Iteration 21 — Cycle 2 Resume: Harvest Feed Imported (68 Candidates, 6 Batches)

Iteration 21 — Cycle 2 Resume: Harvest Feed Imported CYCLE 2 · PR #591

In plain words: the loop is back after the operator merged cycle 1 (20 iterations → PR #578, squashed into main) and, in the meantime, ran a separate harvesting campaign that pre-screened 68 new candidate instruments covering every status boundary. This firing did the plumbing of the restart: verified the old campaign PR is merged, cut a fresh cycle-2 branch from current main, opened the cycle-2 campaign PR (#591), recorded the operator's directive word-for-word in the ledger, and imported the 68-candidate feed into the frontier — STABLE-boundary batches first (that is the stack's one vacant instrument slot), cheapest kills first, with the two e-value entries folded into the already-pending candidate-A entrance exam rather than tracked twice.

Preflight (A0): 2026-07-08 — load1 5.19/32c (≤24) · 22 GiB avail · CH + sidecar healthy → ALL PASS. No compute this firing (directive execution + docs). Zero ClickHouse, zero synthetic data.

What landed

Item	Value
Cycle-1 closure	PR #578 MERGED at 53751b29 (operator milestone); cycle-2 branch cut from main @ a0ba5dfe; cycle-2 campaign PR #591 opened (ledger row 27)
Tier-2 refill import	68 candidates, batches B-01..B-06 (STABLE ×2, FAIL, FRAGILE↔CONDITIONAL, EXPIRY, CONDITIONAL-scoping) → FRONTIER.md Tier 2; per-entry dossiers stay in the harvest campaign's CANDIDATE-CATALOG.md (read-only for this loop)
B-01 consumption order	H-016 seqICP → H-023 IAS → H-019 IPP → H-025 HSIC-X → H-017 StabReg → H-021 EILLS → H-018 Causal Dantzig → H-004 PCMCI+ · H-001/H-002 = candidate A, folded into the pending entrance exam

Item

Value

Cycle-1 closure

PR #578 MERGED at 53751b29 (operator milestone); cycle-2 branch cut from main @ a0ba5dfe; cycle-2 campaign PR #591 opened (ledger row 27)

Tier-2 refill import

68 candidates, batches B-01..B-06 (STABLE ×2, FAIL, FRAGILE↔CONDITIONAL, EXPIRY, CONDITIONAL-scoping) → FRONTIER.md Tier 2; per-entry dossiers stay in the harvest campaign's CANDIDATE-CATALOG.md (read-only for this loop)

B-01 consumption order

H-016 seqICP → H-023 IAS → H-019 IPP → H-025 HSIC-X → H-017 StabReg → H-021 EILLS → H-018 Causal Dantzig → H-004 PCMCI+ · H-001/H-002 = candidate A, folded into the pending entrance exam

▶ Next iteration

Next firing = the STABLE slot opens for business: the candidate-A entrance exam (permutation-calibrated e-value invariance test on the committed perfect-duplicate case — covers H-001/H-002), and if it completes inside the slice, H-016 seqICP's M2 begins (cheapest kill in B-01). Every B-01 candidate faces the same entrance exam before its ladder walk. Standing for the operator: the 73 grid-corroborated draft statuses (cycle-1 proposals, iters 11/13) and the Frontier #4 precondition review — both survive the cycle boundary untouched in the append-only ledger.

Iteration 21 · 2026-07-08 · cycle-2 resume (LEDGER row 27, PR #591, Tier-2 import) · append-only