Consolidate native qualification and harden evidence publication

[Taz33m]

Summary

Consolidates the verified qualification work and follow-up hardening on current main.

• Add a separate L3 book-replay contract, pinned NautilusTrader mirror adapter, deterministic campaign/reducer, and retained injected queue-priority control. This is not submitted-order CLOB qualification.
• Add the pinned native Go intrepidkarthi/orderbook adapter, scoped FIFO qualification, and retained unfillable-FOK contract difference (not an upstream bug claim).
• Harden Rust input quantity encoding, stdio shutdown, L3 no-clobber artifact publication, distribution privacy, release-helper virtual environments, and native CI path coverage.
• Preserve corrected experiment guards and the completed 24-run aggregate. Private gold, evaluator artifacts, provider transcripts, and local navigation material are not included.
• Record the independently reproducible onboarding packet and the invitation in Optional independent FIFO qualification rehearsal with Tracebook intrepidkarthi/orderbook#25. The invitation pins the earlier verified 55e3c4b snapshot; sending it is not evidence of independent adoption.

Local validation

• Python 3.11 final full suite: 612 passed, 1 skipped. Python 3.13: 608 passed, 1 skipped, 82.16% coverage before four final virtualenv tests; final release/helper subset also passed. Python 3.10/3.12 full baseline suites and subsequent added regressions passed.
• Formatting, lint, active-minor mypy, Bandit, compile and whitespace checks pass.
• Both Rust adapters pass native tests, Clippy, and their 5,000-event qualifications with unchanged IDs. Existing orderbook-rs 1,000-event FIFO-full campaign and historical numeric controls pass.
• Exact Go 1.23.5 tests, vet, native build, Linux cross-build and real-binary smoke pass; prior complete Go FIFO and FOK proofs retained.
• Pinned Nautilus fixed 17-event proof, 2,500-event positive campaign and negative control preserve IDs and reduced trace bytes.
• Both distributions pass privacy, metadata, sdist/wheel byte agreement, isolated installed qualification, resolver checks, command ownership/uninstall, and SHA-pinned public 0.5.0 uninstall-first migration.

Remote/release gate

Remote CI must pass on this PR, including the four supported Python minors and all affected native integrations. No package upload or public release is requested by this PR. The current public release remains 0.6.0; the new source features will be prepared as a distinct next version after these checks.

---

Summary by cubic

Consolidates two new native qualification surfaces and evidence hardening onto main, addressing the release-code findings from the PR #84 automated review.

New qualification

• Adds a pinned Go submitted-order CLOB adapter for intrepidkarthi/orderbook v0.26.0 with FIFO/pro-rata qualification and an explicit unfillable-FOK contract difference, not an upstream bug claim.
• Adds a separate l3-book-replay-v1 surface for L3 order-book mirrors that do not cross submitted orders, with a pinned NautilusTrader adapter and deterministic campaigns.
• Adds preregistered agent-qualification experiments (v1 and v2) with the corrected 24-run aggregate; the frozen skill improved the safe-pass rate by 33.3 points.
• Leaves private gold, transcripts, and provider material ignored under experiments/private/ and defers the frozen research findings to issue 85.

Hardening

• Rust adapters encode quantities as exact 12-place integer lots and reject out-of-domain inputs before submission instead of pre-rounding.
• Decimal probe normalization no longer silently rounds exact values.
• L3 report and reduced-trace publication is no-clobber: destinations reserved, artifacts staged, then published together.
• Conformance peers must exit within the timeout after a complete frame, and release helpers preserve platform venv symlink defaults.
• Staging-identity failures report the possible residue path and skipped cleanup, preserve unverified entries, and exit before starting a candidate.
• Both distributions are checked for private paths and stale SOURCES.txt inventories.
• The dashboard requires the optional dashboard dependencies installed before starting.

^{Written for commit 4ba428e. Summary will update on new commits.}

[cubic-dev-ai]

All reported issues were addressed across 92 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[Taz33m]

Addressed the 17 release-code findings in e65416a, with regression tests. The full per-comment disposition is in docs/pr-84-review-disposition.md.

Local verification: 651 Python 3.13 tests passed, one skipped, 82.38% coverage; focused Python 3.10 tests passed; formatting, lint, type/security checks, pinned Rust 1.88 and Go 1.23.5 checks, package privacy, and sdist/wheel agreement passed. Four 5,000-event native qualifications retained their exact IDs. The Go FOK negative result and Nautilus positive/negative campaigns also retained their expected identities and reduced trace. One earlier full Python run hit an existing two-second subprocess startup timeout during concurrent builds; the isolated test and subsequent full confirmation runs passed without changing that timeout.

The 13 historical/frozen research findings remain open for a separate research disposition. No experiment runner, frozen input, retained evaluator artifact, or recorded qualification fixture was changed. No completed provider turn was retried. Passing release checks is not a reason to dismiss those research comments or bypass the merge gate.

This update does not merge the PR, publish a release, resend the maintainer invitation, or change its immutable source/wheel references. Remote checks must be assessed on the new SHA.

[cubic-dev-ai]

All reported issues were addressed across 28 files (changes from recent commits).

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 8 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Re-trigger cubic</sub>