hypertwist/docs/MODEL_B_SOURCE_ACCESS_BOUNDARY.md
2026-05-12 20:27:43 +02:00

6.2 KiB

Model B Source Access Boundary

Created on 2026-04-23

Purpose:

  • make the clean-room Model B boundary explicit for HyperTwist work
  • prevent accidental reading of restrictive mirrors or unrelated sensitive ops material

Allowed for clean-room Model B

  • C:\HyperTwist\UnrealHyperTwist\Source\UnrealHyperTwist
  • C:\Workspaces\HyperTwist\clean-room-specs\*
  • C:\Workspaces\HyperTwist\repos.manifest.json
  • curated first-party docs in:
    • C:\HyperTwist\docs\MODEL_B_SOURCE_ACCESS_BOUNDARY.md
    • C:\HyperTwist\docs\HT_REPO_INCORPORATION_AUDIT_2026-05-11.md
    • C:\HyperTwist\docs\HYPERTWIST_CANONICAL_RESTART_RECONCILIATION_2026-05-12.md
    • C:\HyperTwist\docs\HYPERTWIST_REPO_EVALUATION_RESET_AND_IMPLEMENTATION_SCHEDULE_2026-05-11.md
    • C:\HyperTwist\docs\HYPERTWIST_REPO_STATE_BOARD_2026-05-11.md
    • C:\HyperTwist\docs\REPO_LICENSE_TRACKING.md
    • C:\HyperTwist\docs\arch\HYPERTWIST_IMPORTED_GENERATED_MODE_EXECUTOR_PACKET_2026-05-05.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\README.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\AGENTS.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\API.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\ARCHITECTURE.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\DEVELOPMENT.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\LICENSETRACKING.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\PRD.md
    • C:\HyperTwist\docs\v6_5_deep_manual_pack\HyperTwist\ROADMAP.md
  • curated safe parse/workbook docs in:
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\06-project-agnostic-repo-evaluation-modus-operandi.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\07-benchmark-oracle-vs-clean-room-implementation.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\14-clean-room-model-a-model-b-prompts.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\88-roadmap-implementation-workbook.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\89-phase-0-and-phase-1-bootstrap-packet.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\90-clean-room-safe-project-context.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\122-hypertwist-phase-0r-restart-authority-and-sequencing.md
  • first-party staging docs and acceptance materials in:
    • C:\Workspaces\HyperTwist\implementation-workspaces\scratch\roadmap-bootstrap\docs\contracts\*
    • C:\Workspaces\HyperTwist\implementation-workspaces\scratch\roadmap-bootstrap\docs\architecture\*
  • first-party acceptance tests and public compatibility requirements

Forbidden for clean-room Model B

  • C:\Workspaces\HyperTwist\mirrors\restrictive\*
  • source-backed restrictive dossiers in:
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\42-kash-cubedesk-clean-room-dossier.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\43-onionhoney-roux-trainers-clean-room-dossier.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\44-cubing-alg-js-clean-room-dossier.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\45-cubing-twisty-js-clean-room-dossier.md
    • C:\visual_studio_solutions\multi_project\GPT 5.4 HyperTwist parse\46-hactarce-2x2x2x2-scrambler-clean-room-dossier.md
  • mixed-license restricted subtrees inside otherwise partially permissive repos, for example:
    • C:\Workspaces\HyperTwist\mirrors\restrictive\PostHog\posthog\ee\*
    • C:\Workspaces\HyperTwist\mirrors\restrictive\screenpipe\screenpipe\ee\*
  • copied sensitive ops references such as:
    • C:\HyperTwist\docs\refs\FORGEJO_WOODPECKER_VPS_A_TO_Z_SENSITIVE_RUNBOOK.md
  • secret-bearing or admin-bearing docs such as:
    • C:\HyperTwist\docs\refs\ADMIN_CREDENTIALS.md
    • C:\HyperTwist\docs\refs\AUTH_CONFIGURATION_GUIDE.md
    • C:\HyperTwist\docs\refs\OAUTH_ACTIVATION.md
    • C:\HyperTwist\docs\refs\OVERLEAF_ADMIN_SETUP.md
  • unrelated product ops folders, SSH material, or secret-bearing runbooks

Operational rule

  • if a file exists only because Model A inspected a restrictive repo, Model B must not read it unless it was explicitly scrubbed into clean-room-specs
  • benchmark oracles remain source-forbidden to Model B
  • this rule still applies when the upstream repo says all rights reserved, has no explicit license, or contains contradictory license notices; those repos may still inform Model A, but never become direct Model B source input
  • if a repo mixes permissive and restricted trees, default to forbidding the whole mirrored repo to Model B unless a narrower allowlist has been explicitly created in writing
  • for the imported generated-mode executor packet, Model B should implement from owned HyperTwistTraining runtime/request contracts rather than from donor-facing raw extract files or bundled intake JSON unless those assets were separately scrubbed and allowlisted

Current topology default

HyperTwist currently defaults to:

  • Model A
  • Model B as both clean implementer and clean integrator

The optional third clean mainline integrator instance remains documented, but it is not the default.

Current verified restrictive implementation truth

At the moment, HyperTwist should treat only one restrictive repo as both:

  • properly governed through clean-room separation
  • and already implemented/live in owned first-party surfaces

That repo is:

  • onionhoney/roux-trainers

Implications for Model B:

  • Model B may work with the already-landed first-party Unreal outputs and scrubbed clean-room specs for that lane
  • Model B must still not read the restrictive mirror itself
  • do not describe other restrictive HyperTwist repos as already landed unless the same level of clean-room and live-evidence closure is explicitly documented
  • for future restrictive-lane widening, wait for Phase 0R retention and Phase 1R handoff refresh before selecting the next clean-room target

Brownan / Oracle clarification

Brownan's solver mirror is allowed for Model A research and oracle derivation only. It does not justify Model B access to any VPS runbook, CI document, SSH key, or other unrelated operations material.