Skip to content

Latest commit

 

History

History
134 lines (111 loc) · 8.18 KB

VALIDATION.md

File metadata and controls

134 lines (111 loc) · 8.18 KB

0.6.0 scheduling and saved-draft review validation — 2026-10-11

Status: experimental. Fresh full Python suite: 193 passed, one opt-in live test skipped. Synthetic JXA logic: 77 checks passed. Native Foundation/CommonCrypto: four synthetic protocol cases passed without loading Mail. Diff integrity and publication privacy pattern scans are checked separately.

Seven new scheduling tests cover the stale-audit release/acquire interleaving, fresh evidence while claims are held, failed/unsafe rechecks, changed and unchanged reschedule/cancel targets, and multi-row cleanup preserving previously issued work. The exact interleaving failed before the fix and is rejected after it. These are synthetic state/claim tests, not native Send Later certification. Claims coordinate cooperating current-version CLI workers; they do not lock Mail against manual edits, other apps or later synchronization changes. Recheck visibly before acting.

Independent source review reproduced the race before the fix and found no additional material findings in the reviewed integration. A separate review of the fix confirmed closure and passed 69 synthetic scheduling/evolution tests. This scoped review is not a guarantee against remaining defects or live Mail behavior. The offline suite passed 191 tests with one live test skipped, excluding the two loopback HTTP cases from the full suite.

In separately authorized local work, one existing self-test draft was saved and its exact account-scoped saved RFC Message-ID was reviewed. Envelope fields and decoded attachment name/size/SHA-256 matched the expectations. Repeated source reads and a later review-digest recheck agreed. Stored IDs changed on save, so the review required refreshed identity rather than same-subject editor linkage.

The saved copy had HTML and an empty plain-text part, so the result remained needs_content_review; HTML rendering and the link to the open editor were not verified. An earlier separately approved send attempt stopped at pre-dispatch verification. Its unknown receipt and attempt guard were preserved byte-for-byte. No send was invoked by the saved-review workflow, and no delivery is certified. Private addresses, IDs, message bodies, attachment data and receipts are excluded from this publication. This one-machine observation is not cross-version coverage.

Historical 0.6.0 integration validation — 2026-10-10

Status: experimental. This public-source update combines local 0.5/0.6 development with the existing public 0.4.3 guard implementation. The earlier independent review below is historical; it does not cover this integration.

Check Fresh result Boundary
Full Python suite 166 passed; one opt-in live test skipped Synthetic fixtures, mocked mutations and a local fake HTTP server
Offline Python suite 164 passed; one opt-in live test skipped Excludes the two HTTP loopback tests
Fake JXA logic 47 checks passed Fake Mail objects; includes draft BCC/body/attachment drift, sender checks before draft updates and legacy send bypass rejection
Native Foundation/CommonCrypto protocol Four cases passed Real platform APIs against temporary synthetic state; Mail never loaded
Syntax, launcher, JSON capabilities, skill structure, diff integrity Passed Public source and manifest report 0.6.0

The direct-send workflow retains exact-digest preview confirmation, immutable attachment snapshots, durable intent reservations and backend claims. New operation receipts do not replace that guard. Tests also cover cross-campaign claims, pinned schedule targets, grouped partial reads, exact-account Sent observations, review-only follow-ups, render rollback and existing campaign review regressions. A historical Sent filename/count match remains labelled metadata_matched, but expected attachments without byte evidence now raise attachment_unverified and keep the audit incomplete.

No real Mail message was read, created, sent, scheduled, moved or deleted in this integration run. No real UBD endpoint was called. Live doctor/account checks and opt-in Mail tests were not run. In particular, the new draft readback guard has only synthetic coverage; native Mail property availability, signature/body transformation and asynchronous attachment behavior may block it. This release does not certify native sends, queue persistence or delivery. Publishing it does not update any existing local installation.

Reproduce with the commands in the historical record below, plus:

PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests -q

Keep APPLE_MAIL_CLI_LIVE_TESTS unset for the synthetic full suite. The offline runner disables it regardless. Privacy pattern scans checked the publication tree; they are not a complete secret-detection or security certification.

Historical 0.4.3 validation record

Status: experimental, with a passed independent review of the bounded source/synthetic scope. Validation performed on 2026-10-01 using macOS 26.6.2 arm64, Python 3.9.6 and Node 25.8.1. These versions are observations, not a cross-version support certification.

Check Recorded result Boundary
Guarded Python regressions 83 passed; one live test skipped Synthetic data; subprocess/network/real Mail store access blocked in the offline runner
Fake JXA logic 40 checks passed Existing Node runtime; fake Mail objects
Native Foundation/CommonCrypto protocol Four cases passed Actual JXA platform APIs; temporary private state; Mail never loaded
Independent additional scenarios 10 passed Includes eight concurrent confirmations: one backend invocation, seven blocked attempts
Syntax, launcher, JSON capabilities, patch integrity Passed CLI/backend/plugin agree on 0.4.3
Native read-only index smoke Compatible schema; bounded synthetic subject query succeeded with zero rows Existing authority; no body reads or raw rows retained

Review found no material P1/P2 findings in that scope. The earlier review's three campaign findings and one body-read finding had been fixed and rechecked in 0.4.2. This is a scoped review result, not a guarantee that defects cannot remain. Publication-only metadata/documentation edits do not change the reviewed runtime or tests.

The direct-send review exercised exact-intent binding, attachment byte drift, missing/private record checks, replay rejection, persistence failures, interruptions, unknown outcomes and concurrent reservation. A protocol fixture's accepted result means its guard/claim step passed; it does not mean Mail accepted or delivered a message. The backend success contract separately reports submission_accepted: true and delivery_confirmed: false.

Unverified: live account/sender resolution; actual drafts/replies, To/CC/BCC, body/signature transformations, attachment readback/asynchronous behavior, Mail send/queue/receipt, native mutations and Send Later UI; provider/localized mailbox mappings; denial/partial-failure recovery in real Mail; other macOS, Python or Node versions; CI; loopback HTTP and an external UBD OS deployment. No new Automation/FDA/Accessibility permission was granted. A read-only backend ping does not establish Automation authority.

The source candidate, original 24-file base and installed launcher were verified unchanged. Candidate privacy/provenance scans found no high-confidence secret or non-example address findings in the allowlisted package. Pattern scans are not a complete secret or authorship certification. Private mail, credentials, machine-specific source inventories and review logs are not public fixtures.

Reproduce the supported synthetic checks from the source directory:

PYTHONDONTWRITEBYTECODE=1 python3 scripts/test_offline.py
node tests/test_backend_quality.js
# Optional macOS-only native protocol fixture; does not load Mail:
PYTHONDONTWRITEBYTECODE=1 python3 scripts/test_native_send_protocol.py

The project offline runner excludes the HTTP loopback module and disables the opt-in live module; the stricter no-subprocess/no-socket/no-real-store harness used during review is private review tooling, not bundled here. The native protocol fixture launches JXA for synthetic platform checks. Do not enable live tests or real Mail mutations merely to reproduce this table.