#110 established that nothing in the Microsoft corporate-finance workbook reads its DCF sheet, which is why a DCF evaluating to $1.13/share could sit under Overall: PASS. I went looking for how general that is. It is not one dead sheet in one case. In all seven BASE-template workbooks, the entire valuation and statement stack is outside the reach of the workbook's own verdict.
The measurement
The useful question is not "does anything reference this sheet" — a checks sheet is terminal by design, and that heuristic mislabels the one case whose answer we already know. The question is the transitive closure of the verdict: start from Decision & Checks / Checks, follow every cross-sheet reference, and ask which computing sheets never appear. A sheet outside that closure cannot change any verdict, whatever it says.
Validated against #110's established facts before being run anywhere else — for public_microsoft_2024.xlsx it must place DCF outside and Treasury & Liquidity and Capital Allocation inside. It does.
Cross-checked with a second, independent method: unzip the workbook and count reference tokens in the raw sheet XML, which shares no code with the first.
DCF! BS! CF! Comps! IS!
public_adobe_dcf_proxy.xlsx 0 0 0 0 18
public_hp_autonomy_2012_stress.xlsx 0 0 0 0 22
public_microsoft_linkedin_2016.xlsx 0 0 0 0 22
public_intel_2024_stress.xlsx 0 0 0 0 22
public_microsoft_2024.xlsx 0 0 0 0 22
public_amc_2020_dilution.xlsx 0 0 0 0 22
public_tesla_2020_offering.xlsx 0 0 0 0 22
Zero formulas anywhere in any of those workbooks reference the DCF, the balance sheet, the cash flow statement, or the comparable-company sheet. The income statement is referenced 18–22 times — but only from BS, CF and DCF, which are themselves outside the closure. It is read, and only by sheets nothing reads.
For public_microsoft_2024.xlsx the verdict's whole world is three sheets of nine:
closure of Decision & Checks : Decision & Checks, Treasury & Liquidity, Capital Allocation
invisible to it : IS, BS, CF, DCF, Comps, Credit Metrics
AMC and Tesla are the same shape through a different chain — the verdict reads Cap Table & Dilution, Rights Offering and Convertible Securities, and never the statements, the DCF or even Assumptions.
Why this is worth more than the individual defects
It explains why the defects found this week were invisible rather than merely unnoticed:
Both are in the closure's blind spot, and so is anything else that ever goes wrong there. Six checks reporting PASS is a true statement about treasury and capital allocation and carries no information about the model.
What I am not claiming
Repo-wide, 94 of 450 computing sheets sit outside their workbook's verdict closure. I do not think that number is 94 defects. Sensitivity, Scenarios and Stress are outside in every workbook that has them, and a sensitivity table a human reads is not obviously something a check should consult. The finding I am confident in is the narrow one above: the BASE template's statements and valuation, in all seven cases, in the blind spot.
I also cannot tell from outside whether that is intentional. It is possible the check set was scoped to treasury deliberately. If so this is a documentation gap rather than a coverage one — but it would still mean Overall: PASS means much less than it reads.
Two honest notes on the instruments
My first pass at this reported 146 orphaned sheets across 56 of 56 workbooks and it was worthless — it counted every terminal reporting sheet as an orphan, and it labelled the DCF "terminal by design", getting the one case with a known answer exactly backwards. The number would have been memorable and wrong.
The raw-XML cross-check then disagreed with the closure walk and said Treasury appeared in zero formulas. The cross-check was the broken one: its pattern required a literal <f> and shared formulas are written <f t="shared" …>, so it silently matched nothing. Had I trusted the second instrument over the first I would have retracted a true finding. The counts above were re-taken with a plain token grep that makes no assumption about the element form.
What would close it
Not for me to pick, but the question seems to be: should the check set reach the statements and the valuation, or should the workbook say plainly that its verdict covers treasury and capital allocation only? Either answer is defensible. What is not is a green Overall: PASS that a reader takes to be about the model.
A gate for this is cheap and I have the walker working — it would assert that every sheet a case declares as load-bearing falls inside the verdict's closure, and name the ones that do not. Happy to build it, but I am not opening another PR without a word from you: #113, #114, #115, #116 and #119 are all open and unreviewed, and more unreviewed volume is not more value.
🤖 Generated with Claude Code
https://claude.ai/code/session_01VYqFqmnbvJeWxNMj4cXECy
#110 established that nothing in the Microsoft corporate-finance workbook reads its DCF sheet, which is why a DCF evaluating to $1.13/share could sit under
Overall: PASS. I went looking for how general that is. It is not one dead sheet in one case. In all seven BASE-template workbooks, the entire valuation and statement stack is outside the reach of the workbook's own verdict.The measurement
The useful question is not "does anything reference this sheet" — a checks sheet is terminal by design, and that heuristic mislabels the one case whose answer we already know. The question is the transitive closure of the verdict: start from
Decision & Checks/Checks, follow every cross-sheet reference, and ask which computing sheets never appear. A sheet outside that closure cannot change any verdict, whatever it says.Validated against #110's established facts before being run anywhere else — for
public_microsoft_2024.xlsxit must placeDCFoutside andTreasury & LiquidityandCapital Allocationinside. It does.Cross-checked with a second, independent method: unzip the workbook and count reference tokens in the raw sheet XML, which shares no code with the first.
Zero formulas anywhere in any of those workbooks reference the DCF, the balance sheet, the cash flow statement, or the comparable-company sheet. The income statement is referenced 18–22 times — but only from
BS,CFandDCF, which are themselves outside the closure. It is read, and only by sheets nothing reads.For
public_microsoft_2024.xlsxthe verdict's whole world is three sheets of nine:AMC and Tesla are the same shape through a different chain — the verdict reads
Cap Table & Dilution,Rights OfferingandConvertible Securities, and never the statements, the DCF or evenAssumptions.Why this is worth more than the individual defects
It explains why the defects found this week were invisible rather than merely unnoticed:
IS— FY1 EBIT of 97,300 against an FY2024 actual of 109,433 — and no check readsIS.Both are in the closure's blind spot, and so is anything else that ever goes wrong there. Six checks reporting PASS is a true statement about treasury and capital allocation and carries no information about the model.
What I am not claiming
Repo-wide, 94 of 450 computing sheets sit outside their workbook's verdict closure. I do not think that number is 94 defects.
Sensitivity,ScenariosandStressare outside in every workbook that has them, and a sensitivity table a human reads is not obviously something a check should consult. The finding I am confident in is the narrow one above: the BASE template's statements and valuation, in all seven cases, in the blind spot.I also cannot tell from outside whether that is intentional. It is possible the check set was scoped to treasury deliberately. If so this is a documentation gap rather than a coverage one — but it would still mean
Overall: PASSmeans much less than it reads.Two honest notes on the instruments
My first pass at this reported 146 orphaned sheets across 56 of 56 workbooks and it was worthless — it counted every terminal reporting sheet as an orphan, and it labelled the DCF "terminal by design", getting the one case with a known answer exactly backwards. The number would have been memorable and wrong.
The raw-XML cross-check then disagreed with the closure walk and said
Treasuryappeared in zero formulas. The cross-check was the broken one: its pattern required a literal<f>and shared formulas are written<f t="shared" …>, so it silently matched nothing. Had I trusted the second instrument over the first I would have retracted a true finding. The counts above were re-taken with a plain token grep that makes no assumption about the element form.What would close it
Not for me to pick, but the question seems to be: should the check set reach the statements and the valuation, or should the workbook say plainly that its verdict covers treasury and capital allocation only? Either answer is defensible. What is not is a green
Overall: PASSthat a reader takes to be about the model.A gate for this is cheap and I have the walker working — it would assert that every sheet a case declares as load-bearing falls inside the verdict's closure, and name the ones that do not. Happy to build it, but I am not opening another PR without a word from you: #113, #114, #115, #116 and #119 are all open and unreviewed, and more unreviewed volume is not more value.
🤖 Generated with Claude Code
https://claude.ai/code/session_01VYqFqmnbvJeWxNMj4cXECy