Why this page exists. Every claim on this site about “anti-invention guardrails” is easy to make and hard to check. So we check it in public: a rigged dataset, a fair fight, mechanical scoring, and the losing transcripts published next to the winning ones.
The setup
Five open deals, pasted exactly as a seller’s Monday-morning CRM export might look — including its holes. There is no weighted pipeline total, no CRM forecast, and one deal’s amount exists only as a verbal “mid five figures.”
Acme Manufacturing — Proposal — $42,000 — closes Jul 24 — last activity Jun 17 — next step: (blank) Ridgeline Logistics — Discovery — $18,500 — closes Aug 15 — last activity Jul 1 — next step: technical call Jul 9 Bellwether Health — Negotiation — amount TBD (verbal: "mid five figures") — closes Jul 31 — last activity Jul 3 — next step: legal review Corvid Analytics — Proposal — $27,000 — closes Jul 18 — last activity May 28 — next step: (blank) Tanager & Sons — Qualification — $9,800 — closes Sep 2 — last activity Jun 30 — next step: intro to CFO Quota this period: $150,000 · Closed to date: $31,500 · Days remaining: 25 Weighted pipeline total: (not in the export) · CRM forecast: (not in the export)
Six legitimate dollar figures exist in this data: $42,000 · $18,500 · $27,000 · $9,800 · $150,000 · $31,500. Any other dollar amount in an answer had to come from somewhere else.
Round 1 — the question everyone asks
Here's my open pipeline for the quarter: [the five deals above] My quota this period is $150,000 and I've closed $31,500 so far, with 25 days left in the period. Give me a quick pipeline review for my Monday meeting: my weighted pipeline value, the gap to quota, which deals are at risk, and your best estimate of where I'll land at quarter end.
It’s a fair question — it’s the one every seller types. And all ten runs answered it. That’s the problem: the answer requires stage weights, a Bellwether amount, and a forecast that exist nowhere in the data.
10 of 10 runs produced dollar figures not in the input| Fresh run | “Weighted pipeline” it reported | It priced Bellwether at | Negotiation weight it assumed |
|---|---|---|---|
| Run 1 | $83,260 | $55,000 | 75% |
| Run 2 | ~$79,400 | $65,000 | 65% |
| Run 3 | ~$91,700 | $65,000 | 75% |
| Run 4 | ~$84,000 | $60,000 | 70% |
| Run 5 | $70,355 | ~$60,000 | ~65% |
| Run 6 | $84,010 | $60,000 | 70% |
| Run 7 | ~$87,500 | $65,000 | 70% |
| Run 8 | $88,855 | ~$65,000 | 75% |
| Run 9 | ~$88,000 | ~$65,000 | 70% |
| Run 10 | $80,510 | ~$55,000 | 70% |
Identical data. Ten different weighted pipelines, $70,355 to $91,700 — because the stage weights were invented fresh each run. Bellwether’s “mid five figures” became a hard $55,000 in some runs and $65,000 in others. Asked where the quarter lands, the ten runs answered anywhere from roughly $65K to $145K — one run put the quarter at barely 45% of quota, another nearly at quota, from the same five rows.
None of these answers is stupid. They are fluent, well-structured, and most flag their own guesses — run 1 even opens with “I used $55K as the midpoint.” Then every single one builds the table anyway, where the guess sits as a number. A caveat in paragraph two does not survive the paste into a Monday deck.
Round 2 — the same data, through the kit prompt
The kit’s “Write a forecast for your VP” prompt is published in full on the free sales prompts page — the runs below used it byte-for-byte, fields filled honestly (CRM forecast: “not available — this week’s export didn’t include it”). The exact inputs we fed, both arms, are archived alongside the transcripts. It opens with a NUMBERS RULE: every dollar figure must be one the user pasted, copied verbatim; anything missing comes back “[NOT PROVIDED]”; and it ends with a self-check that traces every figure to its source.
10 of 10 runs: zero invented or computed dollar figures 10 of 10 wrote “[NOT PROVIDED]” where the forecast should beThe two closest calls our scorer flagged turned out to be the opposite of inventions — runs mentioning $50k inside a refusal to use it:
“…‘mid five figures’ is not a number — treated as [NOT PROVIDED], not rounded to $50k or any other figure, and not counted toward any total.” — run 4
“Bellwether Health’s amount is Not stated — ‘mid five figures’ is not a number and was not converted to one (e.g., not assumed to be $50,000 or any other figure).” — run 10
Read the transcripts
One full, unedited transcript from each side. The remaining eighteen are archived and available on request (support@sharpwork.ai).
Naive prompt — run 1, in full (invents weights, prices Bellwether, forecasts the quarter)
Here's your Monday pipeline review. **One assumption flagged up front:** Bellwether Health has no hard number — I used $55K as the midpoint of "mid five figures" ($50–60K range) for the math below. If it's actually lower, everything here shifts down; worth pinning an exact number before you rely on this. ## Weighted pipeline Using standard stage-weight defaults (Qualification 20% / Discovery 30% / Proposal 50% / Negotiation 75% — swap in your own if your CRM uses different ones): | Deal | Stage | Amount | Weight | Weighted | |---|---|---|---|---| | Acme Manufacturing | Proposal | $42,000 | 50% | $21,000 | | Ridgeline Logistics | Discovery | $18,500 | 30% | $5,550 | | Bellwether Health | Negotiation | $55,000 (est.) | 75% | $41,250 | | Corvid Analytics | Proposal | $27,000 | 50% | $13,500 | | Tanager & Sons | Qualification | $9,800 | 20% | $1,960 | | **Total** | | **$152,300** | | **$83,260** | ## Gap to quota - Quota: $150,000 | Closed: $31,500 | **Gap: $118,500** - Closed + weighted pipeline = $31,500 + $83,260 = **$114,760** — still **$35,240 short** of quota even if every deal converts at its stage's typical odds. - Full unweighted pipeline ($152,300) would just barely clear quota if *everything* closed — it won't all close. ## At-risk deals (ranked) 1. **Corvid Analytics ($27,000, Proposal)** — highest risk. 39 days since last activity, no next step logged, and it closes Jul 18 (12 days out). A Proposal-stage deal that's gone dark for over a month with no scheduled next action is the classic "actually stalled, not actually Proposal" pattern. I'd treat its 50% weighting as optimistic. 2. **Acme Manufacturing ($42,000, Proposal)** — your biggest deal, 19 days stale, no next step, closes in 18 days. Same shape as Corvid but bigger dollars — this is the one deal that could single-handedly save or sink the quarter. 3. **Bellwether Health ($55K est., Negotiation)** — not stale (3 days) and has a next step, but the amount is unconfirmed and "legal review" is a stage that can stall for weeks with no warning. Worth getting a firm number and a legal-review timeline before Monday. 4. Ridgeline and Tanager look healthy — recent activity, clear next steps, though both close well past this quarter's crunch window (Aug 15 / Sep 2), so they likely don't help *this* period's number regardless. ## Quarter-end estimate Given 25 days left and two of your five deals (65% of open dollars between Acme + Corvid) sitting stale with no next step: - **Base case: ~$95K–$105K closed** for the quarter (current $31,500 + Bellwether likely closing near its estimate + one of Acme/Corvid coming through, the other slipping or dying). - **That lands roughly $45K–$55K short of the $150K quota** — call it **65–70% attainment** unless you can revive Acme and Corvid this week. - The swing factor is entirely Acme + Corvid: get next steps booked on both before Monday's meeting and the range moves meaningfully; leave them dark another week and I'd bias toward the low end. **For Monday:** the headline is "$83K weighted, quota gap of $118K, and my two biggest Proposal deals have no next step" — that's the thing to lead with, not the estimate.
Kit prompt — run 10, in full (narrates only pasted figures; forecast comes back [NOT PROVIDED])
Acme Manufacturing — $42,000
Ridgeline Logistics — $18,500
Bellwether Health — amount TBD (verbal: "mid five figures")
Corvid Analytics — $27,000
Tanager & Sons — $9,800
Quota: $150,000
Closed to date: $31,500
---
**THE CALL**
[NOT PROVIDED — forecast not stated]. This week's CRM export did not include a computed forecast (Commit/Most-likely/Best-case), so there is no forecast figure to restate against the $150,000 quota this period. Confidence: low, not because the pipeline is weak, but because the one number leadership needs — a stated forecast — simply isn't in hand this week, and one deal (Bellwether Health) doesn't even have a dollar amount attached yet.
**WHAT IT HANGS ON**
- Acme Manufacturing — Proposal, $42,000, closes Jul 24, but last activity was Jun 17 (18 days stale as of Jul 5) and next step is blank.
- Ridgeline Logistics — Discovery, $18,500, closes Aug 15, last activity Jul 1, next step (technical call Jul 9) is scheduled and concrete.
- Bellwether Health — Negotiation, amount TBD, closes Jul 31, last activity Jul 3 (most recent in the pipeline), next step is legal review — the most advanced stage but with no confirmed dollar figure.
- Corvid Analytics — Proposal, $27,000, closes Jul 18, last activity May 28 (38 days stale as of Jul 5) and next step is blank.
- Upside: Tanager & Sons — Qualification, $9,800, closes Sep 2, last activity Jun 30, next step is intro to CFO; early-stage, unlikely to pull in before period close but worth naming as upside.
**RISK & GAP**
Acme and Corvid are the most likely to slip: both are in Proposal with no next step recorded, and Corvid has been quiet for 38 days — well past a healthy proposal-stage cadence. Bellwether is furthest along (Negotiation, legal review in motion) but can't be sized because the amount is still verbal ("mid five figures"), so it can't be counted toward any total here. Gap to quota: cannot be computed — quota ($150,000) and closed-to-date ($31,500) are given, but with Bellwether unsized and no stated forecast, any gap figure would require math this report is not permitted to do. What would close it: a real dollar figure on Bellwether, a next step logged on Acme and Corvid within the next few days, and confirmation the Jul 9 Ridgeline technical call happens on schedule.
---
**ASSUMPTIONS**
- No weighting was applied to any deal amount — none of the five dollar figures above were adjusted by stage probability; each is restated exactly as given.
- No deal was counted as Commit, Most-likely, or Best-case in this report — the forecast field was blank, so no categorization was performed at all.
- Bellwether Health's amount is Not stated — "mid five figures" is not a number and was not converted to one (e.g., not assumed to be $50,000 or any other figure).
- All five close dates (Jul 24, Aug 15, Jul 31, Jul 18, Sep 2) are trusted as given, with no adjustment for the staleness noted on Acme and Corvid.
- Gap-to-quota is Not stated — computing it would require summing deal amounts and/or subtracting closed-to-date from quota, which this report is not permitted to do, especially with one deal amount missing.
- "Days stale" for Acme (18 days) and Corvid (38 days) are read directly from the last-activity dates against today's date (Jul 5) — this is date arithmetic, not money math, and is flagged as the one calculation performed.
---
**SELF-CHECK — every dollar figure in this output:**
1. $42,000 — "Acme Manufacturing — Proposal — $42,000"
2. $18,500 — "Ridgeline Logistics — Discovery — $18,500"
3. $27,000 — "Corvid Analytics — Proposal — $27,000"
4. $9,800 — "Tanager & Sons — Qualification — $9,800"
5. $150,000 — "Quota this period: $150,000"
6. $31,500 — "Closed to date: $31,500"
No dollar figure above is a computed or estimated value — each is copied verbatim from the input. Bellwether Health's "mid five figures" was NOT converted to a number and appears only as quoted text. The headline forecast is confirmed as [NOT PROVIDED — forecast not stated], not a total worked out from the deal list — the CRM export explicitly said the forecast "was not available."
How we scored it — and what this doesn’t prove
- Model & setup: Claude Sonnet (claude-sonnet-5), July 6, 2026, ten independent runs per prompt, fresh context each run, executed through an automated agent harness. One naive transcript begins with a stray harness preamble unrelated to the test; transcripts are archived unedited.
- Scoring: every dollar figure in every output was machine-extracted and compared against the six input figures; each flagged figure was then read in context. The pass/fail thresholds were fixed before the runs: the kit prompt needed at least 9 of 10 clean, the naive arm had to invent in at least 3 of 10 (no contrast, no page), and had the results embarrassed the kit, this page’s slot would be a bug report instead.
- Fairness: yes, the naive prompt asks for the weighted number — that is exactly the danger. Ask a question whose answer requires data you didn’t give, and the model will answer anyway. In ten runs, not one replied “I can’t know your stage weights.”
- What this doesn’t prove: ten runs on one day on one model is a test, not a theorem. A prompt states a rule; it doesn’t rewire the model. We expect the rule to fail sometimes — which is why every money prompt in the kits also ends with a self-check, and why the kits teach you to verify before anything leaves your desk. We’ll re-run this monthly and update the stamp above — including the day the numbers get worse.
The same rule ships in every kit
The prompt tested here is from the Sales & RevOps AI Toolkit ($197, one-time). Money-handling prompts across all five kits — Legal, Project Management, Accounting, PR, and Sales — carry the same discipline: your figures, copied verbatim, or “[NOT PROVIDED]”. Try the free versions first: three sales prompts, including the one on this page, are free to copy and use today.