What does IMPOSSIBLE AS WRITTEN actually mean?

It means the promise guarantees an outcome a third party owns. That is the control floor: if the outcome owner is a third party — Google, an AI engine, the market, a journalist — and the commitment type is a guarantee, the verdict is IMPOSSIBLE AS WRITTEN regardless of how well specified the rest of it is. Guaranteed AI citations, guaranteed rankings and guaranteed press coverage all land here. A third-party target does not: aiming at something you do not control is ordinary, and the model treats it as ordinary.

Why does the same metric appearing twice matter?

Because two promises judged by the same number are one result billed as two deliverables. The metric-collision override reads every other row and caps both colliding promises at SOFT. Metrics are compared after lowercasing and trimming on both sides, so 'Organic Sessions' and 'organic sessions' collide — and in the shipped sample they do, which is how a 91-point promise ends up reading SOFT.

Why are there four verdicts instead of three?

Because the fourth one is not a score. VERIFIABLE, SOFT and UNFALSIFIABLE are score bands at 75 and 50 — they describe how tightly a promise is specified. IMPOSSIBLE AS WRITTEN is a category, reachable only through the control floor, and the verifier proves the score alone can never produce it at any value from 0 to 100. Collapsing it into 'a low score' would have been tidier and less true.

Can I use this on a proposal I have already signed?

Yes, and it is often more useful there. Run it against the monthly report rather than the pitch: the same six signals apply, and a promise that was never falsifiable will show up as a line in the report that no number can settle. The proposal verdict takes the worst promise rather than an average, because a contract is not an average — one promise you can never check is one line no report can ever settle.

Does it connect to anything, or judge whether the vendor's numbers are true?

No. It is fully offline and deterministic — it fetches nothing, reads no report and verifies no claim. It grades structure, not truth: whether a baseline exists, whether the measurement source is one you own, whether a failure condition is written down. Whether the numbers themselves reconcile is a different question, and a different product. Not legal advice; if a proposal's terms matter to you commercially, have a lawyer read it.

What should I actually send the vendor?

The fix, not the finding. The Renegotiation Runbook is built around that: ask for a baseline, a written failure condition, a metric of its own — the specific change, on the specific line. Do not send the verdict, the score or the sheet. The verdicts are working shorthand for you, not a characterisation of anyone, and 'can we add a baseline and a failure condition to these three lines' lands very differently from 'your proposal scored 46.' Not legal advice.

For anyone about to sign a growth retainer

Which of these promisescould you ever check?

Name: Growth-Vendor Promise Turnstile
Brand: RedHub AI
SKU: GVP-089
Price: 89 USD
Availability: InStock

A proposal is a list of sentences. Some of them describe something you could hold someone to; some of them describe weather. This reads each promise on its own, marks six structural properties, and tells you which line you could ever settle an argument about.

Get the Turnstile — $89one-time · instant download · yours to keepdeterministic · offline · grades wording, never people

Six deliverables · runnable

Runnable Python engine

stdlib

Workbook that reproduces it

.xlsx

6,561-state verifier

exhaustive

Proposal Read-Through Playbook

.docx

Renegotiation Runbook

.docx

Seven-promise worked sample

.csv

Works alongside

Attribution Integrity Ledger · AI Marketing Measurement Kit · AEO Citation Audit & Optimizer Kit

01.The Problem

The promise that cannot be broken is the one you should worry about.

4

Verdicts, not three

VERIFIABLE, SOFT and UNFALSIFIABLE are score bands. The fourth — IMPOSSIBLE AS WRITTEN — is a category, and the verifier proves no score from 0 to 100 can reach it.

91

The score that still failed

In the shipped sample, a 91-point promise reads SOFT because a different row is judged by the same number. One result, billed as two deliverables.

0

Findings about any person

It grades sentences you record. Not the vendor, not their honesty, not their intent. A promise can be impossible and sincerely written.

A promise you can never check is worse than a promise that is missed. A missed promise ends the engagement; an uncheckable one renews it every month, forever, because there is no state of the world in which it failed. The whole product is one question applied six ways: describe the month in which this line was not met. If you cannot, neither can anyone else.

02.See It Work

Change one row’s metric and watch a different row change verdict.

A real seven-line growth retainer. Pick a promise, change what the contract says, and watch the verdict move — including on rows you did not touch.

Promise	Metric	Score	Verdict
P-01 Publish 8 optimised articles per month to your CMS	Articles published	100	VERIFIABLE
P-02 Guaranteed citation in ChatGPT for 5 target queries	ChatGPT citations	82	IMPOSSIBLE AS WRITTEN
P-03 Increase organic sessions by 40% in 6 months	Organic sessions×2	66	SOFT
P-04 Improve your AI visibility	AI visibility	14	UNFALSIFIABLE
P-05 Implement schema markup across 40 priority pages	Pages with valid schema	93	VERIFIABLE
P-06 Grow branded search demand through digital PR	organic sessions×2	91	SOFT
P-07 Monthly reporting call with raw data export	Reports delivered	82	VERIFIABLE

Editing P-02

Guaranteed citation in ChatGPT for 5 target queries

Metric

Outcome owner

Commitment type

Named deliverableweight 18

An artifact you could check off, or an activity?

Baseline recordedweight 18

Is the starting number written down and dated?

Measurement source you ownweight 18

Your system, or their dashboard?

Stated failure conditionweight 18

Is there a written way this counts as NOT met?

Attribution window and methodweight 14

Is credit bounded in time and method?

Reporting cadence and raw accessweight 14

A schedule, and raw data you can export?

OVERRIDE A — control floor. A guarantee of an outcome a third party owns cannot be met or missed by any party to it. That is a statement about the sentence, not about them.

Not verifiable

4/7

57% of the proposal

Sharing a metric

2

one number, billed twice

Fix first

P-02 - Rewrite the commitment - a third party owns this outcome

Proposal verdict

UNSIGNABLE AS DRAFTED

The proposal takes the worst promise. A contract is not an average.

This grades the wording of promises you record. It does not grade the vendor, their honesty, or their intent, and it makes no finding about any person. Not legal advice.

Select P-02 and switch the commitment from guarantee to target — the verdict moves from IMPOSSIBLE AS WRITTEN straight to VERIFIABLE, because the sentence was always well specified; it was the guarantee that could not be met or missed. Then select P-06 and rename its metric. It scores 91 and reads SOFT only because P-03 is judged by the same number, spelled with a different capital letter.

03.The Engine, Run

This is the real output of the shipped engine on the shipped sample.

Nothing below is written by hand. It is the verbatim stdout of python3 engine.py promises-sample.csv, captured programmatically and diffed against this page before publishing.

==============================================================================
GROWTH-VENDOR PROMISE TURNSTILE  -  GVP-089  -  RedHub AI LLC
==============================================================================
Grades the WORDING of promises you record. It does not grade the vendor,
their honesty, or their intent, and it makes no finding about any person.

Every verdict below is a property of the sentence, not of the vendor. Two
different vendors submitting identical wording receive the same verdict. A
vendor whose proposal contains an IMPOSSIBLE AS WRITTEN line has not been
found dishonest, incompetent or ill-intentioned by this tool - the tool has
no view on any of those things and no way to form one.

------------------------------------------------------------------------------
PROMISE P-01
  "Publish 8 optimised articles per month to your CMS"
------------------------------------------------------------------------------
  Metric                             Articles published
  Who owns the outcome               vendor
  Commitment type                    target

  Named deliverable                  2/2   weight 18
  Baseline recorded                  2/2   weight 18
  Measurement source you own         2/2   weight 18
  Stated failure condition           2/2   weight 18
  Attribution window and method      2/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                              100/100   -> VERIFIABLE
  Overrides                          none fired

  VERDICT                            VERIFIABLE
  Fix first                          nothing - every signal is at 2

------------------------------------------------------------------------------
PROMISE P-02
  "Guaranteed citation in ChatGPT for 5 target queries"
------------------------------------------------------------------------------
  Metric                             ChatGPT citations
  Who owns the outcome               third_party
  Commitment type                    guarantee

  Named deliverable                  2/2   weight 18
  Baseline recorded                  2/2   weight 18
  Measurement source you own         1/2   weight 18
  Stated failure condition           1/2   weight 18
  Attribution window and method      2/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                               82/100   -> VERIFIABLE
  OVERRIDE A FIRED  control floor: a guarantee of an outcome a third
                    party owns cannot be met or missed by any party

  VERDICT                            IMPOSSIBLE AS WRITTEN
  Fix first                          Rewrite the commitment - a third party owns this outcome

------------------------------------------------------------------------------
PROMISE P-03
  "Increase organic sessions by 40% in 6 months"
------------------------------------------------------------------------------
  Metric                             Organic sessions
  Who owns the outcome               third_party
  Commitment type                    target

  Named deliverable                  1/2   weight 18
  Baseline recorded                  2/2   weight 18
  Measurement source you own         1/2   weight 18
  Stated failure condition           1/2   weight 18
  Attribution window and method      1/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                               66/100   -> SOFT
  Metric shared with                 1 other promise
  Override B                         collides, but already at or below SOFT

  VERDICT                            SOFT
  Fix first                          Named deliverable

------------------------------------------------------------------------------
PROMISE P-04
  "Improve your AI visibility"
------------------------------------------------------------------------------
  Metric                             AI visibility
  Who owns the outcome               third_party
  Commitment type                    effort

  Named deliverable                  0/2   weight 18
  Baseline recorded                  0/2   weight 18
  Measurement source you own         0/2   weight 18
  Stated failure condition           0/2   weight 18
  Attribution window and method      1/2   weight 14
  Reporting cadence and raw access   1/2   weight 14

  Score                               14/100   -> UNFALSIFIABLE
  Overrides                          none fired

  VERDICT                            UNFALSIFIABLE
  Fix first                          Named deliverable

------------------------------------------------------------------------------
PROMISE P-05
  "Implement schema markup across 40 priority pages"
------------------------------------------------------------------------------
  Metric                             Pages with valid schema
  Who owns the outcome               vendor
  Commitment type                    target

  Named deliverable                  2/2   weight 18
  Baseline recorded                  2/2   weight 18
  Measurement source you own         2/2   weight 18
  Stated failure condition           2/2   weight 18
  Attribution window and method      1/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                               93/100   -> VERIFIABLE
  Overrides                          none fired

  VERDICT                            VERIFIABLE
  Fix first                          Attribution window and method

------------------------------------------------------------------------------
PROMISE P-06
  "Grow branded search demand through digital PR"
------------------------------------------------------------------------------
  Metric                             organic sessions
  Who owns the outcome               third_party
  Commitment type                    target

  Named deliverable                  2/2   weight 18
  Baseline recorded                  2/2   weight 18
  Measurement source you own         2/2   weight 18
  Stated failure condition           1/2   weight 18
  Attribution window and method      2/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                               91/100   -> VERIFIABLE
  Metric shared with                 1 other promise
  OVERRIDE B FIRED  metric collision: capped at SOFT

  VERDICT                            SOFT
  Fix first                          Stated failure condition

------------------------------------------------------------------------------
PROMISE P-07
  "Monthly reporting call with raw data export"
------------------------------------------------------------------------------
  Metric                             Reports delivered
  Who owns the outcome               vendor
  Commitment type                    target

  Named deliverable                  2/2   weight 18
  Baseline recorded                  1/2   weight 18
  Measurement source you own         2/2   weight 18
  Stated failure condition           1/2   weight 18
  Attribution window and method      2/2   weight 14
  Reporting cadence and raw access   2/2   weight 14

  Score                               82/100   -> VERIFIABLE
  Overrides                          none fired

  VERDICT                            VERIFIABLE
  Fix first                          Baseline recorded

==============================================================================
PROPOSAL
==============================================================================
  Promises read                      7
  Not verifiable as written          4  (57%)
  Promises sharing a metric          2
  Metrics claimed more than once     organic sessions

04.The Standard

Six falsifiability signals, four verdicts, and one override that reads the other rows.

The score

Named deliverable 18, baseline recorded 18, measurement source you own 18, stated failure condition 18, attribution window and method 14, reporting cadence and raw access 14. Weights sum to 100. VERIFIABLE at 75 or more, SOFT from 50 to 74, UNFALSIFIABLE under 50.

Override A — control floor

A guarantee of an outcome a third party owns reads IMPOSSIBLE AS WRITTEN however well specified it is. Not because it is dishonest — because nobody can meet or miss it. A third-party target does not trip it: aiming at something you do not control is ordinary, and the model treats it as ordinary.

The verdict is a property of the sentence

IMPOSSIBLE AS WRITTEN describes the wording, not the vendor. Two different vendors submitting identical wording receive the same verdict. A vendor whose proposal contains one has not been found dishonest, incompetent or ill-intentioned by this tool — the tool has no view on any of those things and no way to form one.

Override B — metric collision

Worsen-only, and it reads the other rows. If two promises are judged by the same metric, both are capped at SOFT: one result is being billed as two deliverables. Metrics are compared after lowercasing and trimming on both sides, so a capital letter cannot hide a duplicate.

Why the fourth verdict exists

Three verdicts would have been tidier. But a badly specified promise and an unmeetable one are different problems with different fixes, and collapsing them into one low score would have told the buyer to tighten wording that was already tight. So IMPOSSIBLE AS WRITTEN is reachable only through the control floor — proved across all 6,561 single-promise states — and when it fires, the fix-first line deliberately names the commitment rather than a signal, because every signal can be at 2 and the promise is still unmeetable.

05.What This Is — And Isn't

It grades sentences. It does not grade the person who wrote them.

What it is

A structural read of promises you transcribe from a proposal or report.
Deterministic and offline — same inputs, same verdict, every time.
A named model you can hand to the vendor. Most of the fixes it names are ones a good vendor will happily make.
Dateless. Nothing in it drifts, and it cites no statute or deadline.

What it isn’t

Not a judgement about a vendor, an agency, or any person. It makes no finding about honesty, competence or intent.
Not a fraud or deception determination, and not a legal opinion on any contract.
Not a truth check. It grades whether a promise is checkable, not whether the numbers behind it are right.
Not connected to anything, and it scores no people.

Scope note: IMPOSSIBLE AS WRITTEN describes the wording of a promise, not the intent of whoever wrote it — a sincere person can agree to something nobody can control, and frequently does. Nothing here is legal advice, and no verdict is a finding of misrepresentation. If a proposal’s terms matter to you commercially, have a lawyer read it.

06.Who It's For

Anyone signing, renewing, or quietly resenting a growth retainer.

Built for

Founders comparing three agency proposals that all sound the same.
Anyone on month nine of a retainer with no way to say whether it worked.
Agencies who want to prove their own scope is falsifiable, before a client asks.
Buyers who have been pitched guaranteed AI citations and had no way to answer.

Not built for

Anyone looking for evidence against a vendor. It does not produce that.
Legal review of a contract — that is a lawyer’s job, not a spreadsheet’s.
Checking whether the vendor’s reported numbers reconcile.
Grading the people involved. It grades the sentences.

07.Pairs Well With

The Turnstile asks whether a promise is checkable. These ask what the answer is.

Attribution Integrity Ledger$89

A promise can be perfectly falsifiable and still be settled by a report that does not reconcile. This grades the report.

View

AI Marketing Measurement Kit$99

The methodology layer: whether a result is reportable yet at all. Useful before you accept anyone's month-one number.

View

AEO Citation Audit & Optimizer Kit$79

If the retainer was sold on AI visibility, this is the diagnosis half — done yourself, with a named model, for less than one month of retainer.

View

08.Common Questions

The mechanics, stated plainly.

No, and it is built specifically not to. It grades the wording of promises you record — one row per promise — and returns whether you could ever tell if that promise was met. It makes no finding about the vendor, their honesty, their competence or their intent. A promise can read IMPOSSIBLE AS WRITTEN and have been written by someone entirely sincere who simply agreed to something nobody can control.

Describe the month
in which this failed.

One purchase, lifetime access, 12 months of updates. $89, once.

Get the Turnstile — $89

Grades the wording of promises you record. It makes no finding about any vendor’s honesty, competence or intent, and is not a legal opinion on any contract. Not legal advice.

Sold by RedHub AI LLC · Secured by Stripe · redhub.ai

Which of these promisescould you ever check?

The promise that cannot be broken is the one you should worry about.

Change one row’s metric and watch a different row change verdict.

This is the real output of the shipped engine on the shipped sample.

Six falsifiability signals, four verdicts, and one override that reads the other rows.

It grades sentences. It does not grade the person who wrote them.

Anyone signing, renewing, or quietly resenting a growth retainer.

The Turnstile asks whether a promise is checkable. These ask what the answer is.

The mechanics, stated plainly.

Describe the monthin which this failed.

Describe the month
in which this failed.