v0.2.1 — npm distribution (dsh-harness2-*)

v0.2.1 — npm distribution for Round-6

This patch release packages the Round-6 prototype under discoverable npm names.

npm

• dsh-harness2-task-recursion@0.2.1
• dsh-harness2-meta-recursion@0.2.1

dsh plugin --profile <name> add dsh-harness2-task-recursion
dsh plugin --profile <name> add dsh-harness2-meta-recursion

The plugin export name and profile row ids remain harness2-task-recursion /
harness2-meta-recursion; only the npm distribution names changed.

Verification

• npm run verify — 52/52 (task 31, meta 16, loader 5)
• npm run e2e — PASS (2 real headless sessions; PR-5 compose is diagnostic only)
• npm run stage0 — v3 archive 20/20 (gate falsification)
• npm run r6:verify — PASS (13 assertions)
• npm run stage0:r6 / stage0:r6:positive — PASS (negative 0 applied, positive 20/20 applied)
• npm pack / npm publish --dry-run — PASS

Known boundaries (unchanged)

• pr4:integration is fake-matrix calibration, not real-provider efficacy evidence.
• Production apply requires a real-provider accepted evaluation; providerClass is locally self-declared.
• HMR reload is pending-only; probe child has no sandbox; legacy experience is quarantined, not migrated.

v0.2.0 — Round-6 prototype (Gate-v4 + offline apply/eval/meta)

Round-6 prototype — offline profile-patch closed loop + Gate-v4 hardening

This release freezes the Round-6 contract implementation (PR-1 → PR-5.1) as a prototype, not a proven self-improving harness.

Included

• Gate-v4 admission: admissibility, per-candidate matched controls, deadband, token guard, audit-consistent receipts.
• Edit-body sidecar (harness2-edit/v1): content-addressed, atomic, 0o600, GC.
• Offline applier (tools/harness2-apply): plan → preflight → human review → atomic write → verify → rollback; refuses desktop profiles.
• Gate-B evaluation (tools/harness2-eval, OEP/v1 C/R/I/T): pre-registered holdout, exact McNemar, real-provider-only apply, forged-report rejection.
• Meta verified-experience: learns only from real-provider accepted/rejected evaluations with provenance; legacy experience quarantined by default; compose/* is diagnostic only.

Verification (local, 2026-10-21)

• npm run verify — 52/52
• npm run verify:task 31/31, verify:meta 16/16, verify:loader 5/5
• verify:apply 9/9, verify:eval 6/6
• npm run r6:verify PASS (13 gate assertions + spike)
• pr3:integration / pr4:integration PASS
• stage0 v3 archived 20/20 (GATE INVALID — falsification archive, not production evidence)
• stage0:r6 negative PASS (0 applied / 20 rejected), stage0:r6:positive PASS (20/20 applied)

Important boundaries

• No real-provider efficacy evidence is included. pr4:integration is a fake-matrix calibration run. Production apply requires a providerClass:'real' accepted evaluation report.
• providerClass is self-declared locally; the tooling cannot cryptographically prove a report came from a real provider.
• Real runner token/step metrics are estimates.
• Legacy experience is filtered, not migrated; HMR is only marked pending; probe children have no sandbox.
• Packages remain private/unpublished; install from source or GitHub spec. No npm publish in this release.

Full release review: docs/round-6-final-review.md.