Skip to main content

AI Cloud consumer gap audit

Date: 2026-07-11

Purpose​

AI Cloud is Fairway's largest active consumer. This audit converts recurring consumer-side coordination friction into Fairway product requirements. It does not treat old AI Cloud failures as current product failures or authorize changes to AI Cloud task state.

Evidence reviewed​

  • AI Cloud active reconciliation: clean.
  • AI Cloud ready queue and stale wait projection.
  • Review notification mapping failures for product and app-developer.
  • audit failure-routing, which currently reports 932 failed evidence rows and 573 missing follow-ups, including historical and closeout-command noise.
  • automation candidates, including repeated CI, UAT, docs, security, and notification patterns.
  • AI Cloud git status, where managed binaries under .fairway/bin appear as untracked project files.
  • AI Cloud pinned Fairway v0.1.11 binary versus newer source capabilities.

Product gaps​

Reviewer mapping fails too late​

Active tasks can name domains such as product and app-developer that have no configured reviewer route. Fairway discovers this only after a wait becomes notification_failed with mapping_required.

FW-297 adds a preflight that checks domains used by active tasks against role, route, and provider-target configuration before work reaches review wait.

Historical waits remain expensive to triage​

The wait surface correctly preserves immutable history, but old review and completion-handback rows remain operationally noisy. Resolving them one by one is disproportionate and can obscure current blockers.

FW-298 adds lifecycle-aware projection and bounded bulk acknowledgement or supersede decisions. It must preserve the underlying facts and preview every mutation.

Failure routing overstates actionable debt​

The current AI Cloud report treats some passing closeout commands, superseded evidence, and failures with existing follow-ups as new missing-gate findings. The resulting volume is too high to use as an operator queue.

FW-299 makes classification aware of task state, later accepted evidence, supersede facts, and existing follow-up tasks. Raw evidence remains available; only the actionable projection changes.

Managed binaries dirty consumer repositories​

AI Cloud stores release binaries under .fairway/bin, which appears as untracked source unless every consumer adds a local ignore rule. Runtime binary retention should not become repository metadata maintenance.

FW-300 moves the default to a user/OS cache or installs an explicit ignore rule while retaining exact version, path, upgrade, rollback, and cleanup readback.

Consumer capability drift is not explicit​

AI Cloud currently pins Fairway v0.1.11 while current source has newer commands. Operators can inspect versions, but there is no single report that says whether the pinned and running binaries satisfy the project's required capabilities.

FW-301 adds a non-mutating capability/minimum-version readiness report. It must not upgrade binaries automatically.

Existing capabilities that already cover observed needs​

The audit did not create duplicate tasks for these areas:

  • repeated command detection is already covered by automation candidates;
  • common-path command reduction is covered by FW-290 through FW-295;
  • CI and external utility monitoring already have watcher/session adapters;
  • completion handback supersede and generic wait acknowledgement already exist;
  • cross-project read-only reporting and shared-team API/store work already have implemented or reviewed tracks.
  1. FW-299: reduce false-positive operational debt first.
  2. FW-297: prevent new unroutable review waits.
  3. FW-298: clean historical wait projection with immutable decisions.
  4. FW-300: remove consumer worktree pollution.
  5. FW-301: make binary/capability drift explicit.

This epic is separate from the common-path epic. FW-290 reduces actions in a normal task lifecycle; FW-296 improves the accuracy and maintainability of a large consumer deployment.