Crussell/backend/handlers/payments/refunds.go

package payments

import (
	"context"
	"crypto/sha256"
	"database/sql"
	"errors"
	"fmt"
	"log"
	"log/slog"
	"math"
	"sort"
	"strconv"
	"strings"
	"sync"
	"time"

	"crussell/clock"
	"crussell/db"
	"crussell/internal/square"

	"github.com/jackc/pgx/v5"
)

// Refunds are exempt from buyer verification. Square's RefundPayment endpoint
// does not support 3DS/SCA verification tokens (verification is a charge-time
// concept — PSD2 SCA applies to the original payment capture, never to the
// money flowing back out), so no refund path carries a verification token and
// none ever will. The practical already-refunded signal is Square's
// REFUND_AMOUNT_INVALID error (PAYMENT_ALREADY_REFUNDED is no longer
// documented): on a genuinely-refunded payment that code is NOT a decline —
// the handler reconciles via an EXACT-amount COMPLETED-refund check
// (reconcileRefundAtSquareExact, see the A11 disambiguation in
// processChargeGroup / processManualPaymentGroup) and resolves the refund row
// to 'completed' instead of failing it, so the over-refund guard never
// re-issues money Square already returned.
//
// stalePendingRefundAge is the age guard for pending refunds: Square's
// idempotency-key retention is finite (~24h), so a pending refund older than
// this must be RECONCILED against Square (ListPaymentRefunds) before any
// re-issue — re-issuing with a key Square no longer retains would be treated
// as a NEW refund (double refund). A pending row older than
// stalePendingRefundAge that Square shows no COMPLETED refund for is marked
// 'failed' and surfaced for manual arrangement instead.
const stalePendingRefundAge = 23 * time.Hour

// stalePendingB1RefundAge is the age guard for a B1 sweep auto-refund of a
// replay-induced duplicate charge (sweepPendingB1Refunds) that Square left
// PENDING. FAILED/REJECTED are terminal refund states, but the webhook
// FAILED-refund reconciliation that would normally resolve them is OPTIONAL
// (README) — with it unconfigured, hasInFlightSweepDuplicateRefund blocks the
// stale sweeps from resolving the parent forever, no alert fires and every
// sweep run re-polls Square indefinitely. After a B1 refund has been pending
// this long the re-poll pass treats FAILED/REJECTED as terminal itself (failing
// the refund, resolving the parent and alerting), and a still-pending refund
// older than this is escalated to a deduped CRITICAL admin notification and no
// longer re-polled. Never marks anything completed without a COMPLETED Square
// refund.
const stalePendingB1RefundAge = 48 * time.Hour

// maxManualRefundAttempts caps retry attempts for a refund stuck on a
// decline/ambiguous outcome before it is resolved terminally. Both the manual
// sweep (processManualPaymentGroup / resolveManualRefundAtCap) and the
// cancellation charge-group sweeps share the same cap: rows are retried while
// refund_attempts < maxManualRefundAttempts, and at the cap a refund is
// reconciled at Square FIRST (never marked failed while the money state is
// unknown — that would let the over-refund guard exclude money that actually
// left the business), then resolved to 'completed' or 'failed' + admin
// notification. The SQL literals below are formatted with this constant so
// the DB filter and the Go cap can never drift apart.
const maxManualRefundAttempts = 3

// maxConsecutiveReconcileFailures is the number of consecutive cap-time
// reconcile failures (each causing a re-arm under the attempt cap) before a
// refund row is surfaced in the admin notification centre. A reconcile failure
// is an UNKNOWN money state — the row is never marked 'failed' on it — but
// silently re-arming forever would oscillate the row between the cap and
// cap-1 indefinitely with zero admin visibility (A5b/A5c). After this many
// consecutive failures a deduped 'critical_payment_log' admin notification is
// inserted so the owner learns the reconcile is hard-failing.
const maxConsecutiveReconcileFailures = 5

// maxTrackedReconcileFailures caps the in-memory consecutive-failure counter
// so the map never grows past a sane value AND the ==maxConsecutiveReconcileFailures
// notification check can never be skipped by an odd increment pattern (A5): a
// row whose counter is already at the cap stops growing, but the notification
// fired when it first crossed maxConsecutiveReconcileFailures and is deduped by
// the NOT EXISTS guard, so capping loses no visibility.
const maxTrackedReconcileFailures = 10

// manualReconcileFailures counts consecutive cap-time reconcile failures per
// refund row (keyed by refunds.id). resolveManualRefundAtCap and the
// charge-group cap path re-arm a row under the attempt cap on a reconcile
// error so the next sweep re-picks it; without a counter that re-arm loops
// forever with no notification. The counter is in-memory (no schema change —
// the schema is single-source and pre-launch, no ALTERs): it counts
// CONSECUTIVE failures, is reset whenever a reconcile succeeds (the row is
// resolved completed/failed), and after maxConsecutiveReconcileFailures
// failures triggers a deduped critical_payment_log admin notification. A
// process restart merely resets the counter, deferring the notification by a
// few sweeps — never suppressing it (the row stays pending and keeps being
// re-reconciled, so the notification eventually fires).
var (
	manualReconcileFailureMu sync.Mutex
	manualReconcileFailures  = make(map[string]int)
)

// b1EscalatedMu guards b1Escalated, the set of B1 sweep auto-refund rows the
// re-poll pass has escalated (pending past stalePendingB1RefundAge). Escalated
// rows are no longer re-polled: the FAILED/REJECTED-terminal branch already
// resolves them (the row leaves the pending query), and a still-pending
// escalated row keeps a deduped CRITICAL admin notification alive while the
// parent stays pending for manual reconciliation. In-memory (no schema change —
// the schema is single-source and pre-launch, no ALTERs), mirroring
// manualReconcileFailures: a process restart resets the set, which merely
// re-polls the row once and re-escalates it (the notification is deduped), never
// suppressing the alert.
var (
	b1EscalatedMu sync.Mutex
	b1Escalated   = make(map[string]bool)
)

// trackReconcileFailureReArm records one more consecutive cap-time reconcile
// failure for each refund row and, once a row crosses
// maxConsecutiveReconcileFailures, surfaces a deduped 'critical_payment_log'
// admin notification for the affected booking(s) — the admin MUST learn a
// reconcile is hard-failing instead of the row silently oscillating under the
// attempt cap forever (A5b/A5c).
func trackReconcileFailureReArm(ctx context.Context, ids []string) {
	manualReconcileFailureMu.Lock()
	notify := false
	for _, id := range ids {
		// A5: cap the counter so the map never grows unbounded and the
		// ==maxConsecutiveReconcileFailures check below is never skipped by an
		// odd increment pattern — a counter that grew past 5 without a reset
		// (e.g. to 6) would silently stop firing the notification forever.
		if manualReconcileFailures[id] < maxTrackedReconcileFailures {
			manualReconcileFailures[id]++
		}
		if manualReconcileFailures[id] == maxConsecutiveReconcileFailures {
			notify = true
		}
	}
	manualReconcileFailureMu.Unlock()
	if !notify {
		return
	}
	notifyCriticalReconcileFailure(ctx, ids)
}

// resetReconcileFailureCount clears a refund row's consecutive-reconcile-
// failure counter. Called whenever a reconcile SUCCEEDS (the row is resolved
// to 'completed' or definitively 'failed'), so the counter reflects
// consecutive failures only — a success in between breaks the streak.
func resetReconcileFailureCount(ids ...string) {
	manualReconcileFailureMu.Lock()
	for _, id := range ids {
		delete(manualReconcileFailures, id)
	}
	manualReconcileFailureMu.Unlock()
}

// notifyCriticalReconcileFailure inserts ONE 'critical_payment_log' admin
// notification per affected booking (deduped by insertCriticalPaymentNotification)
// so a hard-failing reconcile surfaces in the admin notification centre.
// Rows without a booking (gift-card purchases) collapse into one booking-less
// notification — the money event is surfaced, not lost.
func notifyCriticalReconcileFailure(ctx context.Context, ids []string) {
	rows, err := db.Conn.Query(ctx, `
		SELECT DISTINCT booking_id FROM refunds
		WHERE id = ANY($1) AND booking_id IS NOT NULL
	`, ids)
	if err != nil {
		log.Printf("Failed to query bookings for critical reconcile-failure notification: %v", err)
		insertCriticalPaymentNotification(ctx, nil, nil)
		return
	}
	var bookingIDs []string
	for rows.Next() {
		var b string
		if err := rows.Scan(&b); err == nil {
			bookingIDs = append(bookingIDs, b)
		}
	}
	rows.Close()
	if len(bookingIDs) == 0 {
		insertCriticalPaymentNotification(ctx, nil, nil)
		return
	}
	for _, b := range bookingIDs {
		bid := b
		insertCriticalPaymentNotification(ctx, &bid, nil)
	}
}

// reconcileRefundAtSquareExact checks Square for a COMPLETED refund matching
// the EXACT payment+amount, WITHOUT an age bound. Same exact-match semantics
// as reconcileRefundAtSquare (payment_id AND status COMPLETED AND amount), minus
// the begin_time filter: the REFUND_AMOUNT_INVALID disambiguation (A11) must
// not miss a pre-existing refund that predates our refund row. Only an
// exact-amount COMPLETED refund proves the money for THIS amount already moved
// — a smaller partial refund does NOT, and attributing it would mark the row
// completed when only part of the amount was refunded.
//
// Tri-state return matches reconcileRefundAtSquare:
//
//	(id, nil)   — exact COMPLETED refund found
//	(nil, nil)  — genuinely no exact match
//	(nil, err)  — reconcile failed (network/API error)
func reconcileRefundAtSquareExact(ctx context.Context, chargeID string, amountPence int64) (*string, error) {
	refunds, err := SquareClient.ListPaymentRefunds(ctx, chargeID, time.Time{})
	if err != nil {
		log.Printf("Failed to reconcile charge %s against Square: %v", chargeID, err)
		return nil, err
	}
	for i := range refunds {
		r := &refunds[i]
		if r.PaymentID == chargeID && r.Status == "COMPLETED" && r.Amount == amountPence {
			return &r.ID, nil
		}
	}
	return nil, nil
}

type RefundCalculationResult struct {
	TotalPrePaid          float64 `json:"total_pre_paid"`
	ProtectedDeposit      float64 `json:"protected_deposit"`
	RefundableAmount      float64 `json:"refundable_amount"`
	KeptAmount            float64 `json:"kept_amount"`
	HoursUntilAppointment float64 `json:"hours_until_appointment"`
	Tier                  string  `json:"tier"`
}

// paymentRow is one completed payment on a booking, read into a slice before
// the refund loop runs (pgx.Tx does not support concurrent queries on the same
// connection, so rows must be fully drained before Exec/QueryRow in the loop).
type paymentRow struct {
	ID              string
	Amount          float64
	PaymentMethod   string
	SquarePaymentID *string
	GiftCardID      *string
}

// CalculateRefundForCancellation computes the refund amounts for a cancelled booking.
//
// Track A (universal — same rules for all bookings):
//   - >72 hours notice:        Full refund of all pre-payments
//   - 24-72 hours notice:      Keep protected deposit (up to 50%), refund the rest
//   - <24 hours or no-show:    Keep all pre-payments
//
// The "protected deposit" is defined as min(totalPrePaid, subtotal *
// ProtectedDepositMaxPct).
// This means up to 50% of the subtotal is always treated as a deposit for
// refund purposes, regardless of whether deposit_required was set on the booking.
//
// These tiers are the DEFAULT retention for a cancellation (customer calls to
// cancel, or the admin cancels on the customer's behalf without forgiving
// fees). The admin "forgive fees" path (forceFullRefund) overrides the tiers
// entirely so the business keeps nothing — see ProcessCancellationRefundTx for
// the two intents (excusable cancellation vs business-initiated cancellation)
// that share that override.
func CalculateRefundForCancellation(
	subtotal float64,
	totalPrePaid float64,
	cancellationTime time.Time,
	startTime time.Time,
) RefundCalculationResult {
	hoursUntilAppointment := startTime.Sub(cancellationTime).Hours()

	protectedDeposit := math.Min(totalPrePaid, subtotal*ProtectedDepositMaxPct)

	// Round to 2 decimal places
	protectedDeposit = math.Round(protectedDeposit*100) / 100
	totalPrePaidRounded := math.Round(totalPrePaid*100) / 100

	var refundableAmount, keptAmount float64
	var tier string

	fullHrs := FullRefundThreshold.Hours()
	partHrs := PartialRefundThreshold.Hours()

	switch {
	case hoursUntilAppointment > fullHrs:
		refundableAmount = totalPrePaidRounded
		keptAmount = 0
		tier = FullRefundTier

	case hoursUntilAppointment >= partHrs:
		keptAmount = protectedDeposit
		refundableAmount = totalPrePaidRounded - keptAmount
		if refundableAmount < 0 {
			refundableAmount = 0
		}
		tier = PartialRefundTier

	default:
		keptAmount = totalPrePaidRounded
		refundableAmount = 0
		tier = NoRefundTier
	}

	return RefundCalculationResult{
		TotalPrePaid:          totalPrePaidRounded,
		ProtectedDeposit:      protectedDeposit,
		RefundableAmount:      refundableAmount,
		KeptAmount:            keptAmount,
		HoursUntilAppointment: hoursUntilAppointment,
		Tier:                  tier,
	}
}

// lockCancellationPayments serializes a cancellation refund against the manual
// RefundPayment handler and the sweep. Both hold
// `pg_advisory_xact_lock(hashtext('crussell:refund:' || payment_id))`
// (transaction-level, acquired via acquireAdvisoryXactLockBlocking) on the
// payment ids they touch; a cancellation that computes residuals without the
// same locks can over-refund against a manual refund in flight (the manual
// guard read precedes the cancellation's commit). Locks are acquired in
// ascending payment_id order (matching processChargeGroup) to avoid deadlocks.
// EVERY payment row the cancellation may refund is locked — giftcard and cash
// rows included, not just card methods — so two concurrent refunds of the same
// giftcard/cash payment serialize on the residual computation and can never
// double-credit (H3).
func lockCancellationPayments(ctx context.Context, tx pgx.Tx, payments []paymentRow) error {
	var ids []string
	for _, p := range payments {
		// discount/on_the_house rows are already excluded by the caller's
		// query, so every remaining row is one this loop may refund.
		ids = append(ids, p.ID)
	}
	if len(ids) == 0 {
		return nil
	}
	sort.Strings(ids)
	for _, pid := range ids {
		// Blocking xact lock (NOT the bounded try-lock used elsewhere): this is
		// the admin-only cancellation path, and the manual RefundPayment handler
		// can hold the same key across its up-to-30s Square round-trip. A timed
		// out acquire here would abort the cancellation transaction — the caller
		// (manage.go) would commit the cancellation with ZERO refund rows and no
		// sweep retry could ever recover the money. Blocking guarantees the
		// refund runs; the lock auto-releases at the caller's commit/rollback.
		// See acquireAdvisoryXactLockBlocking for the full rationale.
		if err := acquireAdvisoryXactLockBlocking(ctx, tx, "crussell:refund:"+pid); err != nil {
			return fmt.Errorf("failed to acquire cancellation refund lock for payment %s: %w", pid, err)
		}
	}
	return nil
}

// ProcessCancellationRefundTx is like ProcessCancellationRefund but uses an
// externally-provided transaction. The caller owns the transaction lifecycle
// (commit/rollback). Pass a non-nil pgx.Tx to share an existing transaction.
//
// The admin "forgive fees" checkbox (forceFullRefund) serves TWO distinct
// purposes, and both route here identically — it overrides the notice-tier
// calculation so the ENTIRE net pre-paid amount is refunded regardless of how
// close to the appointment the cancellation happens:
//
//   - (A) Genuinely excusable cancellation: the customer has a legitimate
//     excuse (medical, emergency, technical fault, etc.) and the business
//     chooses to waive its notice-period fees as a goodwill gesture.
//   - (B) Business-initiated cancellation: the salon had to cancel the
//     appointment and chooses NOT to keep the money (the deposit/notice
//     retention would be unfair when the cancellation is the business's
//     doing).
//
// These two intents are deliberately NOT distinguished in the refund
// calculation — both mean "the business keeps nothing". When the admin cancels
// on a customer's behalf or a customer calls up to cancel without a forgivable
// excuse, the notice tiers below apply (forceFullRefund=false); the checkbox
// is the explicit opt-out from that retention.
func ProcessCancellationRefundTx(
	ctx context.Context,
	tx pgx.Tx,
	bookingID string,
	subtotal float64,
	totalPrePaid float64,
	startTime time.Time,
	cancellationTime time.Time,
	reason string,
	actorID *string,
	forceFullRefund bool,
) (*RefundCalculationResult, error) {
	calc := CalculateRefundForCancellation(subtotal, totalPrePaid, cancellationTime, startTime)

	if forceFullRefund {
		calc.RefundableAmount = calc.TotalPrePaid
		calc.KeptAmount = 0
		calc.Tier = "admin_full_refund"
	}

	if calc.RefundableAmount <= 0 {
		return &calc, nil
	}

	// Get the booking's user info for refund routing.
	var bookingUserID string
	var isGuest bool
	bookingUserLookupFailed := false
	if err := tx.QueryRow(ctx, `
		SELECT b.user_id, COALESCE(u.account_role = 'guest', false)
		FROM bookings b
		LEFT JOIN users u ON b.user_id = u.id
		WHERE b.id = $1
	`, bookingID).Scan(&bookingUserID, &isGuest); err != nil {
		log.Printf("Failed to get booking user info for refund: %v", err)
		// Non-fatal for Square refunds (which route by payment, not user), but
		// the balance-credit branches below must know the lookup FAILED (vs a
		// genuine guest) so they never record a 'completed' refund when no
		// money could be credited — see creditFailed.
		bookingUserLookupFailed = true
	}

	rows, err := tx.Query(ctx, `
		SELECT id, amount, payment_method, square_payment_id, gift_card_id
		FROM payments
		WHERE booking_id = $1 AND status = 'completed' AND payment_method NOT IN ('discount', 'on_the_house')
		  AND payment_type <> 'tip'
		ORDER BY created_at ASC
	`, bookingID)
	if err != nil {
		log.Printf("Failed to fetch payments for refund: %v", err)
		return &calc, nil
	}
	defer rows.Close()

	// Read all payments into a slice, then close rows immediately.
	var payments []paymentRow
	for rows.Next() {
		var p paymentRow
		if err := rows.Scan(&p.ID, &p.Amount, &p.PaymentMethod, &p.SquarePaymentID, &p.GiftCardID); err != nil {
			log.Printf("Failed to scan payment row: %v", err)
			continue
		}
		payments = append(payments, p)
	}
	if err := rows.Err(); err != nil {
		log.Printf("Payment row iteration error: %v", err)
	}

	// Serialize against the manual RefundPayment handler and the sweep — the
	// residual calculation below must not race a manual refund in flight. If
	// any lock fails, abort: continuing without the lock reopens the over-refund
	// race. pg_advisory_xact_lock auto-releases at the caller's commit.
	if err := lockCancellationPayments(ctx, tx, payments); err != nil {
		return nil, err
	}

	refundRemaining := calc.RefundableAmount

	// Prior refunds per payment record (completed + pending) — the loop must
	// not re-refund money already returned. Sums by payment_id; pending counts
	// because a Square call may already be in flight.
	//
	// This DB-side over-refund guard (completed + pending) is what prevents
	// Square's REFUND_AMOUNT_INVALID in practice: a refund is never issued past
	// the residual `amount - already`. Both this cancellation path and the
	// manual RefundPayment handler compute residuals while holding the same
	// advisory lock (`hashtext('crussell:refund:' || payment_id)` — see
	// lockCancellationPayments), so a manual refund cannot slip past the guard
	// and a cancellation refund cannot be recorded after the manual guard ran
	// without the two serializing. Square no longer documents
	// PAYMENT_ALREADY_REFUNDED; the realistic already-refunded response is
	// REFUND_AMOUNT_INVALID. The square client reconciles that code against
	// Square's refund list (PaymentWasRefunded) and maps an already-refunded
	// payment to ErrRefundAlreadyProcessed; the charge-group/manual sweep
	// handlers additionally reconcile REFUND_AMOUNT_INVALID via an exact-amount
	// COMPLETED-refund check (reconcileRefundAtSquareExact, defense-in-depth —
	// see the A11 disambiguation) and resolve those rows to 'completed' — never
	// 'failed' + admin_notification, which would let the over-refund guard
	// re-issue money Square already returned.
	priorRefunds := make(map[string]float64)
	prRows, prErr := tx.Query(ctx, `
		SELECT payment_id, COALESCE(SUM(amount), 0) FROM refunds
		WHERE booking_id = $1 AND status IN ('completed', 'pending')
		GROUP BY payment_id`, bookingID)
	if prErr != nil {
		log.Printf("Failed to query prior refunds for booking %s: %v", bookingID, prErr)
	} else {
		for prRows.Next() {
			var pid string
			var amt float64
			if err := prRows.Scan(&pid, &amt); err == nil {
				priorRefunds[pid] = amt
			}
		}
		prRows.Close()
	}

	for _, p := range payments {
		if refundRemaining <= 0 {
			break
		}

		paymentID := p.ID
		paymentMethod := p.PaymentMethod
		amount := p.Amount
		giftCardID := p.GiftCardID

		already := priorRefunds[paymentID]
		residual := math.Round((amount-already)*100) / 100
		if residual <= 0 {
			// already fully refunded — don't consume refundRemaining
			continue
		}
		refundThisPayment := math.Min(residual, refundRemaining)
		var squareRefundID *string

		// creditFailed records that money for this payment did NOT actually
		// move (deleted gift card, failed balance credit) so the refund record
		// is inserted 'failed' instead of claiming a completed refund (M2).
		creditFailed := false

		switch paymentMethod {
		case "online_square", "in_person_card":
			// Square API refund is processed AFTER the transaction commits
			// (see ProcessPendingSquareRefunds). Inside the tx we only record
			// the refund record as "pending" for post-commit processing.
			if isGuest || bookingUserID == "" {
				log.Printf("Guest card refund: booking %s, payment %s, amount £%.2f — will be processed after commit", bookingID, paymentID, refundThisPayment)
			}
			// squareRefundID stays nil — will be set by ProcessPendingSquareRefunds

		case "giftcard":
			if giftCardID == nil || *giftCardID == "" {
				// A giftcard payment with no gift_card_id was made from the
				// user's gift-card ACCOUNT balance (the terminal giftcard path
				// stores no gift_card_id for balance payments). Credit the
				// booking user's balance back — skipping would lose the money
				// while the completed refund row claims it was returned (C3).
				// Guests get no balance credit, mirroring the cash branch.
				if isGuest || bookingUserID == "" {
					if isGuest {
						log.Printf("Guest giftcard refund: booking %s, payment %s, amount £%.2f — no balance credit", bookingID, paymentID, refundThisPayment)
					} else {
						log.Printf("Giftcard payment %s has no gift_card_id and no booking user — cannot refund. Skipping.", paymentID)
					}
					if bookingUserLookupFailed {
						// The user may well exist — the lookup failed
						// transiently, so 'completed' would claim money was
						// credited when none could be. Mark the record failed
						// for admin reconciliation instead.
						creditFailed = true
					}
					break
				}
				log.Printf("Crediting £%.2f to user %s gift-card balance for account-balance giftcard payment %s", refundThisPayment, bookingUserID, paymentID)
				if _, balErr := tx.Exec(ctx, `
				INSERT INTO user_giftcard_balances (user_id, balance, updated_at)
				VALUES ($1, $2, NOW())
				ON CONFLICT (user_id) DO UPDATE SET
					balance = user_giftcard_balances.balance + EXCLUDED.balance,
					updated_at = NOW()
			`, bookingUserID, refundThisPayment); balErr != nil {
					log.Printf("Failed to credit user %s gift-card balance for refund of booking %s: %v", bookingUserID, bookingID, balErr)
					creditFailed = true
				}
				break
			}
			// expiry_date is maintained by EVERY gift-card write that counts as
			// a "use" (balance check, top-up, transfer, redeem, payment, refund
			// credit), so a non-NULL expiry_date means the rolling timer is
			// authoritative. NULL semantics stay fail-open: an unset expiry_date
			// cannot prove the card is expired, so the refund proceeds.
			var expired bool
			if err := tx.QueryRow(ctx, `
				SELECT expiry_date IS NOT NULL AND expiry_date < NOW()
				FROM gift_cards WHERE id = $1
			`, *giftCardID).Scan(&expired); err != nil {
				log.Printf("Failed to check gift card %s expiry: %v — proceeding with refund", *giftCardID, err)
			} else if expired {
				log.Printf("Gift card %s has expired — money retained by salon, no refund due for booking %s", *giftCardID, bookingID)
				// Money-safety (C4): the UPDATE gift_cards credit above is
				// SKIPPED for expired cards (money retained by the salon), so
				// creditFailed must be set here — otherwise the shared tail
				// below inserts the refund record as 'completed' claiming money
				// was returned when it never moved.
				creditFailed = true
				break
			}
			// Refunding to the card is a "use" per the rolling-expiry terms —
			// reset the timer at the same moment the balance is credited.
			gcExpiryMonths, expiryErr := GetGiftCardExpiryMonths(ctx, tx)
			if expiryErr != nil {
				log.Printf("Failed to query gift card expiry months (using default %d): %v", defaultGiftCardExpiryMonths, expiryErr)
				gcExpiryMonths = defaultGiftCardExpiryMonths
			}
			gcTag, gcErr := tx.Exec(ctx, `
				UPDATE gift_cards SET amount_remaining = amount_remaining + $1, last_used_at = NOW(), expiry_date = NOW() + ($3 * INTERVAL '1 month')
				WHERE id = $2
			`, refundThisPayment, *giftCardID, gcExpiryMonths)
			if gcErr != nil {
				log.Printf("Failed to refund £%.2f to gift card %s: %v", refundThisPayment, *giftCardID, gcErr)
				creditFailed = true
				break
			}
			if gcTag.RowsAffected() == 0 {
				// M2: the gift card no longer exists — the UPDATE matched 0 rows
				// and no money moved. Marking the refund 'completed' would claim
				// money was returned when it wasn't.
				slog.Error("CRITICAL: gift card refund UPDATE affected 0 rows — card deleted?, refund NOT credited", "gift_card_id", *giftCardID, "payment_id", paymentID, "booking_id", bookingID, "amount", refundThisPayment)
				creditFailed = true
				break
			}
			if _, err := tx.Exec(ctx, `
				INSERT INTO gift_card_transactions (gift_card_id, transaction_type, amount, reference_type, reference_id, user_id, notes)
				VALUES ($1, 'refund', $2, 'booking', $3, $4, $5)
			`, *giftCardID, refundThisPayment, bookingID, bookingUserID, "Refund from cancelled booking"); err != nil {
				log.Printf("Failed to create gift card transaction for refund: %v", err)
			}

		case "cash":
			if isGuest || bookingUserID == "" {
				if isGuest {
					log.Printf("Guest cash refund: booking %s, payment %s, amount £%.2f — admin must process cash refund at till", bookingID, paymentID, refundThisPayment)
				} else {
					log.Printf("Cash refund: payment %s has no booking user — skipping balance credit", paymentID)
				}
				if bookingUserLookupFailed {
					// The booking-user lookup errored, so this may be a real
					// user whose balance credit was skipped — a 'completed'
					// refund would claim money was returned when it wasn't.
					creditFailed = true
				}
			} else {
				log.Printf("Crediting £%.2f to user %s balance for cash payment %s", refundThisPayment, bookingUserID, paymentID)
				if _, balErr := tx.Exec(ctx, `
				INSERT INTO user_giftcard_balances (user_id, balance, updated_at)
				VALUES ($1, $2, NOW())
				ON CONFLICT (user_id) DO UPDATE SET
					balance = user_giftcard_balances.balance + EXCLUDED.balance,
					updated_at = NOW()
			`, bookingUserID, refundThisPayment); balErr != nil {
					log.Printf("Failed to credit user %s balance for refund of booking %s: %v", bookingUserID, bookingID, balErr)
					creditFailed = true
				}
			}

		default:
			log.Printf("Skipping refund for payment %s with method %q (no money exchanged)", paymentID, paymentMethod)
		}

		recordStatus := "completed"
		if paymentMethod == "online_square" || paymentMethod == "in_person_card" {
			recordStatus = "pending"
		}
		if creditFailed {
			recordStatus = "failed"
		}
		record := RefundRecord{
			PaymentID:      paymentID,
			BookingID:      bookingID,
			Amount:         refundThisPayment,
			SquareRefundID: squareRefundID,
			Status:         recordStatus,
			Reason:         reason,
			Origin:         "cancellation",
			CreatedBy:      actorID,
			CreatedAt:      clock.Now(),
		}

		// Deterministic idempotency key so a scheduler retry can never issue a
		// second Square refund. Format never collides with the handler's
		// "-refund-" keys.
		refundKey := paymentID + "-square-" + strconv.FormatInt(int64(math.Round(refundThisPayment*100)), 10)
		record.IdempotencyKey = &refundKey

		tag, dbErr := tx.Exec(ctx, `
			INSERT INTO refunds (payment_id, booking_id, amount, square_refund_id, status, reason, idempotency_key, created_by, created_at, origin)
			VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10)
			ON CONFLICT (idempotency_key) DO NOTHING
		`, record.PaymentID, record.BookingID, record.Amount, record.SquareRefundID, record.Status, record.Reason, record.IdempotencyKey, record.CreatedBy, record.CreatedAt, record.Origin)
		if dbErr != nil {
			log.Printf("Failed to create refund record for payment %s: %v", paymentID, dbErr)
			continue
		}
		// tag.RowsAffected() == 0 means the same idempotency_key already exists
		// (a prior refund row for this payment+amount in 'failed' state — money
		// never moved, but a row exists). Dedup — skip WITHOUT consuming
		// refundRemaining so the loop can allocate to the next payment, exactly
		// as the pre-ON-CONFLICT UNIQUE-violation path behaved.
		if tag.RowsAffected() == 0 {
			log.Printf("Refund for payment %s amount £%.2f already exists (idempotency dedup) — skipping without consuming refundRemaining", paymentID, refundThisPayment)
			continue
		}

		refundRemaining -= refundThisPayment
	}

	if bookingUserID != "" {
		var loyaltyUsed bool
		if err := tx.QueryRow(ctx, "SELECT EXISTS(SELECT 1 FROM booking_discounts WHERE booking_id = $1 AND discount_source = 'loyalty')", bookingID).Scan(&loyaltyUsed); err != nil {
			log.Printf("Failed to check loyalty stamp refund for booking %s: %v", bookingID, err)
		} else if loyaltyUsed {
			loyaltyTag, loyaltyErr := tx.Exec(ctx, "UPDATE users SET loyalty_stamps = loyalty_stamps + $1 WHERE id = $2", LoyaltyStampCost, bookingUserID)
			switch {
			case loyaltyErr != nil:
				log.Printf("Failed to refund loyalty stamps for booking %s: %v", bookingID, loyaltyErr)
			case loyaltyTag.RowsAffected() == 0:
				slog.Error("CRITICAL: loyalty stamp refund UPDATE affected 0 rows — user not found", "user_id", bookingUserID, "booking_id", bookingID)
			default:
				log.Printf("Refunded %d loyalty stamps to user %s after cancellation of booking %s", LoyaltyStampCost, bookingUserID, bookingID)
			}
		}
		// B13: the full payment refund is issued above; a campaign-loss balance
		// credit granted at charge time (refundLostCampaignAsBalanceCredit) must
		// be reversed here or the customer gets the money back AND keeps the
		// promised-discount credit — a double credit. No-op when the booking
		// never received a B13 credit.
		clawbackB13CampaignCredit(ctx, tx, bookingID, bookingUserID)
	}

	return &calc, nil
}

// clawbackB13CampaignCredit reverses a B13 campaign-loss balance credit
// (refundLostCampaignAsBalanceCredit) when the booking is cancelled: the
// customer receives the full payment refund above, so keeping the promised-
// discount credit on their gift-card balance too would be a double credit. The
// credit is located via its gift_card_transactions audit row (reference_type
// 'b13_campaign_loss', reference_id = booking). The balance debit is guarded
// (balance >= amount) so a balance already spent below the credit is never
// driven negative; an insufficient balance is flagged for manual
// reconciliation instead of silently kept. Idempotent: an existing reversal
// row for the booking prevents a re-run from debiting twice.
func clawbackB13CampaignCredit(ctx context.Context, tx pgx.Tx, bookingID, userID string) {
	if bookingID == "" || userID == "" {
		return
	}
	var alreadyReversed bool
	if err := tx.QueryRow(ctx, `
		SELECT EXISTS(
			SELECT 1 FROM gift_card_transactions
			WHERE reference_type = 'b13_campaign_loss' AND reference_id = $1 AND transaction_type = 'balance_debit'
		)
	`, bookingID).Scan(&alreadyReversed); err != nil {
		log.Printf("B13: failed to check for an existing clawback of booking %s: %v", bookingID, err)
		return
	}
	if alreadyReversed {
		return
	}
	var creditPounds float64
	if err := tx.QueryRow(ctx, `
		SELECT COALESCE(SUM(amount), 0) FROM gift_card_transactions
		WHERE reference_type = 'b13_campaign_loss' AND reference_id = $1
	`, bookingID).Scan(&creditPounds); err != nil || creditPounds <= 0 {
		return
	}
	tag, err := tx.Exec(ctx, `
		UPDATE user_giftcard_balances
		SET balance = user_giftcard_balances.balance - $1, updated_at = NOW()
		WHERE user_id = $2 AND balance >= $1
	`, creditPounds, userID)
	if err != nil {
		log.Printf("B13: failed to claw back £%.2f campaign-loss credit from user %s (booking %s): %v", creditPounds, userID, bookingID, err)
		return
	}
	if tag.RowsAffected() == 0 {
		log.Printf("CRITICAL: B13 campaign-loss credit of £%.2f for booking %s could not be clawed back from user %s (balance < amount — credit partially spent) — MANUAL RECONCILIATION REQUIRED", creditPounds, bookingID, userID)
		return
	}
	// Record the reversal on the same anchor card as the credit for the audit
	// trail (also serves as the idempotency marker above).
	var anchorCardID string
	if err := tx.QueryRow(ctx, `
		SELECT gift_card_id FROM gift_card_transactions
		WHERE reference_type = 'b13_campaign_loss' AND reference_id = $1
		ORDER BY created_at DESC LIMIT 1
	`, bookingID).Scan(&anchorCardID); err == nil && anchorCardID != "" {
		if _, err := tx.Exec(ctx, `
			INSERT INTO gift_card_transactions (gift_card_id, transaction_type, amount, reference_type, reference_id, user_id, notes)
			VALUES ($1, 'balance_debit', $2, 'b13_campaign_loss', $3, $4, $5)
		`, anchorCardID, creditPounds, bookingID, userID, "B13 campaign-loss credit clawed back on cancellation"); err != nil {
			log.Printf("B13: failed to record clawback transaction for user %s (booking %s): %v", userID, bookingID, err)
		}
	}
	log.Printf("B13: clawed back £%.2f campaign-loss balance credit from user %s after cancellation of booking %s", creditPounds, userID, bookingID)
}

// ProcessCancellationRefund calculates and records refunds for a cancelled
// booking, processing refunds against the booking's completed payments up to
// the calculated refundable amount. The wrapper owns its own transaction and
// delegates the refund loop to ProcessCancellationRefundTx
// (forceFullRefund=false) so the standalone and in-transaction callers share
// one implementation; after a successful commit it runs the post-commit
// Square pass (ProcessPendingSquareRefunds). Returns the refund calculation
// and whether any refunds were processed.
func ProcessCancellationRefund(
	ctx context.Context,
	bookingID string,
	subtotal float64,
	totalPrePaid float64,
	startTime time.Time,
	cancellationTime time.Time,
	reason string,
	actorID *string,
) (*RefundCalculationResult, error) {
	calc := CalculateRefundForCancellation(subtotal, totalPrePaid, cancellationTime, startTime)

	if calc.RefundableAmount <= 0 {
		return &calc, nil
	}

	tx, err := db.Conn.Begin(ctx)
	if err != nil {
		log.Printf("Failed to begin transaction for cancellation refund: %v", err)
		return &calc, nil
	}
	defer func() {
		if err := tx.Rollback(ctx); err != nil && !errors.Is(err, pgx.ErrTxClosed) {
			slog.Error("failed to rollback transaction", "err", err)
		}
	}()

	// Delegate the whole refund loop to the transactional variant — the
	// non-tx wrapper exists only to own the transaction lifecycle and fire the
	// post-commit Square pass (ProcessPendingSquareRefunds) after a successful
	// commit. The Tx variant returns an error ONLY on lock failure; every other
	// failure logs internally and returns (calc, nil).
	res, txErr := ProcessCancellationRefundTx(ctx, tx, bookingID, subtotal, totalPrePaid, startTime, cancellationTime, reason, actorID, false)
	if txErr != nil {
		log.Printf("Failed to acquire cancellation refund locks for booking %s: %v", bookingID, txErr)
		return &calc, txErr
	}

	if cErr := tx.Commit(ctx); cErr != nil {
		slog.Error("CRITICAL: failed to commit cancellation refund", "booking_id", bookingID, "err", cErr)
		return nil, fmt.Errorf("failed to commit cancellation refund: %w", cErr)
	}

	// Process pending Square refunds after the transaction commits successfully.
	// This ensures Square API calls only happen if the DB records persist.
	ProcessPendingSquareRefunds(ctx, bookingID, reason)

	return res, nil
}

// ProcessPendingSquareRefunds resolves a booking's pending cancellation card
// refunds AFTER the enclosing transaction has committed — Square API calls
// only happen if the DB records persist.
//
// Pending refunds are aggregated into ONE Square refund per charge
// (square_payment_id): split payment records sharing a single Square charge
// are refunded together, never per-row. See processChargeGroup.
func ProcessPendingSquareRefunds(ctx context.Context, bookingID string, reason string) {
	// (a) Terminal pre-pass scoped to this booking: card refunds with no
	//     Square reference can never be refunded via Square. Mark them failed
	//     so they stop retrying, and surface the affected booking in the admin
	//     notification centre for in-person arrangement. Must run OUTSIDE any
	//     GROUP BY — Postgres lumps NULLs together, so these rows can't be
	//     handled in the charge grouping below.
	rows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		UPDATE refunds r SET status = 'failed'
		FROM payments p
		WHERE p.id = r.payment_id
		  AND r.booking_id = $1
		  AND r.status = 'pending' AND r.refund_attempts < %d
		  AND p.payment_method IN ('online_square', 'in_person_card')
		  AND p.square_payment_id IS NULL
		  AND r.origin = 'cancellation'
		RETURNING r.id
	`, maxManualRefundAttempts), bookingID)
	if err != nil {
		log.Printf("Failed to mark Square-less card refunds failed for booking %s: %v", bookingID, err)
	} else {
		var failedIDs []string
		for rows.Next() {
			var id string
			if err := rows.Scan(&id); err == nil {
				failedIDs = append(failedIDs, id)
			}
		}
		if err := rows.Err(); err != nil {
			log.Printf("Failed to iterate Square-less card refunds for booking %s: %v", bookingID, err)
		}
		rows.Close()
		if len(failedIDs) > 0 {
			log.Printf("Marked %d Square-less card refund(s) failed for booking %s (in-person arrangement needed)", len(failedIDs), bookingID)
		}
		insertRefundFailedNotifications(ctx, failedIDs)
	}

	// (b) This booking's card charges with pending refunds — one aggregated
	//     Square refund per charge.
	for _, chargeID := range queryChargesWithPendingRefunds(ctx, "r.booking_id = $1", bookingID) {
		if _, err := processChargeGroup(ctx, chargeID, fetchPendingChargeRows(ctx, chargeID), reason); err != nil {
			log.Printf("Failed to process charge %s for booking %s: %v", chargeID, bookingID, err)
		}
	}
}

// SweepPendingSquareRefunds is the scheduler job that retries every booking's
// pending cancellation card refunds. Registered in internal/jobs/cleanup.go as
// "sweep-pending-square-refunds".
func SweepPendingSquareRefunds(ctx context.Context) (int, error) {
	// (a) Terminal pre-pass across all bookings: card refunds with no Square
	//     reference can never be refunded via Square → mark failed and surface
	//     for in-person arrangement. Must run OUTSIDE any GROUP BY — Postgres
	//     lumps NULLs together, so these rows can't be handled in the charge
	//     grouping below.
	rows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		UPDATE refunds r SET status = 'failed'
		FROM payments p
		WHERE p.id = r.payment_id
		  AND r.status = 'pending' AND r.refund_attempts < %d
		  AND p.payment_method IN ('online_square', 'in_person_card')
		  AND p.square_payment_id IS NULL
		  AND r.origin = 'cancellation'
		RETURNING r.id
	`, maxManualRefundAttempts))
	if err != nil {
		log.Printf("Failed to mark Square-less card refunds failed: %v", err)
	} else {
		var failedIDs []string
		for rows.Next() {
			var id string
			if err := rows.Scan(&id); err == nil {
				failedIDs = append(failedIDs, id)
			}
		}
		if err := rows.Err(); err != nil {
			log.Printf("Failed to iterate Square-less card refunds during sweep: %v", err)
		}
		rows.Close()
		if len(failedIDs) > 0 {
			log.Printf("Marked %d Square-less card refund(s) failed (in-person arrangement needed)", len(failedIDs))
		}
		insertRefundFailedNotifications(ctx, failedIDs)
	}

	// (b) Charges with pending refunds — one aggregated Square refund each.
	processed := 0
	for _, chargeID := range queryChargesWithPendingRefunds(ctx, "") {
		n, err := processChargeGroup(ctx, chargeID, fetchPendingChargeRows(ctx, chargeID), "scheduled_retry")
		if err != nil {
			log.Printf("Sweep: failed to process charge %s: %v", chargeID, err)
			continue
		}
		processed += n
	}

	// (c) Stale MANUAL pending refunds (the handler's ambiguous-error path) are
	// invisible to the cancellation passes above (they filter origin='manual'
	// out) — without this pass they were never retried, permanently blocking
	// the over-refund guard and depressing booking TotalPaid.
	n, err := sweepManualPendingSquareRefunds(ctx)
	if err != nil {
		log.Printf("Sweep: failed to process manual pending refunds: %v", err)
	} else {
		processed += n
	}
	return processed, nil
}

// pendingChargeRow is one eligible pending cancellation refund row tied to a
// single Square charge.
type pendingChargeRow struct {
	ID        string // refunds.id
	PaymentID string // payments.id — advisory-lock ordering
	Amount    float64
	CreatedAt time.Time
}

// queryChargesWithPendingRefunds returns the distinct square_payment_ids that
// have at least one eligible pending cancellation refund row. extraWhere is an
// optional extra SQL predicate bound by args (e.g. "r.booking_id = $1").
func queryChargesWithPendingRefunds(ctx context.Context, extraWhere string, args ...any) []string {
	q := fmt.Sprintf(`
		SELECT DISTINCT p.square_payment_id
		FROM refunds r
		JOIN payments p ON p.id = r.payment_id
		WHERE r.status = 'pending' AND r.refund_attempts < %d
		  AND p.square_payment_id IS NOT NULL
		  AND p.payment_method IN ('online_square', 'in_person_card')
		  AND r.origin = 'cancellation'`, maxManualRefundAttempts)
	if extraWhere != "" {
		q += " AND " + extraWhere
	}
	rows, err := db.Conn.Query(ctx, q, args...)
	if err != nil {
		log.Printf("Failed to query charges with pending refunds: %v", err)
		return nil
	}
	defer rows.Close()
	var out []string
	for rows.Next() {
		var s string
		if err := rows.Scan(&s); err != nil {
			log.Printf("Failed to scan charge id: %v", err)
			continue
		}
		out = append(out, s)
	}
	if err := rows.Err(); err != nil {
		log.Printf("Charge id iteration error: %v", err)
	}
	return out
}

// fetchPendingChargeRows returns all eligible pending cancellation refund rows
// for a single Square charge, ordered by refund id.
func fetchPendingChargeRows(ctx context.Context, chargeID string) []pendingChargeRow {
	rows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		SELECT r.id, p.id, r.amount, r.created_at
		FROM refunds r
		JOIN payments p ON p.id = r.payment_id
		WHERE p.square_payment_id = $1
		  AND r.status = 'pending' AND r.refund_attempts < %d
		  AND r.origin = 'cancellation'
		ORDER BY r.id
	`, maxManualRefundAttempts), chargeID)
	if err != nil {
		log.Printf("Failed to query pending refunds for charge %s: %v", chargeID, err)
		return nil
	}
	defer rows.Close()
	var out []pendingChargeRow
	for rows.Next() {
		var pr pendingChargeRow
		if err := rows.Scan(&pr.ID, &pr.PaymentID, &pr.Amount, &pr.CreatedAt); err != nil {
			log.Printf("Failed to scan pending refund row for charge %s: %v", chargeID, err)
			continue
		}
		out = append(out, pr)
	}
	if err := rows.Err(); err != nil {
		log.Printf("Pending refund row iteration error for charge %s: %v", chargeID, err)
	}
	return out
}

// isRefundAmountInvalid reports whether err carries Square's
// REFUND_AMOUNT_INVALID code — structurally (the real HTTP client wraps the
// code in a squareAPIError) or by message (the dev mock embeds the code in its
// simulated error).
func isRefundAmountInvalid(err error) bool {
	if err == nil {
		return false
	}
	if square.ErrorCode(err) == "REFUND_AMOUNT_INVALID" {
		return true
	}
	return strings.Contains(err.Error(), "REFUND_AMOUNT_INVALID")
}

// reconcileRefundAtSquare checks Square for a COMPLETED refund matching the
// exact charge-level amount before the age guard / attempt cap marks rows
// 'failed'. Tri-state return:
//
//	(id, nil)   — exact COMPLETED refund found → caller marks rows completed
//	(nil, nil)  — genuinely no match → caller may mark rows failed
//	(nil, err)  — reconcile FAILED (network/API error) → caller MUST leave
//	              rows pending and skip the terminal transition. Marking
//	              failed on an unknown state would let the over-refund guard
//	              exclude money that actually left the business.
//
// Exact equality on amount AND status COMPLETED AND payment_id: a larger
// COMPLETED refund on the same charge is a manual per-record refund,
// attributing it would mark our rows completed when the aggregate money never
// moved.
func reconcileRefundAtSquare(ctx context.Context, chargeID string, totalPence int64, oldestCreatedAt time.Time) (*string, error) {
	refunds, err := SquareClient.ListPaymentRefunds(ctx, chargeID, oldestCreatedAt)
	if err != nil {
		log.Printf("Failed to reconcile charge %s against Square: %v", chargeID, err)
		return nil, err
	}
	for i := range refunds {
		r := &refunds[i]
		if r.PaymentID == chargeID && r.Status == "COMPLETED" && r.Amount == totalPence {
			return &r.ID, nil
		}
	}
	return nil, nil
}

// insertRefundFailedNotifications surfaces failed refunds in the admin
// notification centre — one row per affected booking. The sweep only processes
// 'pending' rows, so this fires once per row transition (no spam); the
// NOT EXISTS guard prevents duplicates on re-runs.
//
// The webhooks package carries the SINGULAR variant of this same insert —
// insertRefundFailedNotification (handlers/webhooks/square.go) — which demotes
// a single webhook-surfaced FAILED refund to the identical 'refund_failed'
// row. The two share the same NOT EXISTS dedup guard on
// (reason='refund_failed', booking_id), so a refund resolved by either path
// can never be double-notified; keep the reason string and dedup predicate in
// lockstep when either changes.
// InsertRefundFailedNotifications is the EXPORTED single source for surfacing
// failed refunds in the admin notification centre, consumed by both the sweep
// path (sweep.go) and the webhook path (handlers/webhooks/square.go). The
// webhook package calls this instead of maintaining its own copy, so the SQL
// and the (reason='refund_failed', booking_id) dedup predicate live in exactly
// one place. It delegates to the package-internal insertRefundFailedNotifications.
func InsertRefundFailedNotifications(ctx context.Context, refundIDs []string) {
	insertRefundFailedNotifications(ctx, refundIDs)
}

func insertRefundFailedNotifications(ctx context.Context, refundIDs []string) {
	if len(refundIDs) == 0 {
		return
	}
	tag, err := db.Conn.Exec(ctx, `
		INSERT INTO admin_notifications (reason, booking_id, created_at)
		SELECT DISTINCT 'refund_failed'::admin_notification_reason, booking_id, NOW()
		FROM refunds
		WHERE id = ANY($1) AND status = 'failed'
		  AND NOT EXISTS (
		      SELECT 1 FROM admin_notifications an
		      WHERE an.reason = 'refund_failed' AND an.booking_id = refunds.booking_id
		  )
	`, refundIDs)
	if err != nil {
		log.Printf("Failed to insert admin_notifications for failed refunds: %v", err)
		return
	}
	if n := int(tag.RowsAffected()); n > 0 {
		log.Printf("Inserted %d admin_notification(s) for failed refunds", n)
	}
}

// pendingRowsAtAttemptCap returns the refund ids still pending at the
// maxManualRefundAttempts cap — the candidates for terminal 'failed' resolution.
func pendingRowsAtAttemptCap(ctx context.Context, ids []string) []string {
	if len(ids) == 0 {
		return nil
	}
	rows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		SELECT id FROM refunds
		WHERE id = ANY($1) AND status = 'pending' AND refund_attempts >= %d
	`, maxManualRefundAttempts), ids)
	if err != nil {
		log.Printf("Failed to query refunds at attempt cap: %v", err)
		return nil
	}
	defer rows.Close()
	var out []string
	for rows.Next() {
		var id string
		if err := rows.Scan(&id); err != nil {
			log.Printf("Failed to scan refund id at attempt cap: %v", err)
			continue
		}
		out = append(out, id)
	}
	return out
}

// processChargeGroup issues ONE Square refund for a charge (square_payment_id)
// covering the sum of its pending cancellation refund rows — the owner
// directive that split records sharing a charge produce a single Square
// refund, never one per row.
//
// Callers pass the charge's current pending rows (fetchPendingChargeRows); the
// rows are re-read under the per-payment advisory lock so a concurrent manual
// refund or another sweep can't double-process.
func processChargeGroup(ctx context.Context, chargeID string, rows []pendingChargeRow, reason string) (int, error) {
	if len(rows) == 0 {
		return 0, nil
	}

	// Serialize against the manual RefundPayment handler: acquire the SAME
	// per-payment advisory lock (`hashtext('crussell:refund:' || payment_id)`)
	// the handler uses, in ascending payment_id order to avoid deadlocks.
	pinConn, err := db.Conn.Acquire(ctx)
	if err != nil {
		log.Printf("Failed to acquire connection for refund lock (charge %s): %v", chargeID, err)
		return 0, nil
	}
	defer pinConn.Release()

	paymentIDs := make([]string, 0, len(rows))
	seen := make(map[string]bool, len(rows))
	for _, r := range rows {
		if !seen[r.PaymentID] {
			seen[r.PaymentID] = true
			paymentIDs = append(paymentIDs, r.PaymentID)
		}
	}
	sort.Strings(paymentIDs)

	locked := 0
	for _, pid := range paymentIDs {
		// Bounded try-lock (R6) so the sweep never blocks a pool connection
		// while the manual handler holds the same key across its Square call.
		ok, lockErr := acquireAdvisoryLock(ctx, pinConn, "crussell:refund:"+pid)
		if lockErr != nil {
			log.Printf("Failed to acquire refund lock for payment %s (charge %s): %v", pid, chargeID, lockErr)
			break
		}
		if !ok {
			log.Printf("Refund lock for payment %s (charge %s) not acquired within bound — a refund is in progress", pid, chargeID)
			break
		}
		locked++
	}
	if locked < len(paymentIDs) {
		// Give up on this charge (the manual handler may hold the lock). The
		// sweep continues to the next charge rather than aborting fatally.
		for _, pid := range paymentIDs[:locked] {
			releasePaymentLock(pinConn, "crussell:refund:"+pid)
		}
		return 0, nil
	}
	defer func() {
		for _, pid := range paymentIDs {
			releasePaymentLock(pinConn, "crussell:refund:"+pid)
		}
	}()

	// Re-read under the lock — only rows still pending and under the attempt
	// cap are eligible (a concurrent manual refund may have resolved some).
	ids := make([]string, 0, len(rows))
	for _, r := range rows {
		ids = append(ids, r.ID)
	}
	pendingRows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		SELECT id, amount, created_at FROM refunds
		WHERE id = ANY($1) AND status = 'pending' AND refund_attempts < %d
	`, maxManualRefundAttempts), ids)
	if err != nil {
		log.Printf("Failed to re-read pending refunds under lock (charge %s): %v", chargeID, err)
		return 0, nil
	}
	var pending []pendingChargeRow
	for pendingRows.Next() {
		var pr pendingChargeRow
		if err := pendingRows.Scan(&pr.ID, &pr.Amount, &pr.CreatedAt); err != nil {
			log.Printf("Failed to scan pending refund under lock (charge %s): %v", chargeID, err)
			continue
		}
		pending = append(pending, pr)
	}
	pendingRows.Close()
	if len(pending) == 0 {
		return 0, nil
	}

	// Age guard: Square's idempotency-key retention is finite (~24h). If the
	// oldest pending row predates stalePendingRefundAge, re-issuing with the
	// same charge key risks Square treating it as a NEW refund → double refund.
	// Reconcile FIRST: money may already have moved at Square (response loss),
	// and failing the rows without checking would let the over-refund guard
	// exclude money that actually left the business. Only when Square shows no
	// exact COMPLETED refund do we mark failed and surface for manual review.
	oldest := pending[0].CreatedAt
	for _, pr := range pending[1:] {
		if pr.CreatedAt.Before(oldest) {
			oldest = pr.CreatedAt
		}
	}
	var totalPence int64
	for _, pr := range pending {
		totalPence += int64(math.Round(pr.Amount * 100))
	}
	if clock.Now().Sub(oldest) > stalePendingRefundAge {
		sqRefundID, rcErr := reconcileRefundAtSquare(ctx, chargeID, totalPence, oldest)
		switch {
		case rcErr != nil:
			// Reconcile failed — unknown whether Square refunded. Leave rows
			// pending for the next sweep; NEVER mark failed on an unknown state
			// (that would let the over-refund guard exclude moved money).
			log.Printf("Reconcile failed for aged charge %s (%v) — leaving %d refund row(s) pending for the next sweep", chargeID, rcErr, len(pending))
			return 0, nil
		case sqRefundID != nil:
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET status = 'completed', square_refund_id = $1
				WHERE id = ANY($2) AND status = 'pending'
			`, *sqRefundID, idsOf(pending)); upErr != nil {
				log.Printf("Failed to mark aged pending refunds completed after Square reconcile (charge %s): %v", chargeID, upErr)
			}
			// A5: terminal resolution — clear the consecutive-failure counter.
			resetReconcileFailureCount(idsOf(pending)...)
			log.Printf("Aged card refunds for charge %s reconciled at Square — COMPLETED refund %s found, marked completed", chargeID, *sqRefundID)
			return len(pending), nil
		default:
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET status = 'failed'
				WHERE id = ANY($1) AND status = 'pending'
			`, idsOf(pending)); upErr != nil {
				log.Printf("Failed to mark aged pending refunds failed (charge %s): %v", chargeID, upErr)
			}
			// A5: terminal resolution — clear the consecutive-failure counter.
			resetReconcileFailureCount(idsOf(pending)...)
			insertRefundFailedNotifications(ctx, idsOf(pending))
			// TODO (P6 email/SMS delivery): notify user AND admin when the email/SMS
			// system lands; until then the admin_notifications row above is the only
			// channel. Verify the Square dashboard first.
			log.Printf("Card refunds for charge %s are older than stalePendingRefundAge (23h) and Square shows no COMPLETED refund — marked 'failed' and admin notified; TODO email user+admin to arrange in-person cash pickup at the salon (give at least a day's notice for cash on hand)", chargeID)
			return len(pending), nil
		}
	}

	// ONE Square refund per charge with a CHARGE-STABLE idempotency key. The
	// key is derived from the charge ID (square_payment_id) ONLY — NEVER from
	// the set of pending row IDs — so it is identical across every sweep run
	// no matter how the pending set evolves. The key is used ONLY for the
	// Square call — never stored in refunds.idempotency_key (the per-row keys
	// remain the audit trail).
	//
	//   SAME-set retry (a crash/response-loss where the pending set is
	//   unchanged) → SAME key → Square's idempotency dedup returns the
	//   ORIGINAL refund, so a retry can never double-refund.
	//
	//   CHANGED set (a new cancellation refund row joined the group while the
	//   old rows still sit 'pending' — the CRITICAL-log path where the
	//   post-refund DB UPDATE failed) → STILL the SAME key. Money for the
	//   original total has ALREADY moved at Square, so a set-derived key would
	//   have hashed to a NEW key and issued a SECOND Square refund on top —
	//   double-refunding the customer (C6). Deduping on the charge key returns
	//   the original refund instead; the new row resolves against it and any
	//   residual gap is a known, admin-visible shortfall rather than lost
	//   money. The stalePendingRefundAge age-guard reconcile above still
	//   protects the cross-sweep case where Square's finite (~24h) key
	//   retention may have lapsed.
	sqResult, sqErr := SquareClient.RefundPayment(ctx, square.RefundPaymentReq{
		PaymentID:      chargeID,
		Amount:         totalPence,
		IdempotencyKey: chargeAggKey(chargeID),
		Reason:         reason,
	})
	// REFUND_AMOUNT_INVALID is Square's ambiguous signal for BOTH a genuinely
	// invalid refund amount AND an already-refunded payment (Square no longer
	// documents PAYMENT_ALREADY_REFUNDED). Disambiguate with the EXACT amount
	// BEFORE the classification switch so the reconciliation applies whether
	// the client classified the code as a definitive decline or as
	// already-processed: only an exact-amount COMPLETED refund at Square proves
	// the money for THIS amount already moved (A11). A smaller partial refund
	// does NOT — marking the rows completed would claim the full amount was
	// refunded when only part of it was.
	if isRefundAmountInvalid(sqErr) {
		sqRefundID, rcErr := reconcileRefundAtSquareExact(ctx, chargeID, totalPence)
		switch {
		case rcErr != nil:
			// Reconcile failed — unknown money state. Keep the client's own
			// classification below (the switch on sqErr) rather than making a
			// money decision on partial data.
		case sqRefundID != nil:
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET status = 'completed'
				WHERE id = ANY($1) AND status = 'pending'
			`, idsOf(pending)); upErr != nil {
				log.Printf("Failed to resolve refunds completed after REFUND_AMOUNT_INVALID on already-refunded charge %s: %v", chargeID, upErr)
			}
			log.Printf("Charge %s already refunded at Square (REFUND_AMOUNT_INVALID + exact-amount COMPLETED refund %s) — marked %d refund row(s) completed, no admin notification", chargeID, *sqRefundID, len(pending))
			return len(pending), nil
		default:
			// No exact-amount COMPLETED refund exists — REFUND_AMOUNT_INVALID
			// here is a genuine decline/invalid amount (the attempt would have
			// over-refunded a partially-refunded payment), NOT an
			// already-refunded signal. Route through the decline branch.
			sqErr = square.ErrRefundDeclined
		}
	}
	switch {
	case sqErr == nil:
		// Resolve by Square's status: only COMPLETED resolves the group;
		// non-terminal states (PENDING — a later sweep reconciles them via
		// ListPaymentRefunds — APPROVED, CANCELED, unknown) leave the rows
		// pending; FAILED/REJECTED is a definitive failure that must not be
		// marked completed (that would block the amount in the over-refund
		// guard forever).
		sqStatus, sqTerminal := SquareRefundStatusToLocal(sqResult.Status)
		if sqResult.Status == "APPROVED" {
			// Loop-B finding (MED-HIGH): APPROVED is NON-terminal — the refund
			// can still transition to FAILED/CANCELED at Square. Resolving it to
			// 'completed' here would strand the row (the FAILED demotion only
			// demotes 'pending'). Keep the rows pending with the square_refund_id
			// recorded so the sweep re-poll and a later FAILED/CANCELED webhook
			// can settle them. COMPLETED stays terminal-completed.
			sqTerminal = false
		}
		if !sqTerminal {
			if sqResult.Status == "APPROVED" {
				// Record the square refund id on the still-pending rows so the
				// sweep's re-poll can look the refund up when it settles.
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET square_refund_id = $1
					WHERE id = ANY($2) AND status = 'pending'
				`, sqResult.ID, idsOf(pending)); upErr != nil {
					log.Printf("Failed to record square refund id %s on pending group for charge %s: %v", sqResult.ID, chargeID, upErr)
				}
			}
			log.Printf("Square refund %s for charge %s is non-terminal (%s) — leaving refunds pending for the sweep", sqResult.ID, chargeID, sqResult.Status)
		} else if sqStatus == "failed" {
			log.Printf("Square refund %s for charge %s FAILED — marking refunds failed", sqResult.ID, chargeID)
		}
		// ATOMIC — one statement for the whole group, never per-row. Keeps
		// crash-retry amounts identical so Square's key-dedup returns the
		// original refund. Only terminal states write a status: a non-terminal
		// outcome must leave the rows pending untouched.
		if sqTerminal {
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET status = $1, square_refund_id = $2
				WHERE id = ANY($3) AND status = 'pending'
			`, sqStatus, sqResult.ID, idsOf(pending)); upErr != nil {
				log.Printf("CRITICAL: Square refund committed (%s) but DB update for charge %s failed — manual reconciliation required: %v", sqResult.ID, chargeID, upErr)
			}
		}
		return len(pending), nil

	case errors.Is(sqErr, square.ErrRefundAlreadyProcessed):
		// PAYMENT_ALREADY_REFUNDED — money already moved at Square.
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET status = 'completed'
			WHERE id = ANY($1) AND status = 'pending'
		`, idsOf(pending)); upErr != nil {
			log.Printf("Failed to resolve refunds after PAYMENT_ALREADY_REFUNDED (charge %s): %v", chargeID, upErr)
		}
		return len(pending), nil

	case errors.Is(sqErr, square.ErrRefundDeclined):
		// Definitive decline — money will never move. Bump attempts; at
		// maxManualRefundAttempts mark failed and surface for manual arrangement.
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET refund_attempts = refund_attempts + 1
			WHERE id = ANY($1) AND status = 'pending'
		`, idsOf(pending)); upErr != nil {
			log.Printf("Failed to increment refund attempts (charge %s): %v", chargeID, upErr)
		}
		if capIDs := pendingRowsAtAttemptCap(ctx, idsOf(pending)); len(capIDs) > 0 {
			if _, upErr := db.Conn.Exec(ctx, fmt.Sprintf(`
				UPDATE refunds SET status = 'failed'
				WHERE id = ANY($1) AND status = 'pending' AND refund_attempts >= %d
			`, maxManualRefundAttempts), capIDs); upErr != nil {
				log.Printf("Failed to mark refunds failed after %d attempts (charge %s): %v", maxManualRefundAttempts, chargeID, upErr)
			}
			insertRefundFailedNotifications(ctx, capIDs)
			// TODO (P6 email/SMS delivery): notify user AND admin when the email/SMS
			// system lands; until then the admin_notifications row above is the only
			// channel. Arrange in-person cash pickup at the salon (a day's notice for
			// cash on hand).
			log.Printf("Card refund for charge %s definitively declined by Square — marked 'failed' and admin notified; TODO email user+admin to arrange in-person cash pickup at the salon (give at least a day's notice for cash on hand)", chargeID)
		}
		return 0, nil

	default:
		// Ambiguous — Square may or may not have processed. Retried by the
		// sweep, capped at maxManualRefundAttempts attempts.
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET refund_attempts = refund_attempts + 1
			WHERE id = ANY($1) AND status = 'pending'
		`, idsOf(pending)); upErr != nil {
			log.Printf("Failed to increment refund attempts (charge %s): %v", chargeID, upErr)
		}
		// At the cap, money may have moved at Square despite the ambiguous
		// responses — reconcile BEFORE marking failed (same bug class as the
		// age guard). An exact COMPLETED refund resolves to completed.
		if capIDs := pendingRowsAtAttemptCap(ctx, idsOf(pending)); len(capIDs) > 0 {
			sqRefundID, rcErr := reconcileRefundAtSquare(ctx, chargeID, totalPence, oldest)
			switch {
			case rcErr != nil:
				// Reconcile failed — unknown whether Square refunded. Leave
				// rows pending for the next sweep; NEVER mark failed on an
				// unknown state (that would let the over-refund guard exclude
				// moved money). But leaving the rows AT the cap strands them:
				// the sweep only re-picks rows with refund_attempts <
				// maxManualRefundAttempts, so capped rows are never
				// re-reconciled and never admin-notified — silently stuck
				// money (A5c). Re-arm the capped rows under the cap (mirroring
				// resolveManualRefundAtCap) so the next sweep re-picks them,
				// and track consecutive reconcile failures: after
				// maxConsecutiveReconcileFailures consecutive failures a
				// deduped 'critical_payment_log' admin notification surfaces
				// the hard-failing reconcile.
				if _, upErr := db.Conn.Exec(ctx, fmt.Sprintf(`
					UPDATE refunds SET refund_attempts = %d
					WHERE id = ANY($1) AND status = 'pending' AND refund_attempts >= %d
				`, maxManualRefundAttempts-1, maxManualRefundAttempts), capIDs); upErr != nil {
					log.Printf("Failed to re-arm capped refunds under the attempt cap after reconcile error (charge %s): %v", chargeID, upErr)
				}
				trackReconcileFailureReArm(ctx, capIDs)
				log.Printf("Reconcile failed for ambiguous charge %s (%v) — re-armed %d refund row(s) under the attempt cap for the next sweep; never marked failed on an unknown state", chargeID, rcErr, len(capIDs))
			case sqRefundID != nil:
				resetReconcileFailureCount(capIDs...)
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'completed', square_refund_id = $1
					WHERE id = ANY($2) AND status = 'pending'
				`, *sqRefundID, capIDs); upErr != nil {
					log.Printf("Failed to mark ambiguous refunds completed after Square reconcile (charge %s): %v", chargeID, upErr)
				}
				log.Printf("Ambiguous card refunds for charge %s reconciled at Square — COMPLETED refund %s found, marked completed", chargeID, *sqRefundID)
			default:
				resetReconcileFailureCount(capIDs...)
				if _, upErr := db.Conn.Exec(ctx, fmt.Sprintf(`
					UPDATE refunds SET status = 'failed'
					WHERE id = ANY($1) AND status = 'pending' AND refund_attempts >= %d
				`, maxManualRefundAttempts), capIDs); upErr != nil {
					log.Printf("Failed to mark refunds failed after %d attempts (charge %s): %v", maxManualRefundAttempts, chargeID, upErr)
				}
				insertRefundFailedNotifications(ctx, capIDs)
				// TODO (P6 email/SMS delivery): notify user AND admin when the email/SMS
				// system lands; until then the admin_notifications row above is the only
				// channel.
				log.Printf("Card refund for charge %s is AMBIGUOUS (Square may have processed it) and no COMPLETED refund found at Square — marked 'failed' and admin notified; TODO email user+admin, VERIFY Square dashboard before arranging in-person cash pickup at the salon (give at least a day's notice for cash on hand)", chargeID)
			}
		}
		return 0, nil
	}
}

func idsOf(rows []pendingChargeRow) []string {
	ids := make([]string, 0, len(rows))
	for _, r := range rows {
		ids = append(ids, r.ID)
	}
	return ids
}

// aggRefundKeySuffix returns a deterministic 12-hex-char suffix for a set of
// identifiers. The identifiers (CHAR(12) refund ids, or an over-length
// square_payment_id) are sorted, sha256'd and truncated, so the SAME input
// always yields the SAME suffix. Used to compress an over-length chargeID into
// a fixed-width prefix for the charge-level idempotency key (see chargeAggKey).
func aggRefundKeySuffix(ids []string) string {
	sorted := append([]string(nil), ids...)
	sort.Strings(sorted)
	h := sha256.Sum256([]byte(strings.Join(sorted, "")))
	return fmt.Sprintf("%x", h)[:12]
}

// chargeAggKey builds the charge-level idempotency key for an aggregated
// refund as <chargeID>-square-agg. The key depends ONLY on the charge ID —
// NEVER on the set of pending row IDs — so it is stable across sweep runs even
// when a new cancellation refund row joins the group (see processChargeGroup).
// Square's idempotency-key limit is 45 chars; the verbatim form needs the
// chargeID ≤34 chars. square_payment_id is an arbitrary TEXT column holding
// Square's real payment ID (typically 20-28 chars), so the verbatim form can
// exceed the limit — and a >45-char key is rejected with a 400
// INVALID_REQUEST_ERROR (classified ambiguous → stuck pending forever). When
// the verbatim form does not fit, the chargeID is sha256'd into a fixed-width
// prefix instead — NEVER truncated verbatim: two charges sharing a truncated
// prefix would collide on Square's global key dedup and silently swallow the
// second charge's refund (lost money).
func chargeAggKey(chargeID string) string {
	key := chargeID + "-square-agg"
	if len(key) <= maxIdempotencyKeyLength {
		return key
	}
	return aggRefundKeySuffix([]string{chargeID}) + "-square-agg"
}

// manualPendingRow is one stale manual refund row eligible for the sweep's
// resolution. It covers BOTH pending shapes the RefundPayment handler can
// leave behind:
//   - square_refund_id set: the handler's synchronous-PENDING response (Square
//     already holds the refund, status='pending') — reconciled, never re-issued.
//   - square_refund_id NULL: the handler's ambiguous-error path — re-issued
//     with the row's OWN stored idempotency key.
type manualPendingRow struct {
	ID        string
	PaymentID string
	// BookingID is the payment's booking_id ("" when NULL = gift-card purchase;
	// such manual refunds are never re-issued, see processManualPaymentGroup).
	BookingID       string
	Amount          float64
	IdempotencyKey  string
	Reason          string
	SquarePaymentID string
	SquareRefundID  string // set when the handler's synchronous-PENDING path stored the refund id
	CreatedAt       time.Time
}

// sweepManualPendingSquareRefunds retries stale MANUAL refunds left 'pending'
// by the RefundPayment handler. The cancellation passes filter
// origin='cancellation', so manual rows were never re-attempted: they
// permanently blocked the over-refund guard and depressed booking TotalPaid.
// Rows are split per-row by their stored square_refund_id: rows WITH one (the
// handler's synchronous-PENDING response) are reconciled at Square — never
// re-issued; rows WITHOUT one (the ambiguous-error path) are re-issued with
// their OWN stored idempotency key (Square dedups same-key retries, so the
// retry is idempotent).
//
// A terminal pre-pass first sweeps legacy manual rows whose payment has NO
// square_payment_id (pre-dating the handler's square-less guard at
// handlers.go:2477-2480). Such rows can never be refunded via Square and would
// otherwise stay 'pending' forever, blocking the over-refund guard. They are
// marked 'failed' and surfaced in the admin notification centre, mirroring the
// Square-less cancellation pre-pass (refunds.go:554-583) with a DISTINCT
// origin='manual' filter so the two passes never double-process a row. Rows
// WITH a square_refund_id are exempt: a sweep auto-refund of a replay-induced
// duplicate charge (B1) attaches its refunds row to the still-pending parent
// payment (which has no square_payment_id) — demoting it here would kill an
// in-flight refund whose parent Square later settles.
//
// A B1 re-poll pass runs FIRST (sweepPendingB1Refunds): it re-polls sweep
// auto-refunds Square left PENDING by square_refund_id and, ONLY when Square
// reports the refund COMPLETED, marks the parent payment/till-sale row failed
// and claws back a funded gift card — the duplicate charge has been reversed
// and only then is the money state settled.
func sweepManualPendingSquareRefunds(ctx context.Context) (int, error) {
	// (a0) B1-origin sweep auto-refunds (refunds rows with a square_refund_id
	//      that Square left PENDING): re-poll each by square_refund_id and
	//      resolve the parent row ONLY on COMPLETED. Runs before the passes
	//      below so an in-flight B1 refund is never demoted or re-issued.
	b1Count, b1Err := sweepPendingB1Refunds(ctx)
	if b1Err != nil {
		log.Printf("Failed to re-poll B1 sweep auto-refunds: %v", b1Err)
	}

	// (a) Terminal pre-pass: legacy MANUAL card refunds whose payment has no
	//     Square reference can never be refunded via Square → mark them failed
	//     so they stop blocking the over-refund guard, and surface the affected
	//     booking in the admin notification centre for in-person arrangement.
	//     Mirrors the cancellation Square-less pre-pass (origin='cancellation',
	//     refunds.go:554-583) with origin='manual' so the filters stay distinct
	//     and no row is swept by both passes. Must run OUTSIDE any GROUP BY —
	//     Postgres lumps NULLs together, so these rows can't be handled in the
	//     per-payment grouping below. Rows WITH a square_refund_id are exempt —
	//     a B1 auto-refund attaches to a parent payment with no square_payment_id
	//     and its in-flight refund must not be demoted to failed.
	rows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		UPDATE refunds r SET status = 'failed'
		FROM payments p
		WHERE p.id = r.payment_id
		  AND r.status = 'pending' AND r.refund_attempts < %d
		  AND p.payment_method IN ('online_square', 'in_person_card')
		  AND p.square_payment_id IS NULL
		  AND r.origin = 'manual'
		  AND r.square_refund_id IS NULL
		RETURNING r.id
	`, maxManualRefundAttempts))
	if err != nil {
		log.Printf("Failed to mark Square-less manual refunds failed: %v", err)
	} else {
		var failedIDs []string
		for rows.Next() {
			var id string
			if err := rows.Scan(&id); err == nil {
				failedIDs = append(failedIDs, id)
			}
		}
		if err := rows.Err(); err != nil {
			log.Printf("Failed to iterate Square-less manual refunds during sweep: %v", err)
		}
		rows.Close()
		if len(failedIDs) > 0 {
			log.Printf("Marked %d Square-less manual refund(s) failed (in-person arrangement needed)", len(failedIDs))
		}
		insertRefundFailedNotifications(ctx, failedIDs)
	}

	// (b) Manual refunds WITH a Square reference — the rows below are the only
	//     ones the retry/reconcile logic can act on.
	rows, err = db.Conn.Query(ctx, fmt.Sprintf(`
		SELECT r.id, r.payment_id, p.booking_id, r.amount, r.idempotency_key, r.reason,
		       p.square_payment_id, r.square_refund_id, r.created_at
		FROM refunds r
		JOIN payments p ON p.id = r.payment_id
		WHERE r.status = 'pending' AND r.origin = 'manual'
		  AND r.refund_attempts < %d
		  AND p.square_payment_id IS NOT NULL
		ORDER BY r.payment_id, r.id
	`, maxManualRefundAttempts))
	if err != nil {
		log.Printf("Failed to query manual pending refunds for retry: %v", err)
		return 0, nil
	}
	var pending []manualPendingRow
	for rows.Next() {
		var pr manualPendingRow
		var key *string
		var sqRefundID *string
		var bookingID sql.NullString
		if err := rows.Scan(&pr.ID, &pr.PaymentID, &bookingID, &pr.Amount, &key, &pr.Reason, &pr.SquarePaymentID, &sqRefundID, &pr.CreatedAt); err != nil {
			log.Printf("Failed to scan manual pending refund: %v", err)
			continue
		}
		pr.BookingID = bookingID.String
		if key != nil {
			pr.IdempotencyKey = *key
		}
		if sqRefundID != nil {
			pr.SquareRefundID = *sqRefundID
		}
		pending = append(pending, pr)
	}
	rows.Close()
	if len(pending) == 0 {
		return 0, nil
	}

	// Group by payment_id, ascending — matching the lock ordering the manual
	// handler and processChargeGroup use so the sweep never deadlocks them.
	groups := make(map[string][]manualPendingRow)
	var paymentIDs []string
	for _, pr := range pending {
		if _, ok := groups[pr.PaymentID]; !ok {
			paymentIDs = append(paymentIDs, pr.PaymentID)
		}
		groups[pr.PaymentID] = append(groups[pr.PaymentID], pr)
	}
	sort.Strings(paymentIDs)

	processed := 0
	for _, pid := range paymentIDs {
		n, err := processManualPaymentGroup(ctx, pid, groups[pid])
		if err != nil {
			log.Printf("Sweep: failed to process manual pending refunds for payment %s: %v", pid, err)
			continue
		}
		processed += n
	}
	return processed + b1Count, nil
}

// b1PendingRefund is one refunds row for a sweep auto-refund of a
// replay-induced duplicate charge (B1) that Square left PENDING — or that a
// webhook promoted to 'completed' while its PARENT payment row is still
// pending (the parent is only ever resolved by this re-poll pass).
type b1PendingRefund struct {
	RefundID        string
	PaymentID       string // refunds.payment_id — the parent payment row
	Amount          float64
	SquareRefundID  string
	IdempotencyKey  string // deterministic "sweepdup-" + duplicate payment id (payments-table rows)
	Reason          string // carries the parent till_sale id for till_sale rows
	SquarePaymentID string // synthetic till-sale payment row's square_payment_id ("" for payments-table rows)
	CreatedBy       string
	CreatedAt       time.Time // refunds.created_at — when the auto-refund was recorded (age for escalation)
	BookingID       string    // parent payment's booking_id ("" when the row has none)
	Status          string    // refunds.status — 'pending', or 'completed' (webhook-promoted, parent unresolved)
}

// sweepPendingB1Refunds re-polls refunds rows for sweep auto-refunds of
// replay-induced duplicate charges (B1, refundSweepDuplicateCharge in sweep.go)
// that Square left PENDING, AND refunds a webhook promoted to 'completed'
// while the PARENT row is still pending (Loop-B finding: the webhook promotes
// the refund status but the parent is resolved only here — querying only
// 'pending' refunds let a promoted refund slip through, the sweep's in-flight
// guard then missed it and re-replayed the expired key, minting ANOTHER charge).
// The refund is NON-terminal while the parent is pending: the parent row must
// stay pending (never marked failed, a funded gift card never clawed back)
// until Square settles. When Square reports the refund COMPLETED the parent is
// finally resolved — the pending payment marked failed, and a till sale's
// funded gift card clawed back (the duplicate charge has been reversed; only
// then is the money state settled). FAILED/REJECTED refunds are left pending
// for the webhook FAILED-refund reconciliation (coordinated); a reconcile
// error is an UNKNOWN state and is never resolved here. The stale-pending
// sweeps skip rows with an in-flight B1 refund (hasInFlightSweepDuplicateRefund,
// sweep.go), so the parent is only ever resolved from here.
func sweepPendingB1Refunds(ctx context.Context) (int, error) {
	rows, err := db.Conn.Query(ctx, `
		SELECT r.id, r.payment_id, r.amount, r.square_refund_id, COALESCE(r.idempotency_key, ''),
		       r.reason, COALESCE(p.square_payment_id, ''), COALESCE(r.created_by, ''),
		       r.created_at, COALESCE(p.booking_id, ''), r.status
		FROM refunds r
		LEFT JOIN payments p ON p.id = r.payment_id
		WHERE r.square_refund_id IS NOT NULL
		  AND r.origin = 'manual' AND r.reason LIKE 'duplicate charge — sweep replay%'
		  AND (
		      r.status = 'pending'
		      OR (r.status = 'completed' AND (
		          p.status = 'pending'
		          OR EXISTS (SELECT 1 FROM till_sales ts
		                     WHERE ts.status = 'pending'
		                       AND ts.id = SUBSTRING(r.reason FROM '\(till_sale ([^)]+)\)'))
		      ))
		  )
		ORDER BY r.id
	`)
	if err != nil {
		log.Printf("Failed to query B1 sweep auto-refunds for re-poll: %v", err)
		return 0, nil
	}
	defer rows.Close()
	var pending []b1PendingRefund
	for rows.Next() {
		var pr b1PendingRefund
		if err := rows.Scan(&pr.RefundID, &pr.PaymentID, &pr.Amount, &pr.SquareRefundID, &pr.IdempotencyKey, &pr.Reason, &pr.SquarePaymentID, &pr.CreatedBy, &pr.CreatedAt, &pr.BookingID, &pr.Status); err != nil {
			log.Printf("Failed to scan B1 sweep auto-refund: %v", err)
			continue
		}
		pending = append(pending, pr)
	}
	if err := rows.Err(); err != nil {
		log.Printf("Failed to iterate B1 sweep auto-refunds: %v", err)
		return 0, nil
	}
	if len(pending) == 0 {
		return 0, nil
	}

	processed := 0
	for i := range pending {
		pr := &pending[i]
		// A row already escalated (pending past stalePendingB1RefundAge) is no
		// longer re-polled — it either resolved terminal (and left the pending
		// query) or stays pending under the deduped CRITICAL notification for
		// manual reconciliation. A refund a webhook promoted to 'completed'
		// AFTER the escalation is re-polled once: the parent can finally be
		// resolved (resolveB1RefundCompleted clears the escalation flag).
		b1EscalatedMu.Lock()
		escalated := b1Escalated[pr.RefundID]
		b1EscalatedMu.Unlock()
		if escalated && pr.Status == "pending" {
			log.Printf("B1 refund %s was already escalated (pending over %s) — not re-polling; manual reconciliation holds the parent pending", pr.RefundID, stalePendingB1RefundAge)
			continue
		}
		if escalated {
			log.Printf("B1 refund %s was escalated but a webhook promoted it to completed — re-polling to resolve the parent", pr.RefundID)
		}
		// The Square payment id the refund targets. A till_sale's refund is
		// attached to the synthetic payments row (square_payment_id = the
		// duplicate charge); a payments-table refund is attached to the
		// still-pending payment row (no square_payment_id), so the duplicate id
		// is recovered from the deterministic refund idempotency key.
		dupPayID := pr.SquarePaymentID
		if dupPayID == "" {
			dupPayID = strings.TrimPrefix(pr.IdempotencyKey, "sweepdup-")
			if dupPayID == pr.IdempotencyKey {
				log.Printf("Cannot re-poll B1 refund %s: no Square payment id and the stored key %q is not the sweepdup form — leaving pending", pr.RefundID, pr.IdempotencyKey)
				continue
			}
		}
		refundList, lErr := SquareClient.ListPaymentRefunds(ctx, dupPayID, time.Time{})
		if lErr != nil {
			log.Printf("Re-poll of B1 refund %s failed (%v) — leaving pending", pr.RefundID, lErr)
			continue
		}
		status := ""
		for j := range refundList {
			if refundList[j].ID == pr.SquareRefundID {
				status = refundList[j].Status
				break
			}
		}
		stale := clock.Now().Sub(pr.CreatedAt) > stalePendingB1RefundAge
		switch status {
		case "COMPLETED":
			if resolveB1RefundCompleted(ctx, pr) {
				processed++
			}
		case "PENDING", "APPROVED":
			if stale {
				escalateStaleB1Refund(ctx, pr)
			} else {
				log.Printf("B1 refund %s is still %q at Square — leaving the parent row pending", pr.RefundID, status)
			}
		case "":
			log.Printf("B1 refund %s (%s) was not found at Square via ListPaymentRefunds — leaving pending; manual reconciliation may be required", pr.RefundID, dupPayID)
		default:
			// FAILED / REJECTED — terminal at Square. While the refund is young
			// the webhook FAILED-refund reconciliation (OPTIONAL per README)
			// owns it; once it has been pending past stalePendingB1RefundAge the
			// sweep treats it as terminal itself so the parent can never be
			// stranded by a missing webhook.
			if stale {
				if resolveB1RefundFailedTerminal(ctx, pr, status) {
					processed++
				}
			} else {
				log.Printf("B1 refund %s is %q at Square — leaving pending for the webhook FAILED-refund reconciliation (or the %s age escalation)", pr.RefundID, status, stalePendingB1RefundAge)
			}
		}
	}
	return processed, nil
}

// resolveB1RefundCompleted resolves a B1 sweep auto-refund that Square reports
// COMPLETED: the refunds row is marked completed (money moved), and the parent
// row — the pending payment, or the pending till sale whose gift-card funding
// is clawed back — is resolved to failed. Returns true when the parent was
// resolved.
func resolveB1RefundCompleted(ctx context.Context, pr *b1PendingRefund) bool {
	if _, err := db.Conn.Exec(ctx, `
		UPDATE refunds SET status = 'completed'
		WHERE id = $1 AND status = 'pending'
	`, pr.RefundID); err != nil {
		log.Printf("Failed to mark B1 refund %s completed after Square settle: %v", pr.RefundID, err)
	}
	// A COMPLETED refund resolved the parent — clear any escalation flag so a
	// restarted/re-polled row can be observed again if it ever re-enters.
	b1EscalatedMu.Lock()
	delete(b1Escalated, pr.RefundID)
	b1EscalatedMu.Unlock()
	return resolveB1ParentFailed(ctx, pr)
}

// resolveB1RefundFailedTerminal resolves a B1 sweep auto-refund that Square
// reports FAILED/REJECTED once it has been pending longer than
// stalePendingB1RefundAge — the terminal treatment the webhook FAILED-refund
// reconciliation (OPTIONAL per README) would have applied when configured. The
// refund is marked failed, the parent resolved to failed (a till sale's funded
// gift card clawed back), and a deduped CRITICAL admin notification raised so
// the missing webhook can never strand the parent silently forever. Returns
// true when the parent was resolved.
func resolveB1RefundFailedTerminal(ctx context.Context, pr *b1PendingRefund, squareStatus string) bool {
	if _, err := db.Conn.Exec(ctx, `
		UPDATE refunds SET status = 'failed'
		WHERE id = $1 AND status = 'pending'
	`, pr.RefundID); err != nil {
		log.Printf("Failed to mark B1 refund %s failed after %q at Square: %v", pr.RefundID, squareStatus, err)
	}
	b1EscalatedMu.Lock()
	b1Escalated[pr.RefundID] = true
	b1EscalatedMu.Unlock()
	insertB1EscalationNotification(ctx, pr)
	resolved := resolveB1ParentFailed(ctx, pr)
	log.Printf("B1 refund %s was %q at Square and pending over %s — marked the refund failed and resolved the parent; MANUAL RECONCILIATION REQUIRED: verify at Square whether the duplicate charge was refunded", pr.RefundID, squareStatus, stalePendingB1RefundAge)
	return resolved
}

// resolveB1ParentFailed resolves a B1 refund's parent row to failed after the
// refund settled definitively (COMPLETED — the duplicate was reversed — or
// FAILED/REJECTED past the age threshold). A payments-table refund's payment_id
// IS the pending parent row; a till_sale's id is encoded in the reason and its
// funded gift card is clawed back. Returns true when the parent was resolved.
func resolveB1ParentFailed(ctx context.Context, pr *b1PendingRefund) bool {
	if pr.Reason == sweepDuplicateRefundReason {
		// payments-table parent: the refund's payment_id IS the pending row.
		if failStaleRow(ctx, "payments", pr.PaymentID) {
			log.Printf("B1 refund %s — marked the pending payment %s failed", pr.RefundID, pr.PaymentID)
			return true
		}
		log.Printf("B1 refund %s — payment %s was already resolved", pr.RefundID, pr.PaymentID)
		return false
	}
	// till_sale parent: the sale id is encoded in the reason.
	if idx := strings.Index(pr.Reason, "(till_sale "); idx >= 0 {
		tillSaleID := strings.TrimSuffix(pr.Reason[idx+len("(till_sale "):], ")")
		ts, ok := loadTillSaleStaleRow(ctx, tillSaleID)
		if !ok {
			log.Printf("CRITICAL: B1 refund %s — parent till sale %s could not be loaded — MANUAL RECONCILIATION REQUIRED: gift card may still be funded", pr.RefundID, tillSaleID)
			return false
		}
		if ts.HasGiftCard {
			if clawbackTillSaleFunding(ctx, ts) {
				log.Printf("B1 refund %s — clawed back till sale %s's funded gift card and marked it failed", pr.RefundID, tillSaleID)
				return true
			}
			log.Printf("CRITICAL: B1 refund %s — clawing back till sale %s's funding failed — MANUAL RECONCILIATION REQUIRED: gift card may still be funded", pr.RefundID, tillSaleID)
			return false
		}
		if failStaleRow(ctx, "till_sales", tillSaleID) {
			log.Printf("B1 refund %s — marked till sale %s failed", pr.RefundID, tillSaleID)
			return true
		}
		log.Printf("B1 refund %s — till sale %s was already resolved", pr.RefundID, tillSaleID)
		return false
	}
	log.Printf("B1 refund %s — the parent row could not be identified from reason %q — MANUAL RECONCILIATION REQUIRED", pr.RefundID, pr.Reason)
	return false
}

// escalateStaleB1Refund surfaces a B1 sweep auto-refund that has been PENDING
// (non-terminal) at Square for longer than stalePendingB1RefundAge: a deduped
// CRITICAL admin notification fires (the webhook FAILED-refund reconciliation
// is OPTIONAL — with it unconfigured nothing else alerts) and the row is no
// longer re-polled. The refund is left PENDING and the parent stays pending —
// marking failed on a non-terminal state could exclude money that may still
// move.
func escalateStaleB1Refund(ctx context.Context, pr *b1PendingRefund) {
	b1EscalatedMu.Lock()
	b1Escalated[pr.RefundID] = true
	b1EscalatedMu.Unlock()
	insertB1EscalationNotification(ctx, pr)
	log.Printf("CRITICAL: B1 refund %s has been PENDING at Square for over %s — leaving the parent pending and STOPPING re-poll — MANUAL RECONCILIATION REQUIRED: verify the refund at Square and resolve the parent", pr.RefundID, stalePendingB1RefundAge)
}

// insertB1EscalationNotification raises the deduped 'critical_payment_log'
// admin notification for an escalated B1 refund (insertCriticalPaymentNotification
// keeps ONE per booking/user until acknowledged). A payments-table refund is
// attributed to its parent payment's booking and the payer; a till-sale refund
// (synthetic payment row, no booking) is attributed to the payer only.
func insertB1EscalationNotification(ctx context.Context, pr *b1PendingRefund) {
	var bookingID *string
	if pr.BookingID != "" {
		b := pr.BookingID
		bookingID = &b
	}
	var userID *string
	if pr.CreatedBy != "" {
		u := pr.CreatedBy
		userID = &u
	}
	insertCriticalPaymentNotification(ctx, bookingID, userID)
}

// loadTillSaleStaleRow reads a till_sale's gift-card context for the B1
// re-poll pass's clawback — the same fields fetchStaleRows/scanStaleRow
// populate for the stale-pending sweep.
func loadTillSaleStaleRow(ctx context.Context, id string) (staleRow, bool) {
	var r staleRow
	var itemID, redeemedBy sql.NullString
	var isCreate *bool
	var hasGiftCard bool
	var total float64
	err := db.Conn.QueryRow(ctx, `
		SELECT ts.id, ts.item_id, gc.redeemed_by, (ts.created_at = gc.created_at) AS is_create,
		       (gc.id IS NOT NULL) AS has_gift_card, ts.total_amount
		FROM till_sales ts
		LEFT JOIN gift_cards gc ON gc.id = ts.item_id
		WHERE ts.id = $1
	`, id).Scan(&r.ID, &itemID, &redeemedBy, &isCreate, &hasGiftCard, &total)
	if err != nil {
		log.Printf("Failed to load till sale %s for B1 refund clawback: %v", id, err)
		return r, false
	}
	r.ItemID = itemID.String
	if redeemedBy.Valid && redeemedBy.String != "" {
		r.RedeemToUserID = &redeemedBy.String
	}
	r.IsCreate = isCreate != nil && *isCreate
	r.HasGiftCard = hasGiftCard
	r.TotalAmount = total
	r.AmountPence = int64(math.Round(total * 100))
	return r, true
}

// ensureRefundKey returns the idempotency key to use when re-issuing a manual
// refund, persisting a generated fallback to the refunds row BEFORE Square is
// called so every retry reuses the SAME key — Square dedups same-key retries,
// so a lost-response retry can never issue a SECOND refund.
//
// A legacy refund row with a NULL idempotency_key used to get a fresh random
// suffix on every resume; if the Square call succeeded but the follow-up DB
// UPDATE to 'completed' failed (the CRITICAL-log path), the row stayed
// 'pending' with its key STILL NULL and the next resume generated a NEW random
// key → a second Square refund (double refund). Persisting the key first
// closes that hole: a crash-retry re-reads the persisted key and reuses it.
//
// The `AND idempotency_key IS NULL` guard makes the persist race-safe: only
// one concurrent caller wins the UPDATE; a loser (0 rows affected) re-reads
// and returns the winner's key. The generated shape (paymentID + "-refund-" +
// amount + "-" + 12 hex chars) stays ≤45 chars: 12 + 8 + up-to-9 + 1 + 12 ≈ 42.
func ensureRefundKey(ctx context.Context, refundID, paymentID string, amount int64, storedKey string) (string, error) {
	if storedKey != "" {
		return storedKey, nil
	}
	key := paymentID + "-refund-" + strconv.FormatInt(amount, 10) + "-" + randomHexSuffix(6)
	tag, err := db.Conn.Exec(ctx, `
		UPDATE refunds SET idempotency_key = $1
		WHERE id = $2 AND idempotency_key IS NULL
	`, key, refundID)
	if err != nil {
		return "", fmt.Errorf("failed to persist refund idempotency key for %s: %w", refundID, err)
	}
	if tag.RowsAffected() == 0 {
		var existing string
		if err := db.Conn.QueryRow(ctx, `SELECT idempotency_key FROM refunds WHERE id = $1`, refundID).Scan(&existing); err != nil {
			return "", fmt.Errorf("failed to re-read persisted refund idempotency key for %s: %w", refundID, err)
		}
		return existing, nil
	}
	return key, nil
}

// processManualPaymentGroup retries one payment's stale manual pending refunds
// under the SAME per-payment advisory lock the manual RefundPayment handler
// holds across its guard read — so a re-issued refund can never double-spend
// against a concurrent manual refund. Rows are re-read under the lock; each is
// issued to Square with its OWN stored idempotency key and stored reason.
func processManualPaymentGroup(ctx context.Context, paymentID string, rows []manualPendingRow) (int, error) {
	pinConn, err := db.Conn.Acquire(ctx)
	if err != nil {
		return 0, err
	}
	defer pinConn.Release()
	// Bounded try-lock (R6) so the sweep never blocks a pool connection while
	// the manual handler holds the same key across its Square call.
	lockOK, err := acquireAdvisoryLock(ctx, pinConn, "crussell:refund:"+paymentID)
	if err != nil {
		return 0, err
	}
	if !lockOK {
		log.Printf("Refund lock for payment %s not acquired within bound — a manual refund is in progress; leaving rows pending for the next sweep", paymentID)
		return 0, nil
	}
	defer releasePaymentLock(pinConn, "crussell:refund:"+paymentID)

	ids := make([]string, 0, len(rows))
	for _, r := range rows {
		ids = append(ids, r.ID)
	}
	// Re-read under the lock — only rows still pending and under the attempt
	// cap are eligible (a concurrent manual refund may have resolved some).
	prRows, err := db.Conn.Query(ctx, fmt.Sprintf(`
		SELECT r.id, p.booking_id, r.amount, r.idempotency_key, r.reason, r.created_at,
		       r.payment_id, p.square_payment_id, r.square_refund_id
		FROM refunds r
		JOIN payments p ON p.id = r.payment_id
		WHERE r.id = ANY($1) AND r.status = 'pending' AND r.refund_attempts < %d
		ORDER BY r.id
	`, maxManualRefundAttempts), ids)
	if err != nil {
		return 0, err
	}
	var pending []manualPendingRow
	for prRows.Next() {
		var pr manualPendingRow
		var key *string
		var sqRefundID *string
		var bookingID sql.NullString
		if err := prRows.Scan(&pr.ID, &bookingID, &pr.Amount, &key, &pr.Reason, &pr.CreatedAt, &pr.PaymentID, &pr.SquarePaymentID, &sqRefundID); err != nil {
			log.Printf("Failed to scan manual pending refund under lock: %v", err)
			continue
		}
		pr.BookingID = bookingID.String
		if key != nil {
			pr.IdempotencyKey = *key
		}
		if sqRefundID != nil {
			pr.SquareRefundID = *sqRefundID
		}
		pending = append(pending, pr)
	}
	prRows.Close()
	if len(pending) == 0 {
		return 0, nil
	}

	// Age guard (mirrors processChargeGroup): Square's idempotency-key
	// retention is finite (~24h). Reconcile FIRST — an exact COMPLETED refund
	// at Square resolves to completed even though our rows are stale.
	oldest := pending[0].CreatedAt
	for _, pr := range pending[1:] {
		if pr.CreatedAt.Before(oldest) {
			oldest = pr.CreatedAt
		}
	}
	if clock.Now().Sub(oldest) > stalePendingRefundAge {
		processedAged := 0
		for i := range pending {
			pr := &pending[i]
			amountPence := int64(math.Round(pr.Amount * 100))
			sqRefundID, rcErr := reconcileRefundAtSquare(ctx, pr.SquarePaymentID, amountPence, pr.CreatedAt)
			switch {
			case rcErr != nil:
				// Reconcile failed — unknown whether Square refunded. Leave
				// this row pending for the next sweep; NEVER mark failed on an
				// unknown state. B18: track consecutive reconcile failures and
				// surface the hard-failing reconcile in the admin notification
				// centre (mirrors resolveManualRefundAtCap) so the row does not
				// oscillate here silently forever.
				trackReconcileFailureReArm(ctx, []string{pr.ID})
				log.Printf("Reconcile failed for aged manual refund %s (%v) — leaving pending for the next sweep", pr.ID, rcErr)
			case sqRefundID != nil:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'completed', square_refund_id = $1
					WHERE id = $2 AND status = 'pending'
				`, *sqRefundID, pr.ID); upErr != nil {
					log.Printf("Failed to mark aged manual refund %s completed after Square reconcile: %v", pr.ID, upErr)
				}
				// A5: terminal resolution — clear the consecutive-failure counter.
				resetReconcileFailureCount(pr.ID)
				processedAged++
			default:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'failed'
					WHERE id = $1 AND status = 'pending'
				`, pr.ID); upErr != nil {
					log.Printf("Failed to mark aged manual refund %s failed: %v", pr.ID, upErr)
				}
				// A5: terminal resolution — clear the consecutive-failure counter.
				resetReconcileFailureCount(pr.ID)
				insertRefundFailedNotifications(ctx, []string{pr.ID})
				// TODO (P6 email/SMS delivery): notify user AND admin when the email/SMS
				// system lands; until then the admin_notifications row above is the only
				// channel.
				log.Printf("Manual refund %s is older than stalePendingRefundAge (23h) and Square shows no COMPLETED refund — marked 'failed' and admin notified; TODO email user+admin to arrange in-person cash pickup at the salon (give at least a day's notice for cash on hand)", pr.ID)
			}
		}
		return processedAged, nil
	}

	processed := 0
	for i := range pending {
		pr := &pending[i]
		amountPence := int64(math.Round(pr.Amount * 100))

		// A payment with NO booking is a gift-card purchase (BuyGiftCard
		// inserts without a booking) — the handler rejects these outright, so a
		// pending row that predates that guard must never be re-issued here
		// (the sweep is a bypass of the handler guard). Reconcile only,
		// exactly like the square_refund_id branch: an exact COMPLETED refund
		// resolves to completed (money already moved); a no-match means Square
		// never refunded — mark failed + admin-notified so the amount unblocks
		// the over-refund guard and the customer is handled via the gift-card
		// section.
		if pr.BookingID == "" {
			sqRefundID, rcErr := reconcileRefundAtSquare(ctx, pr.SquarePaymentID, amountPence, pr.CreatedAt)
			switch {
			case rcErr != nil:
				log.Printf("Reconcile failed for gift-card-purchase manual refund %s (%v) — leaving pending for the next sweep", pr.ID, rcErr)
			case sqRefundID != nil:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'completed', square_refund_id = $1
					WHERE id = $2 AND status = 'pending'
				`, *sqRefundID, pr.ID); upErr != nil {
					log.Printf("Failed to mark gift-card-purchase manual refund %s completed after Square reconcile: %v", pr.ID, upErr)
				}
				// A5: terminal resolution — clear the consecutive-failure counter.
				resetReconcileFailureCount(pr.ID)
				processed++
			default:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'failed'
					WHERE id = $1 AND status = 'pending'
				`, pr.ID); upErr != nil {
					log.Printf("Failed to mark gift-card-purchase manual refund %s failed: %v", pr.ID, upErr)
				}
				// A5: terminal resolution — clear the consecutive-failure counter.
				resetReconcileFailureCount(pr.ID)
				insertRefundFailedNotifications(ctx, []string{pr.ID})
				log.Printf("Gift-card-purchase manual refund %s (payment %s) blocked — payment has no booking; Square shows no COMPLETED refund — marked failed, customer must be refunded via the gift-card section", pr.ID, pr.PaymentID)
			}
			continue
		}

		// Row WITH a stored square_refund_id — the RefundPayment handler's
		// synchronous-PENDING response. Square already holds the refund, so a
		// re-issue would risk a SECOND refund (Square's key dedup does not
		// protect a fresh key). Reconcile instead: an exact COMPLETED refund at
		// Square resolves the row; a genuine no-match means Square never
		// recorded it → failed + admin notification; a reconcile error is an
		// UNKNOWN state → leave pending (never mark failed on an unknown state,
		// that would let the over-refund guard exclude money that may have
		// moved). Mirrors the stalePendingRefundAge age-guard branch above.
		if pr.SquareRefundID != "" {
			sqRefundID, rcErr := reconcileRefundAtSquare(ctx, pr.SquarePaymentID, amountPence, pr.CreatedAt)
			switch {
			case rcErr != nil:
				log.Printf("Reconcile failed for pending manual refund %s (square_refund_id %s, %v) — leaving pending for the next sweep", pr.ID, pr.SquareRefundID, rcErr)
			case sqRefundID != nil:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'completed', square_refund_id = $1
					WHERE id = $2 AND status = 'pending'
				`, *sqRefundID, pr.ID); upErr != nil {
					log.Printf("Failed to mark manual refund %s completed after Square reconcile: %v", pr.ID, upErr)
				}
				processed++
			default:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'failed'
					WHERE id = $1 AND status = 'pending'
				`, pr.ID); upErr != nil {
					log.Printf("Failed to mark manual refund %s failed after Square reconcile showed no refund: %v", pr.ID, upErr)
				}
				insertRefundFailedNotifications(ctx, []string{pr.ID})
				log.Printf("Manual refund %s (square_refund_id %s) has no COMPLETED refund at Square — marked 'failed' and admin notified; TODO email user+admin to VERIFY the Square dashboard before arranging in-person cash pickup at the salon (give at least a day's notice for cash on hand)", pr.ID, pr.SquareRefundID)
			}
			continue
		}

		// Route the stored key through ensureRefundKey: a legacy row with a
		// NULL idempotency_key must NOT reach Square with "" (a 400
		// INVALID_REQUEST_ERROR, classified ambiguous → stuck forever). The
		// helper persists a generated fallback to the row first, so every retry
		// reuses the SAME key — a lost-response retry can never issue a second
		// refund (Square dedups same-key retries).
		idemKey, keyErr := ensureRefundKey(ctx, pr.ID, pr.PaymentID, amountPence, pr.IdempotencyKey)
		if keyErr != nil {
			// The key could not be persisted — Square must not be called with an
			// empty/unknown key. Leave the row pending for the next sweep (never
			// mark failed on an unknown state); the stalePendingRefundAge age
			// guard above will eventually reconcile it.
			log.Printf("Failed to ensure refund key for manual refund %s before re-issue: %v", pr.ID, keyErr)
			continue
		}
		sqResult, sqErr := SquareClient.RefundPayment(ctx, square.RefundPaymentReq{
			PaymentID:      pr.SquarePaymentID,
			Amount:         amountPence,
			IdempotencyKey: idemKey,
			Reason:         pr.Reason,
		})
		// REFUND_AMOUNT_INVALID is Square's ambiguous already-refunded-or-
		// invalid-amount signal. Disambiguate with the EXACT amount (A11): only
		// an exact-amount COMPLETED refund at Square proves THIS amount already
		// moved — a smaller partial refund does NOT, and completing the row
		// would claim the full amount was refunded when only part of it was.
		// Mirrors the processChargeGroup disambiguation so the reconciliation
		// applies whether the client classified the code as a definitive
		// decline or as already-processed.
		if isRefundAmountInvalid(sqErr) {
			sqRefundID, rcErr := reconcileRefundAtSquareExact(ctx, pr.SquarePaymentID, amountPence)
			switch {
			case rcErr != nil:
				// Reconcile failed — unknown money state. Keep the client's
				// own classification below (the switch on sqErr) rather than
				// making a money decision on partial data.
			case sqRefundID != nil:
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = 'completed'
					WHERE id = $1 AND status = 'pending'
				`, pr.ID); upErr != nil {
					log.Printf("Failed to resolve manual refund %s completed after REFUND_AMOUNT_INVALID on already-refunded payment: %v", pr.ID, upErr)
				}
				log.Printf("Manual refund %s: payment already refunded at Square (REFUND_AMOUNT_INVALID + exact-amount COMPLETED refund %s) — marked completed, no admin notification", pr.ID, *sqRefundID)
				processed++
				continue
			default:
				// No exact-amount COMPLETED refund exists — REFUND_AMOUNT_INVALID
				// is a genuine decline/invalid amount here (the attempt would
				// have over-refunded a partially-refunded payment), NOT an
				// already-refunded signal. Route through the decline branch.
				sqErr = square.ErrRefundDeclined
			}
		}
		switch {
		case sqErr == nil:
			// Resolve by Square's status: a synchronous refund response can be
			// PENDING (money in flight, e.g. an async card network) — marking it
			// completed while Square later fails it would permanently block that
			// amount in the over-refund guard. Only a terminal COMPLETED
			// resolves to completed; non-terminal states (PENDING, APPROVED,
			// CANCELED, unknown) stay pending for the sweep to reconcile;
			// FAILED/REJECTED is a real failure. Mirrors processChargeGroup and
			// the RefundPayment handler (handlers.go).
			sqStatus, sqTerminal := SquareRefundStatusToLocal(sqResult.Status)
			if sqResult.Status == "APPROVED" {
				// Loop-B finding (MED-HIGH): APPROVED is NON-terminal — a later
				// FAILED/CANCELED must still be able to demote the row, so it
				// stays pending with the square_refund_id recorded.
				sqTerminal = false
			}
			if !sqTerminal {
				if sqResult.Status == "APPROVED" {
					if _, upErr := db.Conn.Exec(ctx, `
						UPDATE refunds SET square_refund_id = $1
						WHERE id = $2
					`, sqResult.ID, pr.ID); upErr != nil {
						log.Printf("Failed to record square refund id %s on pending manual refund %s: %v", sqResult.ID, pr.ID, upErr)
					}
				}
				log.Printf("Square refund %s for manual refund %s is non-terminal (%s) — leaving the row pending for the sweep to resolve", sqResult.ID, pr.ID, sqResult.Status)
			} else if sqStatus == "failed" {
				log.Printf("Square refund %s for manual refund %s FAILED — marking the row failed", sqResult.ID, pr.ID)
			}
			if !sqTerminal {
				log.Printf("Square refund %s for manual refund %s is non-terminal (%s) — leaving the row pending for the sweep to resolve", sqResult.ID, pr.ID, sqResult.Status)
			} else if sqStatus == "failed" {
				log.Printf("Square refund %s for manual refund %s FAILED — marking the row failed", sqResult.ID, pr.ID)
			}
			if sqTerminal {
				if _, upErr := db.Conn.Exec(ctx, `
					UPDATE refunds SET status = $1, square_refund_id = $2
					WHERE id = $3
				`, sqStatus, sqResult.ID, pr.ID); upErr != nil {
					log.Printf("CRITICAL: Square refund committed (%s) but DB update for manual refund %s failed — manual reconciliation required: %v", sqResult.ID, pr.ID, upErr)
				}
			}
			processed++

		case errors.Is(sqErr, square.ErrRefundAlreadyProcessed):
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET status = 'completed'
				WHERE id = $1
			`, pr.ID); upErr != nil {
				log.Printf("Failed to resolve manual refund %s completed after PAYMENT_ALREADY_REFUNDED: %v", pr.ID, upErr)
			}
			processed++

		case errors.Is(sqErr, square.ErrRefundDeclined):
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET refund_attempts = refund_attempts + 1
				WHERE id = $1
			`, pr.ID); upErr != nil {
				log.Printf("Failed to increment attempts for manual refund %s: %v", pr.ID, upErr)
			}
			if attempts := currentRefundAttempts(ctx, pr.ID); attempts >= maxManualRefundAttempts {
				resolveManualRefundAtCap(ctx, pr, amountPence)
			}

		default:
			// Ambiguous — Square may or may not have processed. Retried by the
			// sweep, capped at maxManualRefundAttempts.
			if _, upErr := db.Conn.Exec(ctx, `
				UPDATE refunds SET refund_attempts = refund_attempts + 1
				WHERE id = $1
			`, pr.ID); upErr != nil {
				log.Printf("Failed to increment attempts for manual refund %s: %v", pr.ID, upErr)
			}
			if attempts := currentRefundAttempts(ctx, pr.ID); attempts >= maxManualRefundAttempts {
				resolveManualRefundAtCap(ctx, pr, amountPence)
			}
		}
	}
	return processed, nil
}

// resolveManualRefundAtCap reconciles a manual refund row that just hit the
// maxManualRefundAttempts cap: money may have moved at Square despite
// decline/ambiguous responses, so reconcile FIRST — an exact COMPLETED refund
// resolves the row to completed; otherwise mark failed and notify the admin.
// The row is never re-issued here (reconcile is a read), so the cap cannot
// cause a double refund.
func resolveManualRefundAtCap(ctx context.Context, pr *manualPendingRow, amountPence int64) {
	sqRefundID, rcErr := reconcileRefundAtSquare(ctx, pr.SquarePaymentID, amountPence, pr.CreatedAt)
	switch {
	case rcErr != nil:
		// Reconcile failed — unknown whether Square refunded. NEVER mark failed
		// on an unknown state (that would let the over-refund guard exclude
		// moved money). Re-arm the row under the cap so the next sweep re-picks
		// it, and track consecutive reconcile failures: after
		// maxConsecutiveReconcileFailures consecutive failures a deduped
		// 'critical_payment_log' admin notification surfaces the hard-failing
		// reconcile — never silently forever (A5b). Re-issue stays safe: the
		// row's idempotency key is persisted (ensureRefundKey), so Square
		// dedups a same-key retry to the original refund — and once the row
		// crosses stalePendingRefundAge the sweep's age guard reconciles it
		// instead of re-issuing.
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET refund_attempts = $1
			WHERE id = $2 AND status = 'pending' AND refund_attempts >= $3
		`, maxManualRefundAttempts-1, pr.ID, maxManualRefundAttempts); upErr != nil {
			log.Printf("Failed to re-arm manual refund %s under the attempt cap after reconcile error: %v", pr.ID, upErr)
		}
		trackReconcileFailureReArm(ctx, []string{pr.ID})
		log.Printf("Reconcile failed for manual refund %s at attempt cap (%v) — re-armed under the cap for the next sweep; never marked failed on an unknown state", pr.ID, rcErr)
	case sqRefundID != nil:
		resetReconcileFailureCount(pr.ID)
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET status = 'completed', square_refund_id = $1
			WHERE id = $2 AND status = 'pending'
		`, *sqRefundID, pr.ID); upErr != nil {
			log.Printf("Failed to mark manual refund %s completed after Square reconcile: %v", pr.ID, upErr)
		}
	default:
		resetReconcileFailureCount(pr.ID)
		if _, upErr := db.Conn.Exec(ctx, `
			UPDATE refunds SET status = 'failed'
			WHERE id = $1 AND status = 'pending'
		`, pr.ID); upErr != nil {
			log.Printf("Failed to mark manual refund %s failed after %d attempts: %v", pr.ID, maxManualRefundAttempts, upErr)
		}
		insertRefundFailedNotifications(ctx, []string{pr.ID})
		// TODO (P6 email/SMS delivery): notify user AND admin when the email/SMS
		// system lands; until then the admin_notifications row above is the only
		// channel.
		log.Printf("Manual refund %s reached %d attempts with no COMPLETED refund found at Square — marked 'failed' and admin notified; TODO email user+admin, VERIFY Square dashboard before arranging in-person cash pickup at the salon (give at least a day's notice for cash on hand)", pr.ID, maxManualRefundAttempts)
	}
}

func currentRefundAttempts(ctx context.Context, refundID string) int {
	var n int
	if err := db.Conn.QueryRow(ctx, `SELECT refund_attempts FROM refunds WHERE id = $1`, refundID).Scan(&n); err != nil {
		log.Printf("Failed to read refund_attempts for %s: %v", refundID, err)
	}
	return n
}