diff --git a/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/clarify/clarify.md b/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/clarify/clarify.md
index adf17484..fc648c42 100644
--- a/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/clarify/clarify.md
+++ b/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/clarify/clarify.md
@@ -472,13 +472,39 @@ _Present each question's options via the structured question tool (e.g. AskUserQ
 - Q3 defaults to B ($1K–$5K) with a report caveat that spend was not confirmed.
 - **Q27 must not be silently defaulted in wizard mode** when Category H fired — if the user skips it, record `startup_program_status: "unknown"` with `chosen_by: "default"` and `source: "default:Q27"`; downstream artifacts must use neutral Activate copy (both tiers, no "your status: eligible_*").
 - Q3.5, Q23–Q25, and unresolved multi-instance conflicts fall back to their documented defaults (Q3.5 → E; Q23 → framework-based; Q24 → session; Q25 → medium; conflicts → most conservative posture) with `chosen_by: "default"`.
-  Then skip to Category E opt-in, then Step 5.
+  Then run the **Answer Recap** (below) with the defaulted rows included, then Category E opt-in, then Step 5.
 
 **Interpret answers** using the interpret rules in the category files. Apply early-exit rules triggered by answers (e.g., Q5 correction to multi-cloud → `compute: "eks"`, Q8 → N/A).
 
+### Answer Recap (Check Your Answers)
+
+After the last essential batch is answered and interpreted (and after Q27 when it fired), play back what was recorded — the essentials are the only answers in the flow the user states rather than confirms, and shorthand ("1A 2C") plus plain-word answers pass through an interpretation step the user never sees. One compact table, questions asked in Step 4 only (sheet rows were already confirmed at Step 2.5 — do not repeat them):
+
+```
+### What I recorded — last check before design
+
+| Question | Your answer | Recorded as |
+| -------- | ----------- | ----------- |
+| Compliance (Q2) | "2A" | None |
+| Cutover window (Q7) | "monthly is fine" | Monthly maintenance window |
+| AI spend (Q15) | "about a grand" | $500–$2K/month |
+
+Anything wrong? Name it ("cutover: weekly") — or say "looks good" and I'll proceed to design.
+```
+
+**Always wait.** Present the recap and **wait for the user's response** ("looks good" or a correction) before continuing — interpretation is exactly where a misread silently becomes a design constraint, and the batch opener invites shorthand and plain-word answers, so nearly every real run involves interpretation. One extra turn at the single highest-stakes commit in the flow is the intended cost. Corrections use the Step 2.5 override grammar and set `chosen_by: "user"`.
+
+**Sequence (fixed — do not reorder or combine):**
+
+1. Answer Recap → wait for the user's response
+2. Category E opt-in (if `billing-profile.json` exists) → wait
+3. Step 5 — write `preferences.json`
+
+**"Use defaults for the rest" still gets a recap:** when the user defaults remaining questions, include those rows with `(default applied)` in the "Your answer" column and the defaulted value in "Recorded as" — bulk-defaulting is the highest-risk interpretation of all, and this is the user's one chance to see what "the rest" actually meant. Then wait as above.
+
 ### Full Flow variant ("ask me everything")
 
-When the user opted out of the wizard, run the progressive-batch flow: present ALL active questions (no dispositions) in up to three batches — Strategic (Q1–Q7, minus Q4), Infrastructure (Q8–Q13b incl. Q11b Graviton + Category B), AI (Q14–Q27, Q23–Q26 only if agentic) — writing `preferences-draft.json` between batches with `metadata.batches_completed` / `metadata.batches_remaining` (values: `"strategic"`, `"infrastructure"`, `"ai"`). Per-question skip and "use defaults for the rest" behave as documented. Set `metadata.clarify_mode: "full"`.
+When the user opted out of the wizard, run the progressive-batch flow: present ALL active questions (no dispositions) in up to three batches — Strategic (Q1–Q7, minus Q4), Infrastructure (Q8–Q13b incl. Q11b Graviton + Category B), AI (Q14–Q27, Q23–Q26 only if agentic) — writing `preferences-draft.json` between batches with `metadata.batches_completed` / `metadata.batches_remaining` (values: `"strategic"`, `"infrastructure"`, `"ai"`). Per-question skip and "use defaults for the rest" behave as documented. Set `metadata.clarify_mode: "full"`. The **Answer Recap** above runs after the final batch here too (all answered questions as rows; skipped/defaulted ones summarized in one line, not per-row — intentional compression so a 20+ question full-flow recap stays scannable; the wizard already shows every defaulted essential as a `(default applied)` row because that set is small. Do not silently drop answered questions into the one-liner).
 
 ### Category E Opt-In
 
@@ -803,6 +829,7 @@ Before handing off to Design:
 - [ ] In wizard mode, the Step 2.5 Assumption Sheet was shown (detected + assumed sections) and the user responded before any essential question was asked
 - [ ] Every constraint with `chosen_by: "extracted"` or `chosen_by: "default"` has a `source` field with the correct prefix (`terraform:`, `billing:`, `inventory:`, `ai-profile:`, or `default:<Qid>`)
 - [ ] Essential questions (Q2, Q7, and conditional Q1/Q3/Q3.5/**Q27**/Q15/Q23–Q25/conflicts) were asked, answered, or explicitly defaulted via "use defaults for the rest"
+- [ ] The Answer Recap was shown after the final Step 4 batch (including defaulted rows on "use defaults for the rest") and the user responded to it before Category E / `preferences.json`
 - [ ] If `bigquery_present` was **true**, the Step 4 BigQuery specialist advisory was shown before questions — **or**, if Step 0 option A (reuse preferences), the same advisory was shown after BigQuery detection
 - [ ] `preferences.json` written to `$MIGRATION_DIR/`
 - [ ] `design_constraints.target_region` is populated with `value` and `chosen_by`
diff --git a/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/discover/discover.md b/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/discover/discover.md
index 334ecc6f..2ceff7cc 100644
--- a/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/discover/discover.md
+++ b/migrate/plugins/migration-to-aws/skills/gcp-to-aws/references/phases/discover/discover.md
@@ -14,6 +14,12 @@ Multiple artifacts can be produced in a single run — they are not mutually exc
 
 ## Step 0: Initialize Migration State
 
+0. **Orientation (fresh runs only):** When this is a new migration (no existing `.migration/` runs, or the user chooses Fresh below), open your first response with this block — before any scanning output. Skip it entirely when resuming an existing run (the breadcrumb in the resume message covers position).
+
+   > **Here's how this works.** Six phases: I scan your Terraform, app code, or billing data; confirm a few assumptions with you; then design the AWS architecture, price it, and generate everything. Your part is small — most answers come from your own files; you'll typically confirm one summary sheet and answer 2–7 questions. At the end you get a migrate-or-stay recommendation with costs, Terraform for the in-scope workloads, and step-by-step migration scripts. You can stop at any point — progress is saved and I'll resume where you left off.
+
+   Do not pad it, restate it later, or block on it — continue directly into discovery in the same turn.
+
 1. Check for existing `.migration/` directory at the project root.
    - **If existing runs found:** List them with their phase status and ask:
      - `[A] Resume: Continue with [latest run]`