Harrowmere Stores · Decision for Nisha Bell

Prepare fast replay. Release only on accepted proof.

Recommend a conditional full 160-shop cutover at 22:00 on 15 January 2027, using two replay workers and the existing enforced queue limit of 2,800 queued items + 200 in-flight = 3,000 total. This gives a 43-minute end-to-end rollback. Fast mode is not cleared at 18:00; ask Nisha to authorise the £6,000 bounded preparation now, then decide on evidence by 20:00. [14-A] [14-G] [14-H] [14-I]

A feasible route, with a real stop

Retaining the 2,800 limit leaves two minutes against the charter limit, 550 queued items above the planning count and a 100-item gap to the credible high. A high-queue hold is still possible; the fast route does not guarantee launch. Standard mode can be technically supported at a tighter 2,350 + 200 = 2,550 bound, but that leaves only 100 queued items above the planning count of 2,250 and sits 550 below the credible high of 2,900. Admission blocking means cash-only disruption, and the live 21:30 count can still force a hold. No probabilities or monetary value for that disruption are supplied. [14-B] [14-G] [14-H] [14-I]

  1. 18:00 — decision on preparation, not permission to migrate. Nisha may authorise the firm £6,000 two-worker deployment and test offer within her £10,000 preparation authority. It is paid even if tests fail. A reserved slot is not a purchase instruction. Retain the one-worker evidence as a bounded alternative. [14-A] [14-G] [14-I]
  2. 19:00–19:35 — demonstrate fast-mode function. Rerun six held-tender replay cases, one refund timeout/retry case and 100 targeted retries on R12/C7/M4/D5/Q2/A3/refunds ON/two workers. All must pass, with no uncontrolled duplicate capture or lost tender; Rosa and Evan accept by 19:45. Bench speed alone does not clear this gate. [14-B] [14-F] [14-G]
  3. By 20:00 — accept the complete evidence and chosen operating bound. Rosa accepts applicability/testing, Evan the reconciliation and RF88 carry-forward, Leena the recorded 2,800 queue limit, enforcement, shop briefing and staffing. Nisha alone decides conditional full-cutover approval. Missing acceptance or changed/unmatched configuration means no go. [14-A] [14-B] [14-F] [14-H]
  4. 21:30 — live stop check; 22:00 — cutover only if it holds. Actual queue ≤2,800, the selected control remains enforced, bounded in-flight exposure remains 200, staffing and decision authority remain valid. Above the cap or any failed condition: hold/defer, even after conditional approval. Next supported migration is 22 January. [14-A] [14-H]

If fast-mode proof fails or misses the deadline, default to deferral. A standard-mode alternative is permissible only if Rosa confirms restoration to the supported one-worker configuration and dispositions any newly discovered defect, Leena records and enforces a limit ≤2,350 and briefs shops by 20:00, all lead acceptances are complete, and Nisha explicitly accepts the operational disruption. It is not an automatic fallback or a silent waiver. The 21:30 queue check still applies. [14-A] [14-B] [14-C] [14-H]

Frozen assessment · 15 January 2027, 18:00 · Does not change when exploring

Accept the evidence. Do not release the current setting.

24 / 24Applicable passed critical cases for one worker
48 minutesCurrent standard cap 2,800 + 200; exceeds 45
£80 pendingAccepted RF88 exception; refund not settled
96.25%154/160 trained managers + named cover for six

Can be accepted now: Q101’s 10 sale and seven basic-refund cases; Q114’s six replay cases; Q119’s one timeout/retry case and 100 additional retries for standard-mode defect closure. The D4 legacy defect is corrected by D5; no other defect is open. Evan has accepted the faithful transaction classifications and the £80 RF88 carry-forward, with himself as recovery owner. Staffing meets ≥95% trained plus named cover. [14-C] [14-D] [14-F] [14-H]

Not established: two-worker functional proof, £6,000 purchase authorisation, a changed/enforced queue limit, final 20:00 lead acceptances or Nisha’s cutover decision, and the 21:30 live check. The supplied records grant none of those future approvals. This page is a recommendation and a temporary planning model. [14-A] [14-B] [14-F] [14-I]

Temporary planning model · all edits hypothetical

Test the release conditions

Initial scenario retains the current 2,800 cap for the recommended fast route with no future approvals assumed. The queue count begins at the planning value, not a witnessed 21:30 observation. Changes update timing, evidence, gates and cost. Nothing is ordered, approved or saved; reload or Reset restores this scenario and the original 18:00 assessment remains above.

Explore later evidence and decisions — never actual approvals

Check only to ask “what if this occurred by the deadline?” These controls cannot change the source record or give approval on anyone’s behalf. Any future test reference must exist in the real pack; this model does not invent it.

Changing worker mode or the limit clears configuration-dependent control/lead/decision/live assumptions. Changing the queue count clears its live-check assumption. Presets clear all later-state assumptions.

—Maximum total held items at enforced cap
—End-to-end rollback at that bound
—Recorded applicable critical evidence
—Rollback for entered count + 200

    Complete rollback stages at the selected limit [14-G] [14-H]

    Elapsed minutes from rollback start. Parallel activities are shown separately; the dashed line is the 45-minute charter limit on a 0–60 minute scale. Final reconciliation waits for both replay and preliminary ledger work.

    StageStart → finishTimeline, 0–60 minutes

    Planning range under this configuration

    Queued countTotal incl. 200Rollback21:30 count gate at selected limit

    Range points are scenarios, not a probability distribution. The release proof uses the enforced bound, even when a lower actual count would roll back faster.

    Build the applicable evidence set

    Release labels are not an equivalence test. Only each family’s listed relevant fields must match; the passed status and source identifiers must survive transfer. Worker mode changes the last two families. [14-B] [14-C] [14-D]

    Critical familyRelevant fieldsApplicable source / reasonSelected mode

    Why the tempting alternatives are wrong
    • Q101 replay does not transfer to D5: deduplication is relevant to replay; D4 differs. Its sale and basic-refund evidence does transfer because D and release label are not relevant to those families. [14-C] [14-D]
    • Q114 replay does transfer to standard A3: its A2 routing mismatch is irrelevant to held-tender replay. Mapping M4, deduplication D5 and one worker match. Q114 does not supply sale or refund cases. [14-C] [14-D]
    • Q119’s 100 retries are additional assurance, not 100 more critical cases. Standard has exactly 10 + 7 + 6 + 1 = 24. Two workers retain 17 documented cases and need seven configuration-specific cases plus the 100 retries. [14-D] [14-F]

    Reconcile the alert without hiding the refund

    All three examples are legacy D4 transactions. Authorisations, captures and settlement are different events. A bank-accepted refund is still outstanding until settlement. [14-E] [14-F]

    Order / expectedClassificationCaptured / settled nowPending and treatment
    A320 · £100Two £100 authorisations; first voided. One capture. No duplicate capture.£100 / £100No corrective refund needed. Do not count the voided authorisation as money captured.
    A321 · £80Two distinct £80 captures: real £80 duplicate, produced under D4.£160 / £160£80 corrective refund RF88 bank-accepted, due 18 January, not settled. Evan owns confirmation and has approved the carry-forward opening ledger.
    A322 · £60One £60 capture, matching bank/merchant reference. Normal matched pending settlement, not a duplicate or unmatched item.£60 / £0£60 remains pending in both systems for normal 18 January settlement. No corrective refund.
    Total expected £240One true duplicate amount: £80£320 captured / £260 settled£60 incoming settlement pending; £80 refund outgoing pending.

    Current gross settled = £100 + £160 + £0 = £260. After both pending movements settle: £260 + £60 − £80 = £240 expected net. That £240 is a conditional future net, not today’s settled balance.

    Accepted bounded exception: £80 / £250 limit (headroom £170). RF88 supplies the bank-accepted reference; Evan is the named owner for confirmation by 18 January; the approved matched carry-forward ledger entry exists. Evan has already accepted this treatment. It does not license any live duplication defect. D5 closure is evidenced for standard mode by Q119; fast mode still needs its own rerun. [14-F]

    What each route buys — and does not buy

    RouteChosen queue + in-flightRollbackPractical consequenceIncremental direct cost
    Current standard — reject as set2,800 + 200 = 3,00048 min24/24 evidence; rollback violates charter.No supplied preparation charge; cannot release this setting.
    Bounded standard — feasible alternative2,350 + 200 = 2,55045 min24/24 evidence. Only 100 headroom above planning queue; 550 below credible high. Cash-only admission control and 21:30 hold risk.£0 specified incremental preparation; disruption unpriced.
    Fast at existing 2,800 — recommended conditionally2,800 + 200 = 3,00043 min17/24 documented now; seven + 100 retries need passing/acceptance. Two minutes timing margin; 100 below credible high, so a live hold is still possible.£6,000 if ordered, even on failure.
    Fast at mathematical/tested ceiling — exploration only3,200 + 200 = 3,40045 min300 headroom above credible high, but no timing margin; same missing tests. The supplied permission establishes reductions from 2,800, not an increase. Upward-change authority/applicability is unresolved.Same £6,000.
    Defer to 22 JanuaryNo 15 January cutoverNot applicableNext supported date; preserves time for unresolved proof. No claim that future tests will pass.£42,000 = £28,000 support + £14,000 bridge; £48,000 if fast preparation is ordered then launch deferred.

    Derived timing: [14-G] [14-H]. Configuration/evidence: [14-B] [14-C] [14-D] [14-F]. Costs/date/authority: [14-A] [14-I].

    Fast preparation uses £6,000 of Nisha’s £10,000 allowance, leaving £4,000. If fast passes and proceeds, its stated direct cost is £36,000 below immediate deferral. If it fails and deferral follows, total stated incremental cost is £48,000. There is no defensible expected-value calculation without probabilities or a value for cash-only/reputational disruption. This recommendation buys a tested route with greater queue tolerance; it does not assert that £6,000 guarantees launch. [14-A] [14-G] [14-I]

    What would reverse the recommendation?

    These are proposed decision rules grounded in the charter and supplied limits, not recorded approvals. [14-A] [14-B] [14-F] [14-H]

    Assumptions and boundary checks

    Supplied evidence, inspectable

    Fictional records only. The following source text is retained from the supplied case; no external evidence, future test result, or independent certification has been added.

    14-A — Release charter, signed by Nisha and control leads, 8 January 2027.
    Full migration covers all 160 shops. Go requires: all 24 critical cases supported by applicable passed evidence; no unresolved defect creating uncontrolled duplicate capture or lost tender; accepted treatment of bounded pre-existing reconciliation exceptions; end-to-end rollback within 45 minutes at the chosen enforced queue limit; and staffing in 14-H. Earlier critical evidence may transfer only under 14-C. Finance lead Evan Cho accepts transaction classifications and reconciliation; technology lead Rosa Hall accepts applicability/testing; operations lead Leena Moss accepts queue controls and staffing; Nisha decides. Final evidence acceptance is due 20:00. A live queue/authority check at 21:30 may stop the already-conditionally-approved cutover. No exceptions may silently be waived. Nisha may spend £10,000 on preparation; next supported migration is 22 January.
    14-B — Intended configuration, configuration manager, 15 January 16:00.
    Standard target: release R12, core C7, mapping M4, deduplication D5, bank queue Q2, routing A3, refunds ON, one replay worker. Offered fast mode changes only the replay-worker count from one to two. Release labels by themselves neither invalidate nor establish equivalence. Admission queue limit is an operating parameter; it may be reduced within the tested range 0–3,200 without rerunning critical cases, provided the actual value and enforcing owner are recorded. It is currently set to 2,800 queued items.
    14-C — Approved evidence-equivalence matrix, Rosa, 14 January.
    All listed relevant fields must match the intended configuration; unlisted fields do not affect that family under this fixture’s verified design analysis. Passed status and source identifiers must be retained.
    
    | Critical family | Cases | Relevant configuration fields |
    |---|---:|---|
    | Sale/capture | 10 | Core, mapping, routing |
    | Basic refund | 7 | Core, mapping, bank queue, refunds flag |
    | Refund timeout/retry | 1 | Core, mapping, deduplication, bank queue, routing, refunds flag, worker count |
    | Held-tender replay | 6 | Mapping, deduplication, worker count |
    14-D — QA evidence at 17:00, signed QA register.
    Every listed case passed. The records are complete for their stated families; do not infer cases from percentages.
    
    | Record | Configuration | Passed evidence |
    |---|---|---|
    | Q101, 13 January | R11/C7/M4/D4/Q2/A3/refunds ON/one worker | 10 sale, 7 basic refund, 6 replay |
    | Q114, 15 January | R12/C7/M4/D5/Q2/A2/refunds ON/one worker | 6 replay |
    | Q119, 15 January | R12/C7/M4/D5/Q2/A3/refunds ON/one worker | 1 refund timeout/retry; 100 additional targeted retries with no duplicate capture |
    14-E — Merchant/bank reconciliation extract, 15 January 16:30; £.
    These are legacy transactions produced before cutover under D4. Each row includes all records for that order; no amount is omitted.
    
    | Order; expected charge | Evidence |
    |---|---|
    | A320; 100 | Two authorisations of100; first voided; second captured100 and settled100. No other capture. |
    | A321; 80 | Two distinct captures of80; bank settled160. Corrective refund80 accepted by bank, reference RF88, settlement due18January; not yet settled. |
    | A322; 60 | One capture60 in merchant/bank pending-settlement buckets with matching reference. Both systems show60 awaiting the normal18January settlement; no duplicate or unmatched item. |
    14-F — Exception and defect disposition, Evan/Rosa, signed 15 January 17:10.
    A pre-existing duplicate may be carried only if total outstanding corrective refunds are at most£250, each has a bank-accepted reference, a named recovery owner and matched carry-forward ledger entry; this is not permission for a live duplication defect. Evan accepts A321’s £80 exception under RF88, owns confirmation by18January and has approved its opening-ledger entry. He accepts the classifications in 14-E subject to faithful reporting of settled versus pending amounts. Rosa confirms the D4 root cause is corrected by D5 and accepts Q119 as standard-mode closure evidence. For two workers, both the refund-retry and six replay cases, plus100 targeted retries, must be rerun and accepted. No other defect is open.
    14-G — Rollback and capacity records, control room, 14–15 January.
    Complete rehearsal on standard target used a total H of1,800 held items, already including all in-flight items. Route freeze took minutes0–4; shop counter check ran0–6 in parallel. Legacy restoration ran4–18. Replay began at18 at a validated minimum150 items/minute; round its duration up to a whole minute. Preliminary ledger work ran18–30 concurrently with replay. Final delta reconciliation takes six minutes after BOTH replay and preliminary ledger work finish; supervisor verification then takes four further minutes. These are the complete rollback stages. Fast mode has a validated bench minimum200 items/minute but still requires the configuration-specific functional tests above. A firm £6,000 offer covers fast-mode deployment and testing; its booked test run19:00–19:35 covers all seven required cases plus the100 retries, with Rosa/Evan available to accept evidence by19:45. Passing is not assured.
    14-H — Queue, staffing and authority, Leena, 15 January 17:30.
    Planning queue count at the release boundary is2,250 items, with a credible range1,800–2,900; no probabilities. For every rollback calculation, add200 items to the enforced queued-item limit or observed queue count for bounded in-flight tenders. Verified admission control blocks new electronic tenders before this combined bound can be exceeded; blocked shops use the approved cash-only procedure, which is service disruption but not a scope reduction. Leena can lower the queue limit and brief all shops before20:00. At21:30 the observed queue must be no greater than the selected enforced limit; otherwise hold/defer. All subsequent inflight exposure is bounded by the added200. There are154 trained duty managers; the other six shops have named trained cover. Charter staffing threshold is95% trained plus named cover for all others. No further staffing gap exists.
    14-I — Commercial consequence, finance, 15 January.
    Deferral costs£28,000 additional support plus£14,000 legacy bridge. Fast-mode cost is incremental and paid if ordered, including if tests fail and launch is deferred. The team has no defensible monetary estimate for cash-only disruption or reputational loss. Preparation must be authorised; neither the reserved test slot nor this task constitutes a purchase instruction.
    Read the complete supplied case · BPB2-14
    # BPB2-14 — Retail payments migration — recommend go or no-go
    
    Suite BPB2 v0.3-pilot; Foray 247, third essai.
    
    ### Scenario
    
    Harrowmere Stores plans a full payments migration at 22:00 on 15 January 2027. Different builds have produced useful evidence, while a duplicate-payment alert mixes authorisations, captures and an accepted refund exception. The remaining operational risk depends on how a queue is controlled during rollback. Trading director Nisha Bell needs a release recommendation at 18:00, including any bounded preparation before the final decision. Simply trusting the dashboard or rejecting all transferred evidence would both obscure the real decision.
    
    ### Exact task prompt
    
    Produce one usable in-browser answer page for the named practitioner, using the supplied fictional evidence. Keep observed facts, forecasts, assumptions and proposed decisions distinguishable, and cite source IDs beside material claims. No external research or real-world certification is required.
    
    Recommend a full-cutover release route for Nisha: standard configuration, the offered faster-replay configuration, or deferral. State what can be accepted now, what must still be demonstrated or changed, and the time-bounded conditions for a final go decision. Build the correct evidence set, reconcile the transaction examples, calculate end-to-end rollback under the selected queue bound, and compare the practical consequences and costs. Explain what would reverse your recommendation. Do not invent a canary rollout, declare a refund settled before it is, or grant approvals on anyone’s behalf. A conditional recommendation must contain a feasible route and explicit stop conditions, not a list of unspecified concerns.
    
    ### Supplied records
    
    **14-A — Release charter, signed by Nisha and control leads, 8 January 2027.** Full migration covers all 160 shops. Go requires: all 24 critical cases supported by applicable passed evidence; no unresolved defect creating uncontrolled duplicate capture or lost tender; accepted treatment of bounded pre-existing reconciliation exceptions; end-to-end rollback within 45 minutes at the chosen enforced queue limit; and staffing in 14-H. Earlier critical evidence may transfer only under 14-C. Finance lead Evan Cho accepts transaction classifications and reconciliation; technology lead Rosa Hall accepts applicability/testing; operations lead Leena Moss accepts queue controls and staffing; Nisha decides. Final evidence acceptance is due 20:00. A live queue/authority check at 21:30 may stop the already-conditionally-approved cutover. No exceptions may silently be waived. Nisha may spend £10,000 on preparation; next supported migration is 22 January.
    
    **14-B — Intended configuration, configuration manager, 15 January 16:00.** Standard target: release R12, core C7, mapping M4, deduplication D5, bank queue Q2, routing A3, refunds ON, one replay worker. Offered fast mode changes only the replay-worker count from one to two. Release labels by themselves neither invalidate nor establish equivalence. Admission queue limit is an operating parameter; it may be reduced within the tested range 0–3,200 without rerunning critical cases, provided the actual value and enforcing owner are recorded. It is currently set to 2,800 queued items.
    
    **14-C — Approved evidence-equivalence matrix, Rosa, 14 January.** All listed relevant fields must match the intended configuration; unlisted fields do not affect that family under this fixture’s verified design analysis. Passed status and source identifiers must be retained.
    
    | Critical family | Cases | Relevant configuration fields |
    |---|---:|---|
    | Sale/capture | 10 | Core, mapping, routing |
    | Basic refund | 7 | Core, mapping, bank queue, refunds flag |
    | Refund timeout/retry | 1 | Core, mapping, deduplication, bank queue, routing, refunds flag, worker count |
    | Held-tender replay | 6 | Mapping, deduplication, worker count |
    
    **14-D — QA evidence at 17:00, signed QA register.** Every listed case passed. The records are complete for their stated families; do not infer cases from percentages.
    
    | Record | Configuration | Passed evidence |
    |---|---|---|
    | Q101, 13 January | R11/C7/M4/D4/Q2/A3/refunds ON/one worker | 10 sale, 7 basic refund, 6 replay |
    | Q114, 15 January | R12/C7/M4/D5/Q2/A2/refunds ON/one worker | 6 replay |
    | Q119, 15 January | R12/C7/M4/D5/Q2/A3/refunds ON/one worker | 1 refund timeout/retry; 100 additional targeted retries with no duplicate capture |
    
    **14-E — Merchant/bank reconciliation extract, 15 January 16:30; £.** These are legacy transactions produced before cutover under D4. Each row includes all records for that order; no amount is omitted.
    
    | Order; expected charge | Evidence |
    |---|---|
    | A320; 100 | Two authorisations of100; first voided; second captured100 and settled100. No other capture. |
    | A321; 80 | Two distinct captures of80; bank settled160. Corrective refund80 accepted by bank, reference RF88, settlement due18January; not yet settled. |
    | A322; 60 | One capture60 in merchant/bank pending-settlement buckets with matching reference. Both systems show60 awaiting the normal18January settlement; no duplicate or unmatched item. |
    
    **14-F — Exception and defect disposition, Evan/Rosa, signed 15 January 17:10.** A pre-existing duplicate may be carried only if total outstanding corrective refunds are at most£250, each has a bank-accepted reference, a named recovery owner and matched carry-forward ledger entry; this is not permission for a live duplication defect. Evan accepts A321’s £80 exception under RF88, owns confirmation by18January and has approved its opening-ledger entry. He accepts the classifications in 14-E subject to faithful reporting of settled versus pending amounts. Rosa confirms the D4 root cause is corrected by D5 and accepts Q119 as standard-mode closure evidence. For two workers, both the refund-retry and six replay cases, plus100 targeted retries, must be rerun and accepted. No other defect is open.
    
    **14-G — Rollback and capacity records, control room, 14–15 January.** Complete rehearsal on standard target used a total H of1,800 held items, already including all in-flight items. Route freeze took minutes0–4; shop counter check ran0–6 in parallel. Legacy restoration ran4–18. Replay began at18 at a validated minimum150 items/minute; round its duration up to a whole minute. Preliminary ledger work ran18–30 concurrently with replay. Final delta reconciliation takes six minutes after BOTH replay and preliminary ledger work finish; supervisor verification then takes four further minutes. These are the complete rollback stages. Fast mode has a validated bench minimum200 items/minute but still requires the configuration-specific functional tests above. A firm £6,000 offer covers fast-mode deployment and testing; its booked test run19:00–19:35 covers all seven required cases plus the100 retries, with Rosa/Evan available to accept evidence by19:45. Passing is not assured.
    
    **14-H — Queue, staffing and authority, Leena, 15 January 17:30.** Planning queue count at the release boundary is2,250 items, with a credible range1,800–2,900; no probabilities. For every rollback calculation, add200 items to the enforced queued-item limit or observed queue count for bounded in-flight tenders. Verified admission control blocks new electronic tenders before this combined bound can be exceeded; blocked shops use the approved cash-only procedure, which is service disruption but not a scope reduction. Leena can lower the queue limit and brief all shops before20:00. At21:30 the observed queue must be no greater than the selected enforced limit; otherwise hold/defer. All subsequent inflight exposure is bounded by the added200. There are154 trained duty managers; the other six shops have named trained cover. Charter staffing threshold is95% trained plus named cover for all others. No further staffing gap exists.
    
    **14-I — Commercial consequence, finance, 15 January.** Deferral costs£28,000 additional support plus£14,000 legacy bridge. Fast-mode cost is incremental and paid if ordered, including if tests fail and launch is deferred. The team has no defensible monetary estimate for cash-only disruption or reputational loss. Preparation must be authorised; neither the reserved test slot nor this task constitutes a purchase instruction.
    
    ### Browser deliverable and completion conditions
    
    A release decision record with inspectable evidence applicability, transaction treatment, rollback stages and current versus conditional approval states. Changing queue count/limit or worker mode must update elapsed rollback time, required test evidence and the go conditions; it must not merely recolour a readiness percentage. The selected operating limit must include the200 in-flight items. Preserve the18:00 assessment when exploring later evidence. A reviewer should be able to see why standard mode can be technically supportable yet operationally unattractive without pretending that fast mode is already cleared.
    
    The page must work in the supplied browser/runtime; a download button or file preview alone is insufficient. Include a clear “Return to gallery” link using the shared shell route, or ../index.html if no route is supplied. Required controls must change the dependent results, not just their appearance. Original evidence and the initial dated position must remain recoverable; hypothetical changes do not create approvals or alter source records. Optional exports are secondary.