River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Consulting Deliverable Review Checklist

Four documents and four sheets that turn a draft into a claim list, then close each claim by reproducing it from its own source.

Free download  ·  No account needed

Fact-check Register  ·  Cordell Freight Group  ·  Network Cost Diagnostic v3  ·  164 claims  ·  reviewed 11 Nov

Claim C-031, read at all four depths it exists at

LayerWording as writtenVerdict
Working file
linehaul.xlsx!B14
Reno to Sacramento linehaul runs $2.75 a mile against a $2.41 regional benchmark, 14 percent aboveRe-performed
Analysis page
p.19
The Reno to Sacramento lane runs 14 percent above the regional benchmarkRe-performed
Executive summary
p.3
Linehaul cost per mile runs 14 percent above the regional benchmarkScope widened
Spoken headline
readout line 2
You are paying 14 percent over market on linehaulScope widened

The lane is 14.1 percent over. The network, 2,246,000 miles at $5,735,140, is 6.0 percent over. Slide 4 sized the prize off page 3 at $704,315; the lane extract supports $322,280.

$382,035 overstated on a $310,000 engagement, with the analysis correct throughout. Seventeen of the 45 failures were that shape, and a read-through of either document passes every one of them.

Search this and every result hands you a list about a document. Scope completeness, client name and branding, required sections, professional presentation, and one line reading: are all facts, data points and citations correct. The best of them, a senior-consultant QC guide, goes further and asks for a source footnote on every chart and a final re-read of the executive summary. A footnote proves a source was named. A re-read is the same judgement that wrote the summary, performed with less time.

So the unit here is the claim, not the document. Every sentence carrying a number, a comparison or a causal assertion becomes a row. A row only closes when a second person finds the source, works forward, and lands on the same figure without asking the author anything. That is the bar audit documentation is held to: enough for an experienced auditor with no previous connection to the engagement to follow what was done and what was concluded.

Cordell Freight Group, a six-terminal freight carrier, arrived as a 38-page report and a 22-slide deck. 164 claims, 119 re-performed clean, 45 failed, 5 hours 28 minutes for one reviewer. Seventeen of the failures were claims the analysis got right and the deliverable got wrong, and one of those moved the headline prize by $382,035. Pair it with the evidence log that discovery produced, or with the status reporting pack governing the week this lands in.

164 claims, six ways to fail, and the 17 the analysis got right

The Fact-check Register, the Layer Trace, the Finding Log, and the Release Note.

Fact-check Register

One row per claim. The source the deliverable names and the value the reviewer got from it are separate columns, because they fail separately.

IDSource namedLocatedRe-performed valueVerdict
C-031Lane cost extract FY25Yes, 04 Nov$2.5535/mi network, 6.0% overScope widened
C-034Derived from C-031Yes$322,280Scope widened
C-047Trade association rate seriesYes, national onlyNo regional cut publishedSource does not support it
C-088Timekeeping export FY24 and FY25Yes, 29 Oct−7.0%, not −9.0%Derivation error
C-104September accessorial fileReissued 03 Oct71%, not 62%Superseded
C-112Facility cost model v6Yes, 06 Nov$180,000 to $260,000Re-performed
C-119NoneNoInterview impression, stated as measuredNo source named
C-141Client org chart 12 SepYesSix terminalsRe-performed

164 claims, 119 re-performed clean, 45 failed, one reviewer, 5 hours 28 minutes. Two minutes a claim holds across deliverable types. What blows the budget is not the checking, it is hunting for sources the deliverable never named.

Layer Trace

The 23 claims that appear at more than one depth, compared word for word. Nine of them changed.

IDAnalysis pageExecutive summaryWhat changedDeep layer
C-031The Reno to Sacramento lane runs 14 percent above the benchmarkLinehaul cost per mile runs 14 percent above the benchmarkOne lane became the networkCorrect
C-112Consolidation could release between $180,000 and $260,000Consolidation releases $220,000Range became its midpointCorrect
C-055Detention is rising on a third of accountsDetention is rising across the book11 of 34 became the bookCorrect
C-071Q3 fuel spend was $1.91mAnnual fuel spend runs near $1.9mA quarter became a yearCorrect
C-093Two terminals missed the September targetTerminals are missing targetsTwo of six became the networkCorrect
C-088Dock productivity fell 9 percent year over yearProductivity is down 9 percentNothing. Both inherit the same base errorWrong

Five of those six rows are correct one layer down. Each document is internally consistent with itself, which is why reviewing either one alone returns nothing. The differences are a word long: could to releases, lane to linehaul, quarter to annual.

Finding Log

What goes back to the author. Ordered by severity, then by how many places the claim appears.

RefTypeCorrect valueAppears inSeverity
F-01Scope widenedLane 14.1%, network 6.0%3 placesBlocking
F-02Scope widened$322,280, not $704,3152 placesBlocking
F-03Derivation error−7.0 percent, not −9.04 placesBlocking
F-04Superseded71 percent recovery2 placesBlocking
F-05Source does not support itPrior engagement comparator3 placesBlocking
F-06OverstatedRestore $180,000 to $260,0001 placeLogged
F-10No source namedRewrite with the interview count1 placeLogged

The appears-in column is what saves the reissue. F-03 lives in four places, and an author who fixes the analysis page and misses the summary bullet has shipped a deliverable that disagrees with itself. That list comes straight off the Layer Trace.

Release Note

Cordell Freight Group, Network Cost Diagnostic v4. Reviewer not on the engagement team.

Failure typeCountBlocking
No source named140
Source does not support it61
Derivation error51
Overstated at a later layer91
Scope widened at a later layer82
Superseded31
Failed455
Re-performed clean119
Unverified at close0

Verdict at first pass: hold. After rework: release. Nineteen of the 45, the no-source and derivation rows, were findable without opening a source. The other 26 were not.

The note also carries what nobody checked, and why. Not enough time, source not provided, client would not release the record, or judgement outside the reviewer’s competence. All four are legitimate answers. A blank is what a signed checklist produces.

What's in the pack

01

Fact-check Register

One row per claim, with the source the deliverable names and the value you got from it in separate columns.

02

Layer Trace

The claims appearing at more than one depth, wording compared word for word, with each difference typed.

03

Review Checklist

The conventional pass, filtered by deliverable type, kept off the register so a tick can never stand in for a check.

04

Finding Log

What goes back to the author: typed, ordered by severity, and naming every place in the deliverable each claim appears.

05

Review Standard

What counts as a claim, what closes one, the six failure types, and the rule that decides what stops a release.

06

Reviewer Guidance

How to work a register at two minutes a claim, how to scope down against a deadline, and what to do when you are the author.

07

How a Claim Fails

The six types with a worked instance of each. Findings that survive it go into the findings pack.

08

Release Note

The signed output, opening on six reconciling counts and closing on every claim nobody managed to check.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send the draft with its working files, or download the Word documents and CSV sheets and run the review yourself.

  2. 2

    Count the claims before reading

    Extract every claim into a row with its layer, judging none of them. A read-through inherits the author's order and arrives at the summary least informed.

  3. 3

    Work from the source forward

    Open the source, get to a figure, then compare. Reading the claim first turns a check into a search for confirmation, and it finds one.

  4. 4

    Compare the layers word for word

    Put each multi-layer claim's wordings side by side. The differences are one word long, and reading rather than comparing smooths them away.

Frequently asked questions

Is this template free?

Yes. The zip is Word documents and CSV sheets, no account and no card. Edit with AI is the optional half: the agent enumerates the claims, finds each source and re-performs it. The rest of the packs sit in the template library.

What format are the downloaded files?

Four Word documents and four CSV sheets in one zip, nothing to convert. The register carries the columns that do the work: source named, source located, value in the source, re-performed value, verdict. Excel, Numbers and Google Sheets all open them.

What does Edit with AI actually do?

It reads the draft and the working files, writes one row per claim, locates each named source and re-performs the figure from it. Then it puts the multi-layer claims side by side and types every difference. You get the register, the trace and a typed finding log back.

I am the author. Does self-review count?

Not for closing a claim, and pretending otherwise is worse than skipping it. What works is separating the jobs in time: enumerate the claims the day you finish writing, then re-perform them the next morning from the sources with the draft closed. That catches most of what a second reader would.

There is no time to check 164 claims. What then?

Scope down by layer rather than by page. Check every claim in the executive summary and the spoken headline, plus anything the recommendation sizes money off. That is about a fifth of a register and it is where the blocking findings live. Name the rest as unverified.

Why does severity depend on where the claim sits?

Because a summary gets quoted, forwarded and repeated after everybody has forgotten which lane it came from. Courts treat a summary of voluminous records as usable only where the originals are made available for examination. A rounding error on page 31 is logged. A widened headline stops the release.

What if the claim is true but weakly supported?

That is its own row, and it belongs in the release note rather than in the finding log. The test expert opinion is held to is that it rests on sufficient facts and a reliable application of the method. A benchmark that is really one prior client should say so on the page. Sequence the work with the workplan pack.

Check the claims, not the document

Take the Word documents and CSV sheets blank, or send River the draft and its working files and get the register, the layer trace and a typed finding log back.

Edit with AI