Fix nightly vlt canary: support vlt 1.3.0-1.3.7

[Mikola Lysenko (mikolalysenko)]

Problem

The nightly vlt patch compatibility canary has been red on main four nights in a row:

Night	Run	canary (ubuntu)
2026-10-04	37181534032	failure
2026-10-05	37265527469	failure
2026-10-06	37414450417	failure
2026-10-07	37572304168	failure

Every night the failing step is the same: Release and lockfileVersion watchdogs, Linux-only:

##[error]vlt 1.3.0 is published but neither supported nor excluded in docs/testing/vlt-compatibility.md (Releases)
... (same for 1.3.1 through 1.3.7)
vlt 1.3.7: lockfileVersion 1

The vlt@latest through every capstone step passed on ubuntu, macOS and Windows all four nights, and so did the macOS and Windows canary jobs. vlt is releasing fast: 1.3.0 came out 2026-09-28 and 1.3.7 on 2026-10-06. Nobody has added those versions to the Releases table, so the watchdog fails and the red nightly hides any real failure.

Root cause

The Releases table, and the leg manifest generated from it, stopped at 1.2.0. vlt published eight 1.3.x releases after that.

Fix

• docs/testing/vlt-compatibility.md: list 1.3.0 … 1.3.7 as supported, extend era F to 1.2.0 … 1.3.7, and say where the 1.3.x evidence comes from.
• crates/socket-patch-cli/tests/vlt-leg-manifest.json: regenerated with scripts/check-vlt-legs.py --derive. The only change is the 8 versions added to supported.
• scripts/vlt-historical-integrity.json: pins for 1.3.0 … 1.3.7. I downloaded each tarball, hashed it and checked the hash against the registry's dist.integrity. The existing test test_pinned_integrity_covers_every_supported_release requires a pin for every supported release.
• scripts/tests/test_backtest_harnesses.py: the "unlisted" example changes from 1.3.0 (now supported) to 9.9.9. The era test now also asserts that 1.3.7 maps to F. No assertion is weaker.

Proof

I ran all five capstones locally on Linux for every new release, exactly as the canary step runs them (<suite> vlt_pinned_matrix --ignored, then check-vlt-legs.py against the new manifest):

vlt	e2e_redirect_vlt_build	e2e_vendor_vlt_build	mode_migration_vlt	e2e_safety_vlt	e2e_vlt
1.3.0 … 1.3.7 (each)	38 passed	34 passed	14 passed	10 passed	9 passed

On all 40 runs check-vlt-legs printed ok … manifest diff clean. That includes the era assertion that each release writes lockfileVersion 1 with the tilde DepID grammar.

• The watchdog's unlisted computation over the 137 versions npm currently publishes goes from 8 releases (1.3.0…1.3.7) to [].
• python3 -B -m unittest discover -s scripts/tests: 254 tests OK. Before the test update, test_exclusions and test_pinned_integrity_covers_every_supported_release failed, as expected.
• check-vlt-legs.py --derive output matches the committed manifest byte for byte.

Where tests run

No test is removed or moved. This only makes the nightly canary pass again. The next vlt release will still trip the watchdog until it is qualified the same way, which is what the watchdog is for.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CvCSdYG1Nk8Ys2T2mZr9Z7

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 886dddb. Configure here.}

[Mikola Lysenko (mikolalysenko)]

coverage-merge (CI run 37688404448) was cancelled at its 15-minute timeout while its Install lcov step (sudo apt-get update && sudo apt-get install -y lcov) was still running, before any coverage file was downloaded or merged. The Ubuntu package mirror stalled. This is not caused by this PR: the diff only touches the vlt docs table, the generated leg manifest, the integrity pins and one Python harness test, none of which this job reads before it got stuck. The last 20 coverage-merge jobs that ran on CI completed this step. There is no fix to port. I'll re-run the job once when the rest of the run finishes.

---

Generated by Claude Code