the unconsumed, stale-by-one "all 54" aggregate (the gate reads the shard lists, never this; chore(ci): retire redundant *_row_count cross-checks (merge-skew liability) #4959 retired its cross-check). Real count was 55; the datum still said "54".

These are the *_row_count datums #4959's commit message described removing but whose diff only touched the shell scripts — they were left dangling and unconsumed. The gates fold the typed shard List directly, so a dropped row simply isn't enrolled; a parallel count datum can't detect that and only drifts under concurrent merges. Teeth retained: per-row pass + perturb-RED + ExpectFail-drift + empty-roster fail-closed guards are untouched.

Also (unblock): dsl/std/compute_fabric.dag:541 declared data operator_fleet_runner_slot_total: Int = with the RHS on the next line; the parser rejects =-then-newline, failing closed on every gate that resolves a dsl/ closure (corpus/lens gates) on all PRs. Single-lined it. (Parent's #4960 deletes this dead datum + moves the file to dsl/product/compute_fabric.dag; whichever lands first heals main, delete wins on the trivial conflict.)

Verification

• gunbc compile --source-root src/v4 --target dag → exit 0 (whole-tree v4 parse).
• Grep: zero readers of every deleted symbol across .dag/.sh/.rs/.yml.
• scripts/v4-claim-witness-corpus-gate.sh --shard b locally → GATE EXIT 0, 35 rows matched (34 ExpectPass, 1 ExpectFail).
• The 8 CI reds are fleet-contention (sccache fail-closed-by-design + compile-admission concurrency-group cancellations) in the CI-Rework lane, not merit failures — confirmed with the parent.

Deliberately NOT in this PR (scoped follow-on)

The operator asked to also delete the individual corpus_ci_gate_row_* enrollment rows and let reflection discover them. Ground-truth discover_owned_data coverage proof: 34/54 are discoverable, 20 are NOT (9 dsl/test/claim/* rows are outside the discoverer's src/v4/test/claim scan-dir; 11 src/v4 rows — 7 timeseries, mvp1-emit-matches-serialize, sg2-fold-row-delimiter, ts_effect_io_emit, spice — are registered in the TestClaim = BoolWitnessClaim { witness: BoolWitness { holds: fn() } } shape the discoverer's transport projection doesn't capture). Deleting those 20 would silently drop coverage (fabricated-green), which the fail-closed bar forbids. Safely deleting the rows requires: extend discovery scan to dsl/, make the 11 src/v4 witnesses discoverable, handle the 1 ExpectFail residual (#4801), and rewire the gate's enrollment to the discovery manifest. Tracked as the discovery-rewire follow-on.