Prove · the case
The case
Every other surface on this platform shows what the estate does now. This one asks what it was like before, and refuses to answer where nobody wrote the before down. Each sub-function is matched to the strongest pre-agentic reference the estate holds — a locked baseline where one exists, otherwise a recorded study, benchmark or owner estimate — and the difference against the current agentic unit cost is computed at the stored weekly volume. A line is called provable only when its baseline is locked and signed and finance has verified at least one benefit line drawn against it. Everything else is called claimed. The two are shown apart and never added.
Health
How much of the case survives being asked for a source
A minority of the weekly difference stands on a signed measurement with verified money behind it. The rest is arithmetic on figures nobody has agreed to.
What this page will stand behind
Locked · signed · verified in the ledger
What it will only call claimed
A reference exists; nobody has agreed to the money
The reference register
What is behind the before-cost, tier by tier
Every sub-function is placed in exactly one tier by the strongest reference the estate holds for it. Only the first tier can ever be provable, and only 20 of the 68 sub-functions in it currently are.
Reference tiers
Ranked by what the tier is worth as evidence
Function by function
Where the provable part actually sits
The same split, by function. A large weekly difference next to a small provable share is not a strong function, it is an unexamined one.
| Function | Subs | Counted | Excluded | Provable /wk | Claimed /wk | Provable share |
|---|---|---|---|---|---|---|
| AI GBS & Global Business Services | 11 | 9 | 2 | $1.20m | $1.52m | 44.2% |
| Supply Chain | 11 | 10 | 1 | $257k | $2.31m | 10.0% |
| Finance | 11 | 11 | 0 | $808k | $1.63m | 33.1% |
| Engineering | 11 | 9 | 2 | $463k | $1.90m | 19.6% |
| Information Technology | 11 | 10 | 1 | $682k | $1.59m | 29.9% |
| Administration | 11 | 11 | 0 | $0 | $2.22m | 0.0% |
| Revenue Operations | 11 | 10 | 1 | $1.53m | $655k | 70.0% |
| Human Resources | 11 | 10 | 1 | $0 | $2.18m | 0.0% |
| Marketing | 11 | 9 | 2 | $104k | $1.84m | 5.3% |
| Procurement & Source-to-Pay | 11 | 11 | 0 | $283k | $1.46m | 16.2% |
| Customer Service | 11 | 10 | 1 | $0 | $1.72m | 0.0% |
| Legal | 11 | 8 | 3 | $359k | $1.20m | 23.1% |
| Research & Development | 11 | 4 | 7 | $0 | $1.54m | 0.0% |
| Sales | 11 | 9 | 2 | $167k | $1.27m | 11.6% |
Where the reference figure came from
The weakest joint in the whole case
On all 131 attributed lines the reference figure and the before-cost this platform already displayed are identical to the cent. Two figures obtained independently do not agree that precisely, so the honest reading is that they share a single origin: the reference restates the displayed number rather than testing it.
Nor does a single one of the 131 attributed lines show the agentic unit costing more than its reference. Real estates contain sub-functions that got worse under automation, so the complete absence of one is a property of how these figures were seeded, not a finding about the work.
26 of the 35 attribution records carry a cost with no manual hours behind it. A cost with no effort behind it cannot be challenged on its method, only accepted or rejected whole.
23 sub-functions counted at nothing
The before-cost is displayed; the case refuses it
The ledger side
What finance has actually agreed to
The provable column depends entirely on this. 655 benefit lines exist across 5 periods; 23 of them carry the status verified, and they rest on 20 distinct baselines.
Benefit ledger by status
The benefit ledger is kept in quarters and the reference records carry a plain date. No arithmetic is attempted across the two, and no figure on this page is presented as a quarterly or annual amount.
Actions
What is waiting on a person
None of these is a number that needs recomputing. Each one is a piece of the case that a buyer can push on today, and the work behind it is measurement work, not arithmetic.
lines whose reference equals the before-cost already on display, to the cent
Every attributed line agrees exactly. Two figures measured independently do not agree to the cent this often, so the honest reading is that the reference and the displayed before-cost share one origin. The reference restates the number rather than testing it. Nothing here corroborates anything until a figure arrives from a source this platform did not already hold.
attributed lines whose difference has never been verified as money
Each one has a reference and an agentic cost, so a difference can be computed. What it does not have is a locked, signed baseline with at least one verified line in the benefit ledger drawn against it. Together they carry $23,041,329 a week that this page will only ever call claimed.
locked and signed baselines with no verified line in the benefit ledger
The hardest half of the work is done on these: the measurement is frozen and a named person stands behind it. The money it implies has still never been agreed by finance, so the line stays claimed. This is the shortest path to moving the provable fraction.
attribution records carrying a cost with no hours behind it
A cost per unit with no manual effort behind it cannot be challenged on its method, only accepted or rejected whole. Recording the hours is what makes an estimate arguable.
sub-functions with no reference of any kind
The platform still shows a before-cost for each of them and this page still refuses to count it. They cover 115,550 units a week. Most are blocked by a system that is not connected, which this prototype does not attempt to fix.
baselines that can still move under the case
17 are provisional and 6 are contested. Anything computed on top of them moves when they move, which is why none of them is counted as provable no matter how large the difference looks.
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 page reads the sub-function table, the baseline register, the gap register, the Wave 29 attribution records and the benefit ledger, and writes nothing at all. It can compute a difference wherever a reference exists and it can say whether that difference was ever verified. It cannot say the difference was caused by the agents, because nothing here holds a controlled comparison, and it cannot turn a weekly rate into a period result, because no figure on this platform has passed through a general ledger.
Live observability
What the record shows right now
All 154 sub-functions by the strongest reference behind them. Only the first tier can be provable, and only when the benefit ledger agrees.
Current distribution
154 sub-functions
Is policy and strategy coming to fruition
Whether the written intent is holding here
No. 20.3% of the weekly difference is provable, and the reference layer beneath it mostly restates the figure it was meant to test.
Not holding on the record
The strategy said the estate would be able to show what agentic operation is worth. What it can show today is $5,852,968 a week against $23,041,329 claimed, so 20.3% of the total rate. That fraction is small for two separate reasons and they should not be confused. The first is ordinary and fixable: 48 signed baselines are simply waiting on a verified benefit line. The second is structural. On every attributed line the reference matches the before-cost the platform already displayed, to the cent, which means the reference and the display are the same number under two names. Not one of the 131 attributed lines shows the agentic unit costing more than the reference. A real estate has sub-functions that got worse, so the absence of a single one is a property of how these figures were seeded rather than a result. And not one reference in the estate carries a pre-agentic error rate, so the case is made on cost and cycle alone and says nothing about whether the work is done better.
Every figure on this page is a weekly rate computed from stored per-unit costs and stored weekly volumes. It is not a period result, it has not been closed, and none of it has passed through a general ledger.
The benefit ledger is kept in quarters and the reference records carry a plain date. No arithmetic is attempted across the two, and no figure on this page is presented as a quarterly or annual amount.
There is no controlled comparison anywhere in this estate. Nothing here was run twice, once with agents and once without, so the difference on every line is a difference between two recorded states and not a measurement of what the agents caused. Volume, mix, price and scope all moved in the same window and none of them is held constant.
The word savings does not appear beside any unverified figure on this page, and the provable and claimed columns are never summed into a single headline. $5,852,968 and $23,041,329 are different kinds of number; adding them would produce a figure that is neither.
No reference in the estate records what the work cost in errors before agents took it, so this page compares cost and volume and says nothing whatsoever about whether the work is done better. A cheaper wrong answer is still cheaper.