Absorbing Everything Is Not How You Keep a Regulator. Or a Budget.

The Quarterly Briefing Was Routine. Until Item Four.

Month 8 of 18. Phase 2, Build and Integration, is past its midpoint and holding. Eleven of nineteen microservices are feature-complete in at least one market. The velocity wobble from Post 11 has settled since the RACI decomposition. ADR-001 is closed and filed. The contingency reserve stands where it stood in July: $328K unallocated, with the allocated buckets drawn per the register. The next steering committee is Thursday, 2 October, first Thursday of the month as always, chaired by Fatima Idris.

And it is quarterly briefing week. Post 07 established PMO-CC-002: a formal, minuted, quarterly regulatory briefing per market, created after an informal coffee between a vendor lead and a regulator nearly became a governance incident. The mechanism was designed to give regulators a structured channel so that nothing about Falcon reached them through side doors. Mechanisms like this are insurance. Most quarters, the premium feels expensive and nothing happens.

The Market C briefing convenes at 10:00 local time in the central bank's offices. Across the table: Olumide Adebayo, Director of Banking Supervision, Market C Central Bank. He has chaired all three of Falcon's quarterly briefings in his market. He is precise, courteous, and reads every pre-read in full, which Fasil Alemeye Abate learned in Q1 when Adebayo quoted paragraph numbers back at him.

Items one through three are progress, risk posture, and the certification runway for the March 2026 audit. Item four is new. Adebayo slides a document across the table.

The date is the problem inside the problem. 1 February 2026 lands in the middle of Phase 3, UAT and Certification, which runs January through March. Whatever Falcon does about this directive, it does it while preparing for the certification audit the directive's own issuer will conduct. The two threads are now the same thread.

Fasil thanks Adebayo, confirms the program will respond through the formal channel within ten business days, and asks one question: whether the central bank distinguishes between the monitoring rule changes and the 200ms real-time screening requirement, or treats the amendment as indivisible. Adebayo's answer matters more than anything else said in the room.

That distinction is the seam the entire Change Request will be built along. The regulator did not ask for everything by February. The room, as Fasil is about to discover, heard everything.

The Sponsor Wants to Absorb All of It

Fasil briefs Fatima Idris the same afternoon, before the summary email can reach her through anyone else. Her reaction is immediate and strategic, which is what makes it dangerous.

Idris is not wrong about the relationship. She is wrong about what protects it. Her instinct is the instinct of a sponsor who has watched banks lose certification fights over posture rather than substance. But “absorb everything quietly” has a cost structure she has not seen yet, and it converts the program's $14.2M baseline into a number that no longer describes anything. The next variance report would be explaining overruns against a budget that was silently re-scoped without a paper trail. In a regulated program, that is not flexibility. That is the audit finding.

Fasil's counter rests on the charter signed 2 February 2025, Day 19, with its three firm commitments and seven explicit tolerances. Tolerance T4 was written for precisely this morning:

The directive, fully absorbed, prices at $920K to $1.06M for Market C. T4 caps silent absorption at $500K. The math is not a negotiating position; it is a signature Idris herself put on the charter. Fasil's proposal: absorb the mandatory monitoring provisions ($340K, within T4, fully traceable) and take the 200ms screening build to steering on 2 October as a formal Change Request, exactly the credible plan Adebayo said his office expects.

Idris does not concede in the meeting. She concedes two days later, on 27 September, after reading the draft CR with the tolerance trace and the cost table. What moved her was not the argument; it was seeing T4 cited by clause number with the drawdown ledger attached. The charter language she signed in February turned out to be the thing protecting her from her own instinct in September.

200 Milliseconds, Two Paths, One Contract Clause

While the sponsor question settles, the technical question opens. The 200ms real-time screening requirement needs an engineering answer before it can have a cost line, and the answer splits the room within an hour of the working session starting on 26 September.

Amara Okonkwo, Head of Compliance, argues the capability already half-exists. The fraud-monitoring service screens transactions today; extend its rule engine, tighten its latency, done. Dmitri Volkov, the platform integrator's lead, lets her finish and then reads from the contract.

This is the same Volkov who refused to sign for dates in Post 05 and asked for a softer risk rating in Post 10. He is being neither difficult nor helpful here; he is being contractual, and on this occasion the contract discipline is doing the program a favor: it forces the build-versus-buy decision into the open before it becomes a change order dispute in January. Bank legal counsel confirms the clause reading the same day. Lena Marquez's contribution lands where Post 07 actually put her role: the regulator-facing response letter routes through her pre-approved communications path, while the amendment itself would move under the vendor's standing change-order framework, which is why the buy option is faster despite being a commercial negotiation.

The Two Options, Priced

Neither option wins on the matrix, which is the honest result. Cost and ownership favor build; schedule and certification risk favor buy, and certification risk carries unusual weight in a year when the component's first auditor is the regulator who requested it. The matrix goes into CR-001 unresolved by design. The $140K gap and the regulated-control-point question sit above both the PM's $100K single-item authority and the sponsor's appetite to decide alone. This one belongs to steering, and the Change Request's job is to make steering's decision a fifteen-minute decision.

Drafting CR-001 with PEEL

A Change Request is not a form. It is a sequence of arguments, each of which must survive an audit it cannot see coming. Impact, cost, schedule, risk, recommendation: every section makes a claim, and in a regulated program every claim eventually meets someone, an auditor, a steering member, a successor PM, who asks where it came from. The drafting framework has to enforce traceability at the paragraph level, not just at the document level.

The Draft, and the Flag That Mattered

A representative extract. The full draft covered all nine sections; the two below show the PEEL discipline at work, and the flag box at the end is where the AI earned its seat at this particular table.

Both flags survived contact with reality, which is rarer than it sounds. The first flag caught the draft doing the thing drafts do: quietly converting a comparison into a recommendation because one number was smaller. The second flag caught a funding distinction that Thomas Richter confirmed the next morning: R01's allocation is scoped to certification costs only, and routing directive work through it would have created exactly the kind of ledger smudge the March audit exists to find. The AI did not know the answer. It knew there was a question, and it refused to bury it. That is the correct division of labor.

CR-001, as Issued