A CPI of 0.98 Is Not a Crisis. Hiding It Would Be.

Two Weeks After Steering Said Build, the Quarter Asks for the Bill

Q3 2025 was the most expensive quarter of Falcon's life, and most of the expense was decisions. CR-001 absorbed $340K of regulatory directive under Tolerance T4. Steering approved the $580K txn-screen build on 2 October. The program is nine months into eighteen, which makes this review the halfway accounting: the one where earned value either tells a coherent story or reveals that nobody has been keeping one.

Falcon can keep one, and the reason is an unglamorous decision from June. On 20 June, the program stood up workstream-level cost tagging: $22K to implement, $4K a month to run, every invoice and timesheet coded to a workstream and a deliverable. It was Post 09's financial plan made operational, and it is why this dashboard exists at a resolution that means something. It is also, as of this quarter, a visible PMO line item with a CFO's pen hovering over it. That collision is half of this post.

The other half is what the instrumentation reports. Through 30 September: planned value $7.45M, earned value $7.18M, actual cost $7.31M. Cost variance negative $130K. Schedule variance negative $270K. CPI 0.982, SPI 0.964, and an estimate at completion of $14.46M against a $14.2M baseline: a projected overrun of $260K, printed in a quarter when the program just asked steering for $580K more. Nobody at the table needs the acronyms explained. The question on the table is older than EVM: what do we show the board, and how?

The review has four jobs, and the dashboard is only the first. Second: decide R08's residual. The Market B legacy-adapter rework risk drew $180K from its allocated bucket in July; the rework closed in September at $120K actual, leaving $60K parked against a risk whose driver no longer exists. Third: run the quarterly reserve adequacy mechanism for the first time with a materially changed register, because R12 joined it in September and Phase 3's certification season is now visible. Fourth: re-baseline. Steering's $580K decision changed what the program is funded to deliver; a performance baseline that ignores its own approved changes stops being a baseline and starts being a grievance.

Richter Finds the Line Item That Counts the Other Line Items

Thomas Richter, CFO, arrives at the review having read the PMO cost breakdown line by line, which is what CFOs are for. He stops on one.

Richter's challenge deserves a better answer than “it's best practice,” because “absorb it into overheads” is how measurement quietly dies. A line item folded into overheads survives exactly until the next cost review, when it gets trimmed by someone who cannot see what it does. Fasil Alemeye Abate's defense is not philosophical; it is a receipt. When Directive 2025-04 landed in September, the program priced the $340K absorption in ten days, to a number Richter's own team validated, because every comparable piece of historical work was already tagged to workstream and deliverable. Without the instrumentation, that estimate is a workshop and a shrug. With it, CR-001 walked into steering with a cost figure nobody could argue with, and the sponsor signed in two days. The dashboard Richter is currently using to challenge the program exists because of the line he wants to cut.

The resolution costs nothing and fixes everything: the line item survives, visible, renamed “cost instrumentation,” with a standing footnote listing the decisions it has priced, starting with CR-001's $340K estimate. Richter converts from the line's prosecutor to its sponsor in one meeting, which is what happens when a control can produce its receipts. The lesson goes in the register: controls do not die from attack. They die from being illegible to the person who pays for them.

The Threshold Rule Meets the Critical Path

Ahmed Hassan, PMO Lead, runs the dashboard and runs it by the book. The book is PMO-FIN-002: any quarter closing with SPI below 0.97 triggers a mandatory schedule recovery plan to the next steering. SPI is 0.964. Hassan tables the trigger.

They are both right, which is the interesting kind of disagreement. Raman's decomposition shows what SPI structurally cannot: the negative $270K sits almost entirely in two non-critical workstreams, training content and operational documentation, deliberately deferred in August to protect build capacity during the CR-001 surge. The critical path, the nineteen-then-twenty services and their integration milestones, is on schedule to the week. SPI is a cost-weighted schedule metric; it counts dollars of planned work not yet earned, and it cannot tell a deferred training manual from a slipped integration milestone. This is not a flaw someone discovered in Falcon; it is a documented limitation of the metric, and the program is meeting it on schedule, at month nine, the way every program eventually does.

The resolution honors both positions. The rule fires: a response goes to November steering, as written. But Fasil routes a governance amendment alongside it: an SPI breach now triggers a variance decomposition review first, with a full recovery plan mandated only where the variance is critical-path-driven. Hassan keeps a rule that binds; Raman gets a rule that reads the data before it mobilizes; and the deferred workstreams get dated recovery windows in Phase 3 rather than a war room they did not need.

Drafting the Board Narrative with 4S

The dashboard is for the program. The narrative is for the board risk committee, and those are different documents pretending to be one. The board page gets one read, by people fluent in money and allergic to acronym soup, and it must carry an EAC over baseline, a reserve revision, and a re-baseline without sounding like either panic or perfume. The drafting framework is 4S: Substance, Style, Structure, Specifics. It sits at the lightly documented end of the prompt-framework spectrum, a practitioner pattern rather than a cited method, and it earns its slot here because this task's risk is not logical, it is rhetorical: the same true numbers can read as control or as chaos depending on register and order. Substance pins what must be said. Style pins the reader. Structure pins the order arguments land in. Specifics pins the exact figures and chart definitions, because a board narrative that rounds differently from its own dashboard has volunteered for its hardest question.

The Draft, and the Refusal That Improved It

The first version of this prompt, before the one shown, carried an instruction Fasil inherited from a pre-read discussion: lead with the re-baselined position. The machine declined in the most useful way a machine can.

The flag is the post's hook wearing a tool's clothing. The instruction to lead with v2.0 was not malicious; it was the ordinary gravitational pull every program feels toward the version of the truth that photographs better. The machine, having no stake in the photograph, applied the only test it had, can the requested story be reconciled with the provided numbers, and answered no, twice, with reasons a risk committee would have found on their own in about four minutes. Fasil rebuilt the prompt around the flag, which is why the version shown on the previous page already contains the corrected structure. Keeping the first-pass flag in this post, rather than pretending the prompt was born clean, is the same principle the narrative itself runs on.

QR-2025-Q3, as Presented to the Board Risk Committee