Raw view: Architecture Review of an ADR

No invisible, zero-width, or bidi-control characters detected.

Review the Architecture Decision Record below as a staff engineer asked to approve it. Your job is to judge the quality of the decision and the record, not to redesign the system. A review that says "I would have chosen differently" without showing where the ADR's own reasoning fails is not useful.

<adr>
{{ADR}}
</adr>

<system_context>
{{SYSTEM_CONTEXT}}
</system_context>

<org_constraints>
{{ORG_CONSTRAINTS}}
</org_constraints>

<reviewer_concerns>
{{REVIEWER_CONCERNS}}
</reviewer_concerns>

<system_context> describes the surrounding system today. <org_constraints> may include team size, approved technologies, budget, compliance, and existing platform commitments. <reviewer_concerns> may be empty; if present, address each one explicitly by name in your output.

Review procedure:

1. Quote the decision statement verbatim. If the ADR has no single sentence that states the decision, that is the first finding.
2. Test the context section: could an engineer in two years, who has never met the authors, reconstruct why this was decided? List facts that are asserted without evidence (load numbers, cost figures, "the team is familiar with X") and facts that are missing but load-bearing.
3. Test the alternatives: for each one, ask whether a competent team might actually have picked it. Strawman alternatives (obviously bad options included to make the choice look inevitable) are a finding. A missing alternative that <system_context> or <org_constraints> makes obvious is a finding.
4. Test the consequences: are negative consequences stated with the same specificity as positive ones? Look for the costs the ADR does not mention: operational burden, migration, lock-in, hiring, the next decision this one forecloses.
5. Check fit with <org_constraints>. Each constraint is either honored, explicitly traded off with a reason, or silently violated. Silent violations are the most important finding type.
6. Check reversibility: does the ADR say what would cause it to be revisited, and how expensive reversal would be? If not, estimate both yourself and mark them as your estimate.
7. Form a verdict on the decision itself only after steps 1-6. If you disagree, show which step exposed the weakness; if you agree, say what would have to change for you to disagree.

Separate observations about the record (clarity, completeness) from observations about the decision (is it right). Both matter, but the author needs to know which is which.

## Output format

**Decision as stated** - the verbatim quote, or "No single decision statement found."

**Verdict** - exactly one of: `Approve`, `Approve with required edits`, `Revise and resubmit`, `Reject`. One sentence on the deciding factor.

**Findings on the decision** - numbered list. Each: the weakness, which review step exposed it, evidence from <adr> or the inputs, and what would resolve it (a number to measure, an alternative to evaluate, a consequence to acknowledge).

**Findings on the record** - numbered list of clarity and completeness gaps, each with the exact edit to make.

**Constraint check** - table: constraint from <org_constraints>, status (`honored` / `traded off` / `violated` / `not addressed`), note.

**Reversibility** - two sentences: the trigger for revisiting and the estimated cost of reversal, marked `[from ADR]` or `[reviewer estimate]`.

**Response to reviewer concerns** - only if <reviewer_concerns> is non-empty: each concern, addressed in 1-3 sentences.