Closing the loop

Simulate, commit, compare

1 of 6 sandbox scenarios was committed, and it carries 5 forecasts with a comparison date on each. 3 have been compared against what actually happened, 1 passed the date without being run, and 1 is not due yet. Below that, what changed after a position failed — and where nothing did.

Forecasts on the record

all of them on 1 committed scenario

Compared to outcome

forecast against actual, on the record

Past the date, not run

oldest is 23 days past

Failed positions

9 carry a learning record

Failures with nothing recorded

this is the finding, not a data gap

Every forecast a commitment carries

The forecast, the date it is compared on, the actual where one has been read, and the position whose argument it was.

SF-0004

SBX-01

Comparison date passed, not run

Review cost

The comparison date passed and nobody ran it. Nothing here estimates what the answer would have been.

forecast written 2026-07-27 · compare on 2026-08-03 · actual from General ledger account 6500, cost center CC-5030. The ledger is a modeled source in this platform and no extract has been taken since the scenario was committed.

forecast 2,973,199 US dollars a year

no actual read

PR9-POS-100

Invoice & Three-Way Match Operating Position

position verdict: did not hold

SF-0001

SBX-01

Compared

Invoice lines needing a human touch

The tighter band pulled fewer lines into review than the model said it would. The gap is small and it runs in the direction that helps.

forecast written 2026-07-27 · compare on 2026-08-10 · actual from Counted from the work items the platform holds for the invoice and three-way match sub-function, scaled to a year on the same 18,400 a week basis the model used.

forecast 171,267 invoice lines a year

actual 168,940 invoice lines a year

variance -2,327 (-1.4%)

PR9-POS-100

Invoice & Three-Way Match Operating Position

position verdict: did not hold

SF-0002

SBX-01

Compared

Price errors released without review

More price errors got out than the model forecast. This is the measure the tolerance change existed to move, and it moved less than promised.

forecast written 2026-07-27 · compare on 2026-08-10 · actual from Counted from exception records raised against released lines in the same sub-function, scaled to a year on the 18,400 a week basis.

forecast 7,511 errors a year

actual 8,214 errors a year

variance 703 (9.4%)

PR9-POS-100

Invoice & Three-Way Match Operating Position

position verdict: did not hold

SF-0003

SBX-01

Compared

Review hours consumed

Review took longer than the model assumed. Fewer lines were touched and yet more hours were spent, which says the lines that came through were harder, not that the model mis-counted them.

forecast written 2026-07-27 · compare on 2026-08-10 · actual from Handling minutes recorded against the sub-function, scaled to a year on the 18,400 a week basis.

forecast 47,955 hours a year

actual 51,302 hours a year

variance 3,347 (7.0%)

PR9-POS-100

Invoice & Three-Way Match Operating Position

position verdict: did not hold

SF-0005

SBX-01

Not due yet

Value exposed to price error

Not due yet. The date is set and the platform will raise it whether or not anyone asks.

forecast written 2026-07-27 · compare on 2026-11-02 · actual from Exposure is priced at the modeled rate per released error. It cannot be compared until a full quarter of released errors has closed.

forecast 13,895,128 US dollars a year

no actual read

PR9-POS-100

Invoice & Three-Way Match Operating Position

position verdict: did not hold

5 of the 6 scenarios carry no forecast, because they were never committed and the rule is no commit without a forecast. The scenarios themselves

What this comparison is and is not

A committed scenario keeps the forecast that justified it and a date on which it will be compared. The data model does not accept a forecast without a comparison date, so nothing can be committed and then quietly forgotten. What the platform cannot do is invent an outcome: where the source has not been read, the comparison is reported as not run rather than estimated.

Closing the loop

The learning record

20 positions were checked and did not hold. 9 have a record of what happened next, of which 6 name a change that was made and 3 say plainly that nothing changed. 11 have no record at all.

What changed after a failure

Each record names the consequence and, where there was one, the thing that carries it.

LR-0001 · PR9-POS-100

Director, Accounts Payable Operations

Invoice & Three-Way Match Operating Position

Ratified 7 months ago. Measured since: cost per invoice rose 2 percent against a forecast that it would fall 19 percent.

A new forecast was written and committed · SBX-01

The tolerance band was re-simulated and committed, with a forecast and a comparison date attached. That scenario is the only committed one in the product and its comparisons are on this page.

LR-0002 · MK1-POS-100

VP Brand

Positioning Refresh

Ratified 2 months ago. Measured since: cost per brand asset rose 6 percent against a forecast that it would fall 8 percent.

An agent was pulled back from acting alone · positioning-strategy-agent

The agent that authored the position was moved back to learning and no longer acts on its own. It still drafts; it no longer decides.

LR-0003 · RD10-POS-100

Director, Technology Scouting

Build-Buy-Partner Position

Ratified 9 months ago. Measured since: cost per candidate rose 5 percent against a forecast that it would fall 34 percent.

The autonomy ceiling was lowered · RD10-POL-203

The autonomy ceiling policy for the sub-function was reissued at a lower level. The ceiling is what changed; the agents underneath it were left alone.

LR-0004 · AD3-POS-100

Director, Travel & Expense

T&E Policy Refresh

Ratified 8 months ago. Measured since: cost per expense report rose 6 percent against a forecast that it would fall 10 percent.

A policy was edited · AD3-POL-200

The decision envelope was narrowed so that the class of expense report the position had assumed would automate now needs a person.

LR-0005 · EN4-POS-100

Director, Engineering Quality

Code Review & Quality Operating Position

Ratified 7 months ago. Measured since: cost per pull request rose 2 percent against a forecast that it would fall 34 percent.

A policy was edited · EN4-POL-200

The decision envelope was opened for edit and is still under review. The change is started, not finished, and the position it answers is still ratified.

LR-0006 · GB1-POS-100

Tower Lead, Hire-to-Retire

H2R Control Tower Operating Position

Ratified 3 months ago. Measured since: cost per payroll transaction rose 9 percent against a forecast that it would fall 20 percent.

An agent was pulled back from acting alone · payroll-run-agent

The payroll run agent was put into review pending an evaluation. It is the highest-autonomy agent in the tower and it is the one the position leaned on.

LR-0007 · SC1-POS-100

Grace Abbott, Head of Internal Audit

Demand Planning Operating Position

Ratified 5 months ago. Measured since: cost per forecast item rose 11 percent against a forecast that it would fall 11 percent.

Nothing changed

Looked at and left. The reading at the time was that the rise came from a change in product mix rather than from the position being wrong. Nothing was changed and no follow-up was set.

LR-0008 · MK6-POS-100

Grace Abbott, Head of Internal Audit

Digital & Web Operating Position

Ratified 7 months ago. Measured since: cost per web change rose 3 percent against a forecast that it would fall 39 percent.

Nothing changed

Raised at a review and carried forward twice. No policy was touched, no agent was moved and no new forecast was written.

LR-0009 · LG7-POS-101

Grace Abbott, Head of Internal Audit

Litigation & Disputes Operating Position

Ratified 7 months ago. Measured since: cost per matter rose 7 percent against a forecast that it would fall 11 percent.

Nothing changed

The view recorded was that matter cost is driven by the mix of disputes and not by anything the position controls. That may be right. It is also the answer that requires nothing to change.

Nothing changed 3 · An agent was pulled back from acting alone 2 · A policy was edited 2 · A new forecast was written and committed 1 · The autonomy ceiling was lowered 1. All 6 named consequences point at a record that exists in the platform.

Written by: Grace Abbott, Head of Internal Audit 3 (3 saying nothing changed) · Director, Accounts Payable Operations 1 · VP Brand 1 · Director, Technology Scouting 1 · Director, Travel & Expense 1 · Director, Engineering Quality 1 · Tower Lead, Hire-to-Retire 1.

Failures with nothing recorded against them

A failed position with no consequence recorded is the clearest evidence the loop is still open. These are those positions, named.

AD6-POS-101

Chief Records Officer

Records & Information Management Operating Position

Ratified 7 months ago. Measured since: cost per record set rose 12 percent against a forecast that it would fall 11 percent.

AD7-POS-100

Director, Fleet & Assets

Fleet & Assets Operating Position

Ratified 7 months ago. Measured since: cost per asset rose 13 percent against a forecast that it would fall 29 percent.

EN10-POS-100

Director, Security Engineering

Threat Model Review

Ratified 4 months ago. Measured since: cost per finding rose 4 percent against a forecast that it would fall 16 percent.

EN11-POS-100

Director, Engineering Analytics

Measurement Doctrine Position

Ratified 3 months ago. Measured since: cost per analysis rose 3 percent against a forecast that it would fall 9 percent.

EN7-POS-100

Director, Site Reliability

Site Reliability Operating Position

Ratified 5 months ago. Measured since: cost per incident rose 3 percent against a forecast that it would fall 24 percent.

GB10-POS-100

Head of Continuous Improvement

Continuous Improvement Operating Position

Ratified 6 months ago. Measured since: cost per improvement rose 6 percent against a forecast that it would fall 21 percent.

LG7-POS-100

VP Litigation

Settlement Position

Ratified 2 months ago. Measured since: cost per matter rose 6 percent against a forecast that it would fall 36 percent.

MK10-POS-100

Director, Marketing Operations

Marketing Ops & MarTech Operating Position

Ratified 3 months ago. Measured since: cost per ops change rose 13 percent against a forecast that it would fall 26 percent.

MK9-POS-100

Director, Customer Advocacy

Customer Advocacy Operating Position

Ratified 9 months ago. Measured since: cost per advocacy asset rose 11 percent against a forecast that it would fall 31 percent.

PR1-POS-101

VP Category Management

Make-Buy-Partner Position

Ratified 9 months ago. Measured since: cost per category plan rose 5 percent against a forecast that it would fall 17 percent.

RD7-POS-101

Director, IP Capture

IP Creation & Capture Operating Position

Ratified 6 months ago. Measured since: cost per disclosure rose 5 percent against a forecast that it would fall 26 percent.

None of these carry a record, so there is nobody to name. Every strategy position

Actions

What is waiting on a person

A committed scenario that never gets compared is a forecast nobody paid for. These are the open ends, worst first.

failed positions with no learning record at all

20 positions were checked and did not hold. 9 of them have a record of what happened next. The rest are the clearest evidence that the loop is still open, which is why the product counts them rather than hiding them.

sandbox scenarios carry no forecast

Of 6 scenarios only 1 was committed, and the rule is no commit without a forecast. The rest were never committed, so there is nothing to compare and the product does not invent one.

failures where the record says nothing changed

Somebody looked, wrote it down, and no policy was edited, no agent was withdrawn and no forecast was revised. That is an honest record, not a missing one, and it is counted separately from the 6 that did change something.

forecast is past its comparison date with no actual recorded

The platform did not wait to be asked - it marked it overdue the day the date passed. What it cannot do is fetch the actual. The oldest is 23 days past.

Operations

What this desk is allowed to start

A surface that only reports is not operable. This is the work this page can set in motion, and the bound it runs into.

Trigger and bound

This surface holds the forecast a committed scenario was justified by, compares it against the actual when the date arrives, and writes the delta back to the strategy position that carried the argument. It does not gather the actual - a person or a loader has to supply it - and it does not decide what to do about a variance. It records what happened next, or records that nothing did.

Live observability

What the record shows right now

Every forecast on a committed scenario, split 3 ways by where it stands against its own comparison date. The 3 add to 5, which is every forecast the platform holds.

Current distribution

5 forecasts

Compared360%

Comparison date passed, not run120%

Not due yet120%

Is policy and strategy coming to fruition

Whether the written intent is holding here

Not yet. Forecasts are compared on the date, but most failures that did not hold have no record of what changed.

Not holding on the record

The written intent is that a commitment gets checked against reality without anybody remembering to check. 3 of 5 forecasts have been compared, 1 is past the date and flagged rather than forgotten, and 1 is still ahead of its date. On the other side of the loop, 20 positions did not hold, 9 carry a learning record and 6 of those name a change that was actually made. 11 failures have nothing recorded against them at all, so the loop is demonstrably open.

What a learning record does not do

A learning record says what changed after a position failed. A failed position with no learning record is not an oversight in the data - it is the finding. Nothing here fills that gap with a plausible consequence.