Phase 2: lock and reproduce the scientific analysis environment #651

Merged
erikinkinen merged 2 commits from phase2/p2-14-a-analysis-environment into main 2026-10-21 01:32:30 +02:00
Owner

Task

Closes #116 (P2-14-A).

Summary

Freeze the established protocol 2.2 analysis environment as a verified
Linux/amd64 image. The declared .venv-research/bin/python path launches that
image, so analysis uses the recorded interpreter, package artifacts and thread
settings. Clean installation, image export and restoration have executable gates.

Scope

Included

  • Python 3.14.8, the six exact primary package versions, all 35 established
    distribution pins, wheel hashes/URLs, and the identified official base image.
  • A hash-required offline image recipe; closed lock/recipe checks; installed-file,
    numerical/import/rendering probes; a managed launcher preserving stdin and cwd.
  • OCI export verification of blob hashes and decompressed filesystem layers,
    separate build/export manifest identities, reproduction documentation and
    hashed acceptance evidence.

Explicitly excluded

  • Statistical-method implementation or inference, owned by P2-14-B through E
    and the separate primary-bound leaves.
  • Pilot/production data, manuscript findings and modeled recovery evidence.

Design intent

Preserve the existing software versions and record every installation input.
Use a pinned base manifest, no network during installation, no layer reuse,
fixed build timestamp and a complete hash-required wheel inventory. Reproduction
must match the frozen configuration/rootfs identity and runtime report.

The archive manifest differs from the build-store manifest because export
compresses layers. Both are recorded and bind the same image configuration and
filesystem. The canonical container_digest uses that common identity. A fresh
separate VFS storage root verified actual archive restoration.

The sole 4 KB vendored wheel, autograd-gamma 0.5.0, is retained from the established
environment. Its Python module and included MIT license match the SHA-identified
upstream source. Its frozen bytes are the install input; a byte-identical new
source-to-wheel build is not claimed.

Phase discipline

Protocol 2.2 / backlog 2.2.1; prerequisite #84 is closed. This is a prospective
software lock before pilots or production analysis. Package or image changes
require the governing change-control process. Environment smoke fixtures do not
accept the later statistical methods or establish scientific results.

Verification

  • Signed implementation source 7eb6a62350d10234ea0271bbd473482700ca5b4d.
  • Three independent clean offline installs reproduced the same image identity
    and byte-identical runtime report: 35 distributions, 8670 installed-file hashes,
    fixed threads, numerical/survival/rendering/schema/SQLite smoke probes.
  • Actual OCI export verified seven blobs and five decompressed layers. Restoration
    into separate storage passed through the real verifier with an identical report.
  • The new CTest passed 29 targeted lock rejections, optimized-verifier rejection
    and four OCI identity-graph cases. Actual fetch, build, launcher stdin/cwd,
    export and restored-image commands all passed at the signed source.
  • All eight local hooks and the frozen workspace scope passed. C++ implementation
    is unchanged; the complete registered suite is 92 tests and remains required in
    all eight final-head Forgejo checks before the one-step checklist update/merge.
  • Downloaded and hash-verified acceptance evidence.
    Tracked records are under artifacts/backlog/2.2.1/P2-14-A/.

The large wheelhouse and OCI image are retained locally under
artifacts/environments/protocol-2.2/; their exact paths/sizes/hashes are in the
acceptance evidence. The attachment contains verification receipts, source/release
metadata and logs. It does not claim those large files were uploaded. The committed
recipe and fixed artifact sources provide clean reproduction.

# Task Closes #116 (P2-14-A). ## Summary Freeze the established protocol 2.2 analysis environment as a verified Linux/amd64 image. The declared `.venv-research/bin/python` path launches that image, so analysis uses the recorded interpreter, package artifacts and thread settings. Clean installation, image export and restoration have executable gates. ## Scope ### Included - Python 3.14.8, the six exact primary package versions, all 35 established distribution pins, wheel hashes/URLs, and the identified official base image. - A hash-required offline image recipe; closed lock/recipe checks; installed-file, numerical/import/rendering probes; a managed launcher preserving stdin and cwd. - OCI export verification of blob hashes and decompressed filesystem layers, separate build/export manifest identities, reproduction documentation and hashed acceptance evidence. ### Explicitly excluded - Statistical-method implementation or inference, owned by P2-14-B through E and the separate primary-bound leaves. - Pilot/production data, manuscript findings and modeled recovery evidence. ## Design intent Preserve the existing software versions and record every installation input. Use a pinned base manifest, no network during installation, no layer reuse, fixed build timestamp and a complete hash-required wheel inventory. Reproduction must match the frozen configuration/rootfs identity and runtime report. The archive manifest differs from the build-store manifest because export compresses layers. Both are recorded and bind the same image configuration and filesystem. The canonical `container_digest` uses that common identity. A fresh separate VFS storage root verified actual archive restoration. The sole 4 KB vendored wheel, autograd-gamma 0.5.0, is retained from the established environment. Its Python module and included MIT license match the SHA-identified upstream source. Its frozen bytes are the install input; a byte-identical new source-to-wheel build is not claimed. ## Phase discipline Protocol 2.2 / backlog 2.2.1; prerequisite #84 is closed. This is a prospective software lock before pilots or production analysis. Package or image changes require the governing change-control process. Environment smoke fixtures do not accept the later statistical methods or establish scientific results. ## Verification - Signed implementation source `7eb6a62350d10234ea0271bbd473482700ca5b4d`. - Three independent clean offline installs reproduced the same image identity and byte-identical runtime report: 35 distributions, 8670 installed-file hashes, fixed threads, numerical/survival/rendering/schema/SQLite smoke probes. - Actual OCI export verified seven blobs and five decompressed layers. Restoration into separate storage passed through the real verifier with an identical report. - The new CTest passed 29 targeted lock rejections, optimized-verifier rejection and four OCI identity-graph cases. Actual fetch, build, launcher stdin/cwd, export and restored-image commands all passed at the signed source. - All eight local hooks and the frozen workspace scope passed. C++ implementation is unchanged; the complete registered suite is 92 tests and remains required in all eight final-head Forgejo checks before the one-step checklist update/merge. - [Downloaded and hash-verified acceptance evidence](https://git.erikinkinen.fi/attachments/63165341-7efa-4822-b217-b4cf717f692f). Tracked records are under `artifacts/backlog/2.2.1/P2-14-A/`. The large wheelhouse and OCI image are retained locally under `artifacts/environments/protocol-2.2/`; their exact paths/sizes/hashes are in the acceptance evidence. The attachment contains verification receipts, source/release metadata and logs. It does not claim those large files were uploaded. The committed recipe and fixed artifact sources provide clean reproduction.
erikinkinen added this to the Phase 2 milestone 2026-10-21 01:18:04 +02:00
Phase 2: record verified analysis environment acceptance (#116)
All checks were successful
markdownlint / markdown-lint (pull_request) Successful in 17s
clang-format / check-format (pull_request) Successful in 35s
ci / smoke (pull_request) Successful in 9m38s
ci / sanitizers (pull_request) Successful in 13m54s
markdownlint / markdown-lint (push) Successful in 14s
clang-format / check-format (push) Successful in 27s
ci / smoke (push) Successful in 6m35s
ci / sanitizers (push) Successful in 10m27s
c23b658cd0
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erikinkinen/AES!651
No description provided.