Usiku uliporudi, operations room ilibadilika kuwa sehemu ya majaribio. Aisha hakutaka kuzima auto-reserve kwa network nzima. Victor angekuwa sahihi kusema kwamba kufanya hivyo kungeweza kusababisha shortage nyingine na kuvuruga legitimate risk controls.
Kwa hiyo walichagua cohort ndogo: agents kumi na mbili, borrower profiles thelathini, zikiwemo recycled-number cases mbili, na control group ya entries za kawaida.
Wekesa aliandika change ticket kwenye projector.
“Pause auto-reserve rule kwa dakika thelathini,” alisema. “No other changes.”
Njeri wa finance control alikaa nyuma, mikono imekunjwa. “Kama settlement inaharibika, rollback ni immediate.”
“Agreed,” Aisha alisema.
Saa 01:50, counters ziliwekwa zero. Saa 01:58, reserve engine ilianza kawaida kwa control group. Test cohort ikabaki bila journal.
Saa 02:03, repayment dashboard ilikuwa flat.
“Subiri,” Aisha alisema.
Saa 02:07, control group ilipata spike ya repayments. Test cohort haikupata.
Wekesa akaonyesha raw queue. “Borrower files zipo. Hakuna source funds.”
Saa 02:11, operator log ilitoa warning: **performance threshold unmet**.
Dakika mbili baadaye manual journal file ikaonekana kwenye shared folder.
“Hii haikutoka rule engine,” Wekesa alisema.
“Nani ka-upload?”
“Service account.”
Aisha akazuia mtu yeyote ku-run. “Preserve file metadata.”
Njeri akakasirika. “Aisha, service accounts upload overnight corrections kila siku.”
“Ndiyo. Tunataka kuona maker na checker.”
File ilikuwa na rows za reserve allocations zinazofanana na entries walizokuwa wamehold. Mama Atieno borrower ID ilikuwa ndani.
Saa 02:14, Wekesa aligundua kwamba system ilikuwa inasubiri second approval.
“Usually checker ni treasury,” alisema.
“Usual si tonight,” Aisha akajibu.
Waliangalia approval request. Ilikuwa imetumwa kwa account iliyokuwa tagged `EXEC-OPS`.
Njeri akainama. “Hiyo ni COO office.”
“Account si mtu,” Aisha alisema. “Tuone credential history.”
Dakika kumi na nane baada ya test kuanza, data ilikuwa wazi vya kutosha kusema kitu kimoja: bila auto-reserve journals, ghost repayment rate ilishuka. Lakini hawakuweza kusema kila repayment ya kampuni ilikuwa fake. Test ilikuwa ndogo, intentional na bounded.
Aisha akaomba cohort iendelee dakika kumi zaidi ili kuona kama normal customer repayments zitaingia. Ziliingia. Hiyo ilisaidia kutenganisha real payments na reserve-generated entries.
Njeri alianza kuandika summary.
“Auto-reserve causes KPI inflation,” akaandika.
Aisha akamzuia. “Badili. ‘In this test cohort, pausing auto-reserve reduced reported repayments attributable to reserve journals.’”
“Unaua kila sentence nzuri.”
“Naifanya isikufe ikikutana na lawyer.”
Wekesa alicheka kwa chini, lakini macho yake yalibaki kwenye approval queue.
Saa 02:28, request ile ile ilipata override.
“Wait,” alisema.
Checker requirement ilirukwa kwa emergency flag.
“Nani?”
System ilionyesha credential fingerprint, timestamp na office zone. User label ilikuwa `V.ODEDE-EXEC`.
Njeri akasema, “Sasa tuna mtu.”
Aisha bado hakukubali.
“Tuna credential. Tunahitaji kuthibitisha access haikuwa delegated, compromised au service mapping.”
Wekesa akaonyesha device record. Login ilitoka managed workstation ya COO office.
Hapo hata Aisha alikaa kimya sekunde chache.
Victor Odede alikuwa amekuwa akisema auto-reserve ni optimization. Sasa account yake ilikuwa kwenye approval path ya journal iliyojaribu kurejesha repayment performance wakati rule hiyo iliposimamishwa.
“Run?” Wekesa akauliza.
“Hapana. Leave pending. Test yetu ilikuwa kuona behavior bila reserve, si kumtega mtu.”
Walifunga change ticket saa 02:31 na kurudisha rule kwa control conditions. Hakuna outage iliyotokea. Kiosks za test cohort zilibaki operational.
Aisha alifanya export ya logs na kuweka custody kwa finance control pamoja na IT security. Hakuchukua copy nyumbani. Hakutuma screenshot kwa familia.
Njeri akamwuliza, “Unaelewa acquisition inaweza kusimama kwa hii?”
“Acquisition haijasimama kwa log. Inaweza kusimama kwa ukweli ambao log inaonyesha.”
Saa 02:37, Victor aliingia operations room bila jacket.
“Nani aliruhusu test hii?”
“Change ticket iko hapo,” Aisha alisema.
Alisoma summary, uso wake haukubadilika.
“Auto-reserve is approved liquidity management.”
“Basi approval chain itatuonyesha hivyo.”
Victor akatazama Wekesa, kisha Aisha. “Release executive journal.”
“Checker requirement ilirukwa. Tunahitaji review.”
“That's an instruction.”
Aisha akasukuma keyboard mbali na Wekesa. “Na hiyo ndiyo reason ya review.”
Victor alitoka bila kusema zaidi.
Kwenye screen, override history ilibaki.
Mara ya kwanza, swali halikuwa tena kwa nini marehemu alikuwa amelipa.
Baada ya Victor kutoka, Njeri alitaka watoe notification kwa board mara moja. Aisha akaomba dakika kumi za verification ya test artifacts kwanza. Wekesa alilinganisha hash ya journal file kabla na baada ya override; content haikuwa imebadilishwa. IT security ikathibitisha clock synchronization ya workstation, ili mtu asiweze kusema timestamp ya COO credential ilikuwa kutoka server yenye saa mbaya. Control group nayo ilionyesha customer repayments halisi zikiendelea wakati test cohort imepause. Hivyo experiment haikuwa “tulizima mfumo na metric ikashuka”; ilikuwa comparison ya source classes chini ya window iliyorekodiwa. Ndipo Aisha akakubali summary iende kwa independent review mailbox.
Aisha pia aliomba Wekesa aweke screenshot ya permission matrix ya siku hiyo, si ya documentation ya zamani. Matrix ilionyesha executive emergency role inaweza kupita checker katika circumstances maalum, lakini reason code ilipaswa kuwa documented. Override ya disputed batch ilikuwa na reason field tupu. Njeri akaona tofauti: tatizo halikuwa tu kwamba Victor alikuwa na authority; ilikuwa kwamba authority hiyo iliingia kwenye customer-repayment reporting bila reason ambayo reviewer mwingine angeweza kuichunguza. Aisha aliongeza swali hilo kwenye independent review badala ya kuamua jibu ndani ya operations room.
Change ticket ya test ilifungwa na before/after counters, rollback status na list ya people waliokuwa room. Aisha aliomba Wekesa aweke pia observation kwamba normal repayments ziliendelea, ili review isije ikasoma test kama network shutdown. Njeri akasaini kama finance witness hata kama hakukubaliana na interpretation yote. Record yenye disagreement iliyosainiwa na watu tofauti ilikuwa bora kuliko summary ya Aisha peke yake.
Wekesa aliweka test cohort IDs kwenye sealed audit attachment na dashboard summary ikatumia counts tu. Hivyo mtu wa acquisition angeweza kuona effect bila kupata customer identifiers. Aisha alitaka test iwe transparent kwa mechanism na restrained kwa privacy kwa wakati mmoja.
Swali lilikuwa kwa nini account ya COO ilikuwa imejaribu kulazimisha malipo hayo yarudi kwenye dashboard.