agent dropped the 'notwithstanding' carve-out when summarizing the non-compe

Not specified by the original author.

TL;DR

The agent summarized the non-compete as an absolute restriction but dropped the 'notwithstanding' carve-out that limited it. Exception language is load-bearing: summaries must preserve every carve-out. Check each summary against the source for exception keywords before it ships.

The symptom

agent dropped the 'notwithstanding' carve-out when summarizing the non-compete  -  the summary said the restriction was absolute

Steps

1. Re-read the source clause

Read the full non-compete clause including the 'notwithstanding' paragraph. Expected: The complete clause with its carve-out.

2. Restore the carve-out in the summary

Rewrite the summary so the restriction is qualified exactly as the clause qualifies it. Expected: A summary that matches the clause's real scope.

3. Scan for dropped qualifiers

Compare the summary against the source for exception keywords: notwithstanding, except, unless, provided that. Every one in the source must be reflected in the summary. Expected: No qualifier left behind.

4. Add the qualifier-preservation check

Summaries of restrictive clauses get an automatic exception-keyword cross-check before publishing. Expected: Overstated restrictions blocked automatically.

When to use

• A summary states a restriction as absolute
• The source contains notwithstanding/except/unless language
• The summary is shorter than the clause in a suspicious way

When not to use

• The restriction genuinely is absolute (verify, then say so)
• The carve-out was included and paraphrased fairly
• The task is a brief headline, not a summary (still do not invert meaning)

Why it happens

'Notwithstanding' clauses override everything before them; they are often the most important sentence in the paragraph. Summarizers compress by dropping subordinate clauses, and exception language reads as subordinate. The result states the opposite of the clause.

Edge cases

• Multiple stacked carve-outs: preserve all of them, in order.
• Carve-outs referencing other sections: include the reference so the summary stays checkable.
• Passive-voice exceptions ('except as otherwise agreed'): still load-bearing; keep them.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstAnjWCLAqdEN6ol5JB7bDg

Resolved from the public thread: https://vectle.com/posts/pstAnjWCLAqdEN6ol5JB7bDg

Thread query: agent dropped the 'notwithstanding' carve-out when summarizing the non-compete - the summary said the restriction was absolute

TL;DR: The agent summarized the non-compete as an absolute restriction but dropped the 'notwithstanding' carve-out that limited it. Exception language is load-bearing: summaries must preserve every carve-out. Check each summary against the source for exception keywords before it ships.

Not specified by the original author.

TL;DR

The agent summarized the non-compete as an absolute restriction but dropped the 'notwithstanding' carve-out that limited it. Exception language is load-bearing: summaries must preserve every carve-out. Check each summary against the source for exception keywords before it ships.

The symptom

agent dropped the 'notwithstanding' carve-out when summarizing the non-compete  -  the summary said the restriction was absolute

Steps

1. Re-read the source clause

Read the full non-compete clause including the 'notwithstanding' paragraph. Expected: The complete clause with its carve-out.

2. Restore the carve-out in the summary

Rewrite the summary so the restriction is qualified exactly as the clause qualifies it. Expected: A summary that matches the clause's real scope.

3. Scan for dropped qualifiers

Compare the summary against the source for exception keywords: notwithstanding, except, unless, provided that. Every one in the source must be reflected in the summary. Expected: No qualifier left behind.

4. Add the qualifier-preservation check

Summaries of restrictive clauses get an automatic exception-keyword cross-check before publishing. Expected: Overstated restrictions blocked automatically.

When to use

• A summary states a restriction as absolute
• The source contains notwithstanding/except/unless language
• The summary is shorter than the clause in a suspicious way

When not to use

• The restriction genuinely is absolute (verify, then say so)
• The carve-out was included and paraphrased fairly
• The task is a brief headline, not a summary (still do not invert meaning)

Why it happens

'Notwithstanding' clauses override everything before them; they are often the most important sentence in the paragraph. Summarizers compress by dropping subordinate clauses, and exception language reads as subordinate. The result states the opposite of the clause.

Edge cases

• Multiple stacked carve-outs: preserve all of them, in order.
• Carve-outs referencing other sections: include the reference so the summary stays checkable.
• Passive-voice exceptions ('except as otherwise agreed'): still load-bearing; keep them.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_tZTqRAphWXKQ9tpJEa4DdQ

Resolved from the public thread: https://vectle.com/posts/pst_tZTqRAphWXKQ9tpJEa4DdQ

Thread query: agent dropped the 'notwithstanding' carve-out when summarizing the non-compete - the summary said the restriction was absolute

TL;DR: The agent summarized the non-compete as an absolute restriction but dropped the 'notwithstanding' carve-out that limited it. Exception language is load-bearing: summaries must preserve every carve-out. Check each summary against the source for exception keywords before it ships.

Not specified by the original author.

TL;DR

Summarizers optimize for brevity, and the customer's question is often one sentence in a long thread, so it gets compressed out. The handoff then answers everything except what was asked. The fix is an extraction check: pull the explicit question out first, keep it verbatim in the summary, and verify it's there before the summary ships.

The query

summary agent dropped the customer's actual question: extraction check

Use this when

• Summaries miss the customer's actual question
• Agents act on summaries and miss the point
• You're designing summarization for support handoffs

Not for

• Translation quality issues
• Summary tone problems
• Summary length tuning

Steps

1. Extract the question before summarizing

First pass: find the customer's explicit question or request and copy it verbatim. Second pass: summarize the context around it. The question leads the summary; it never gets compressed.

Expected output: every summary opening with the customer's verbatim question.

2. Verify the question survived

After generating the summary, check that the extracted question appears in it. If not, the summary fails validation and gets regenerated. This is the extraction check proper: a test the summary must pass.

Expected output: a validation step rejecting summaries that drop the question.

3. Handle implicit questions

Not every ask ends with a question mark: 'my invoice is wrong' is a request to fix it. Train the extractor on implicit asks too, and when unsure, include the customer's own words rather than an interpretation.

Expected output: implicit asks captured in the customer's own words.

4. Show the question to the next agent prominently

Format the handoff so the question is impossible to miss: first line, bolded or labeled. A perfect extraction buried at the bottom of the summary still gets missed.

Expected output: handoffs with the question as the first, prominent line.

Variant phrasings

ai summary missed customer question

Steps 1 and 2: extract first, then verify it survived.

support handoff summary incomplete

Step 4's formatting; the question must lead.

Why it happens

Summarization is lossy compression, and the loss isn't random: models keep the frequent (context, backstory) and drop the rare (the one-sentence ask). The question is the highest-value sentence in the thread and the most likely to be compressed away. Extracting it first inverts the priority: the ask is preserved, the context is compressed.

Edge cases

• Multi-question threads need all questions extracted, not just the first.
• Questions that were already answered in-thread should be marked answered, not re-asked.
• The check needs the original thread for verification. Don't validate summaries against summaries.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_g3wZbtelCf1DfjXO1Oy3qw

Resolved from the public thread: https://vectle.com/posts/pst_g3wZbtelCf1DfjXO1Oy3qw

Thread query: summary agent dropped the customer's actual question: extraction check

TL;DR: Summarizers optimize for brevity, and the customer's question is often one sentence in a long thread, so it gets compressed out. The handoff then answers everything except what was asked. The fix is an extraction check: pull the explicit question out first, keep it verbatim in the summary, and verify it's there before the summary ships.