Harden Isogloss architecture, performance, and security

[owengregson]

Summary

• hardens the exact PR 6 head across correctness, robustness, performance, filesystem safety, protocol binding, browser isolation, and CI supply-chain controls
• replaces per-fragment runtime functions with scalar straight-line emission and reentrancy-safe pooled wrappers
• makes identifier preprocessing AST-native and bounded, indexes pure-region and call-graph analyses, and adds enforced performance budgets
• adds fail-closed resource limits and exactness checks across source transforms, BPRF generation, planning, and browser worker inputs
• binds custody signatures to complete requests, lineage, covers, and exact target representations; uses hiding commitments for owner-secret material
• adds contained two-phase filesystem protection, symlink/hardlink and physical-alias rejection, staged publication, rollback, and whole-tree replacement
• keeps the nonlocal planner unpublished until compile, lowering, and evidence can be emitted atomically and bound to the exact contract digest
• pins all workflow actions and runtimes, consolidates Pages deployment, refreshes vulnerable transitive dependencies, and removes the stale npm lockfile
• adds a 166-item first-principles performance and security backlog in docs/performance-security-hardening-backlog.md

Review baseline

This branch starts at PR 6 head commit 02c481d. The changes were driven by independent correctness/robustness, performance, and security audits, followed by adversarial post-fix verification.

Validation

• bun run typecheck
• bun run test: 191 passed, 0 failed, 34,449 assertions
• bun run build: library, browser worker, and Next.js static site
• bun audit --production: no vulnerabilities
• bun install --frozen-lockfile --dry-run
• full benchmark budget: 15,220-byte protected output; 2.94 ms per 50,000 protected calls; checksum agreement
• YAML parse and git diff --check
• npm pack --dry-run: 40.2 KB package, 169.5 KB unpacked

Notable measured improvements

• pathological pure-region planning at 320 regions fell from roughly 4.55 s and 265 MB to 372 ms and 6.8 MB
• SCC inventory and call-risk aggregation now scale effectively linearly in measured 500/1,000/2,000-unit cases
• large-expression protected execution improved from about 0.56/3.20/10.07 microseconds per call to 0.135/0.669/1.47 at 20/80/160 operations
• published package output is substantially smaller after keeping experimental planner internals off the root export surface

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of lines (20,000). Try reducing the number of changed lines and requesting a review from Copilot again.

[github-actions]

Dependency Review

The following issues were found:

• ✅ 0 vulnerable package(s)
• ✅ 0 package(s) with incompatible licenses
• ✅ 0 package(s) with invalid SPDX license definitions
• ⚠️ 1 package(s) with unknown licenses.

See the Details below.

License Issues

apps/web/package.json

Package	Version	License	Issue Type
next	^15.5.21	Null	Unknown License

OpenSSF Scorecard

Package	Version	Score	Details
npm/next	^15.5.21	Unknown	Unknown

Scanned Files

• .github/workflows/deploy.yml
• apps/web/package.json
• packages/ruam/package-lock.json

[Copilot]

Copilot wasn't able to review this pull request because it exceeds the maximum number of lines (20,000). Try reducing the number of changed lines and requesting a review from Copilot again.