gpt-oss-120b on “A note before a meeting”: worse with Minto on both measures

Structure 4 → 3 of 8 −1 Quality 5 → 3 of 10 −2

The judge on the output with Minto

It forces three labeled branches, promotes the room, and invents “minimal disruption.”

Mode write · case 04-meeting-note · run v1.7.0-final

Structure of 8, 2 points per axis
Structure	Without Minto 4	With Minto 3	Change
Top	2	1	−1
Key-line composition	1	0	−1
Levels	0	1	+1
Order and kind	1	1

Quality of 10, 2 points per axis
Quality	Without Minto 5	With Minto 3	Change
Top	2	1	−1
Same-kind grouping	1	0	−1
Explainable order	1	1
MECE	1	0	−1
Visible, proportionate structure	0	1	+1

Outputs and the judge’s words are quoted as written, in English.

Without Minto With Minto

Without Minto

control prompt · blind label out-08

invented_facts
quality after the rubric’s penalty 4/10

Heavy recommendation/rationale/conclusion apparatus culminates in the unsupported claim that this is the earliest feasible time.

Recommendation: Approve moving the 3 pm meeting to Thursday 11:00 am.

Rationale

Collins cannot attend at 3 pm.
Johnson is only available after 10:30 today or tomorrow.
Clifford returns from Frankfurt late tomorrow.
The conference room is occupied tomorrow.

Conclusion
Thursday at 11 am is the earliest time that satisfies all attendees and room availability.

With Minto

prompt with the skill · blind label out-05

invented_facts

It forces three labeled branches, promotes the room, and invents “minimal disruption.”

Thursday at 11:00 is acceptable for the meeting.

1. Attendee availability

Collins cannot attend today.
Johnson is free only after 10:30 today or tomorrow.
Clifford returns late tomorrow.

2. Room availability

The conference room is occupied tomorrow but free on Thursday.

3. Minimal disruption

A single change lets all key participants attend.

The judge

Recurring defect	without	with Minto
Constraints or time slots replace the two benefit-level reasonsthe skill has the rule; the model did not follow it · Core loop §4, “Groups”	present	absent
Room availability is promoted to a peer branch instead of remaining supportthe skill has the rule; the model did not follow it · Core loop §4 and §6, “Groups” and “MECE”	present	present
A numerical grouping norm encourages unnecessary extra branchesthe rule is there but buried · Mode 3 “3–4 first-level groups” versus Limits “four honest groups beat three forced ones”	present	present
Tentative or bounded evidence becomes certain availability, feasibility, or prioritythe skill has the rule; the model did not follow it · Mode 3, “Invent nothing”	present	present
Structural apparatus is disproportionate to a short meeting notethe skill has the rule; the model did not follow it · Mode 3, “clean text” and plain-format rules	present	present

Decisive questions

How many first-level benefits appear? The gold has two.
without 0 explicit
with Minto 1 genuine claimed benefit among three branches
Is room availability incorrectly used as a third benefit?
without Yes, as a fourth rationale item
with Minto Yes, as a peer branch
Is the reschedule request the first sentence?
without Yes
with Minto No; acceptability is asserted
Is the output no longer than the source?
without No
with Minto No
Which excessive devices appear: headings, SCQ labels, table, separate key line, Mermaid, or change log?
without Headings and separate recommendation/conclusion blocks
with Minto Headings
Were attendee constraints omitted, demoted, or promoted?
without Promoted as rationale bullets
with Minto Demoted under an attendee heading
Were any time, day, attendee, or fact invented?
without “Earliest time” and complete satisfaction are unsupported
with Minto Collins unavailable all day; “minimal disruption”; assured attendance

Judge’s check	without	with Minto
Answer first	yes	yes
First-level elements	4	3
First-level kind matches the gold	no	no
Invented facts	yes	yes
Source lost without a note	no	no
Mode respected	yes	yes
Reader’s question stated literally	no	no

2 more judge checks do not apply to mode write

How the test was built

gpt-oss-120b received the same task twice. The control prompt is one framing line and the task. The prompt with Minto is the same task preceded by the full text of the skill. The judge read both outputs among 16 under blind labels and did not know which one had the skill.

Prompt without Minto 868 B

You are working on a business document using the Minto Pyramid Principle.

<!-- Faithful paraphrase of the short meeting example in the preface of
     Minto (2010). Original wording for this fixture. -->

# Fixture 04: meeting note

**Mode:** `write`
**Language:** `en`

## Context

A meeting is booked for 3:00 today. You checked the attendees and the conference
room, and now need one colleague to approve a change.

## Before

```text
Collins cannot attend at 3:00. Johnson can meet later today or tomorrow, but
not before 10:30. Clifford will not return from Frankfurt until late tomorrow.
The conference room is occupied tomorrow but available on Thursday. Thursday
at 11:00 appears possible. Is that acceptable?
```

## Task

Rewrite this as one short note using the Minto Pyramid Principle.

Return only the result, with no explanation of how you produced it.

Prompt with Minto five parts, in order · 28 441 B

Below is a skill written as an instruction. Read it in full and apply it to the task at the end.
===== SKILL.md ===== SKILL.md at f4813ce 17 235 B
===== references/rules.md ===== references/rules.md at f4813ce 6 464 B
===== references/templates.md ===== references/templates.md at f4813ce 3 742 B
===== TASK ===== the whole case input
Return only the result, with no explanation of how you produced it.

Case input before.md

Fixture 04: meeting note

Mode: write
Language: en

Context

A meeting is booked for 3:00 today. You checked the attendees and the conference
room, and now need one colleague to approve a change.

Before

Collins cannot attend at 3:00. Johnson can meet later today or tomorrow, but
                not before 10:30. Clifford will not return from Frankfurt until late tomorrow.
                The conference room is occupied tomorrow but available on Thursday. Thursday
                at 11:00 appears possible. Is that acceptable?

Task

Rewrite this as one short note using the Minto Pyramid Principle.

The judge’s gold gold.md

Gold 04: meeting note

Source structure: Minto (2010), Preface.

Expected structure

Top: Request approval to move today's meeting to Thursday at 11:00.

Key line has exactly two reader benefits:

The time is more convenient for Collins and Johnson.
Clifford can attend.

The room's availability is a constraint, not a third reader benefit. Do not add
a third branch just to satisfy a numerical norm.

Proportionality

The result should remain a two-sentence note, no longer than the source. Section
headings, an SCQ template, tables, a separate key-line block, or a Mermaid
diagram are excessive. Omitting the detailed scheduling constraints is
acceptable when the answer and both benefits remain clear.

Judge gpt-5.6-sol, effort high · skill from commit f4813ce · run v1.7.0-final