Executed on Windows with WFL reporting 26.9.12, built from combined upstream
revision ae5395d9bd215d0d9fa1c039e666f03c8897cf88. Executable SHA-256:
76F612CA1BBAF4484B2CC2BDA86D876588BC765DE9C38B7DAA8951BAD143B911.
Application source baseline: Scriptorium
4ec5c88d9e4ae27041599ad8293a28130b56fbf2, extracted into ignored
target/http-baseline-4ec5c88. Both that revision and the current checkout pin
Scribe 93d62af5a6ed6c3ce257ef888107fc3ca1e2dc1d. Git archive provisioned the
baseline source; WFL created every actual site, configuration, database, upload,
request, assertion and cleanup operation.
Each suite used this command shape from G:/repos/Scriptorium:
G:/repos/wfl/target/codex-worktrees/integration/target/release/wfl.exe --test tests/integration/<suite>.test.wfl G:/repos/wfl/target/codex-worktrees/integration/target/release/wfl.exe G:/repos/Scriptorium/target/http-baseline-4ec5c88
| Suite | Passed | Retained local log |
|---|---|---|
| server-port | 3/3 | target/http-baseline-ports.log |
| authentication | 2/2 | target/http-baseline-authentication.log |
| content-users | 2/2 | target/http-baseline-content-users.log |
| media | 2/2 | target/http-baseline-media.log |
| throttle | 1/1 | target/http-baseline-throttle-green.log |
| recovery | 1/1 | target/http-baseline-recovery.log |
Total: 11/11. The media case exercises both configured and legacy storage
layouts. No assertions were removed or relaxed to obtain these results.
target/test-artifacts/http contained zero residual fixture directories after
the runs. Each WFL stop helper asserts its owned process is no longer running.
The first throttle run failed because the test reused login_cookie outside
the try scope that declared it. Moving the cookie/token declarations to the
test's enclosing scope fixed that fixture defect; its original log remains
target/http-baseline-throttle.log. This is not a production defect or Red
evidence for the application. Earlier parse/smoke attempts using separate,
incomplete prerequisite binaries could not resolve all new APIs and are not
counted as integration evidence.
The interpreter emits advisory undefined-action warnings for actions supplied by the included helper and unused-variable warnings for values read by test assertions. They remain visible in the logs; every suite exits zero and reports the passing counts above. A separate OS process-list audit was unavailable because Windows denied the read; cleanup evidence is the owned-process checks and empty fixture directory, not a claim of an independent OS-wide audit.
The same six suites ran against the updated application working tree on 2026-09-20 with the same combined runtime, omitting the baseline-source argument. All 11/11 passed on their first candidate run:
| Suite | Passed | Retained local log |
|---|---|---|
| server-port | 3/3 | target/http-candidate-ports.log |
| authentication | 2/2 | target/http-candidate-authentication.log |
| content-users | 2/2 | target/http-candidate-content-users.log |
| media | 2/2 | target/http-candidate-media.log |
| throttle | 1/1 | target/http-candidate-throttle.log |
| recovery | 1/1 | target/http-candidate-recovery.log |
These runs cover the first integrated ORM/migration adapters before the later
independent nullable-key and write-metadata compatibility fixes. Those fixes
have separate Red/Green tests in TestPrograms/db-review-contracts.test.wfl.
The application was still an uncommitted working tree, so this evidence is not
a claim that an immutable final application commit has passed. The complete
runner must also test the final proposed commit. Linux/nightly container
evidence and its immutable image digest remain separate required checks;
local Windows results do not establish those results.
The original three Python port cases map directly to the three WFL port cases, including installer status/content type/form/CSRF, configured/legacy/default ports and startup messages. The Python file was removed only after its WFL replacement passed against both baseline and the updated application.
An independent agent reviewed fixture process ownership and the stopped-site backup/restore protocol. It found no blocking source issue: unique owned directories, cleanup after failures, reaping before copying all data/sidecars, and HTTP plus SQLite checks after restore. This is technical review, not Maintainer approval or release acceptance.
legacy-upgrade.test.wfl passed 1/1 on 2026-09-20 with the same combined
runtime against the working candidate after the compatibility fixes. Log:
target/http-candidate-legacy-upgrade.log. This adds a twelfth HTTP case.
WFL creates the original seven-table schema without sessions.csrf_token or a
migration ledger, a synthetic Argon2 administrator, an installed flag, published
content, a live legacy session and an extension-owned payload. It copies a WFL
extension into the disposable site's documented seam and generates a configured
custom theme. Actual boot adopts the schema and creates extension boot markers.
HTTP proves both installer methods remain locked, the extension owns / before
stock dispatch, the custom theme renders extension and stock post routes, the
retained user can log in, and a legacy empty CSRF token cannot mutate settings.
A stopped/reaped restart preserves identical ledger rows/checksums/timestamps, retained extension payload and managed rows; only the extension's expected boot count increments. Integrity and foreign-key checks pass. The fixture has no requirement for the ignored Git-archive baseline and no non-WFL test logic.
Independent source review found no fixture ownership or application-boundary issue. It requested ordered ledger snapshots so map iteration order cannot cause false failures; the test now serializes ordered lists of explicit fields.
The original 16 MiB upload received an HTTP transport error on Linux in
Red run 35507634634, job 106070018120
at head 07e8adcb7785c168c37fd56cc35e88492809a820. The same media suite passed
2/2 on Windows. WFL rejects an oversized advertised Content-Length before
reading the body; closing an unread large upload can reset the connection before
the client observes 413. This was a fixture portability assumption, not evidence
that the upload was accepted. The test did not accept the transport error as a
passing result.
The corrected fixture asserts that the shipped configuration and normal fixture both retain the 10 MiB ceiling, then sets only its disposable site to 1 KiB. It sends a real 2 KiB multipart file and requires 413, zero media rows, zero upload files, and a subsequent authenticated GET returning 200. Existing configured and legacy upload workflows remain covered. No retry, skip, transport-error fallback, application change or runtime change was introduced. This does not claim that eager 16 MiB HTTP/1 uploads always expose a 413 response.
Green run 35508309315, job 106071763602
ran the actual media suite on Blacksmith Linux at
147b15c88c5fbd29e4826a32a0db5daf747ae59d: 2/2 passed. Its freshly resolved
official runtime was WFL 26.9.14, source
8d82d785ea59300834de1c48b04a7ba0e187a1cd, image
bsbyrdwfl/wfl@sha256:8498abec67995274cb4d6cc2ee9580be11f70e15af20497bb62e51d4b2058e3c.
Scribe remained 93d62af5a6ed6c3ce257ef888107fc3ca1e2dc1d. The same revised
suite passed 2/2 using the official Windows 26.9.14 runtime; retained local
log: target/media-probe/media-small-official-windows.log.