Knowledge Health

How to check what Coppermind knows about a client, identify gaps in coverage, and fill them with targeted actions.

Why Check Knowledge Health?#

After onboarding a new client or ingesting a batch of documents, it is natural to wonder: "Does Coppermind actually know enough to be useful?" The knowledge health tool answers that question with a structured report across seven dimensions. Instead of guessing, you get a concrete picture of what is rich, what is sparse, and what is missing entirely.

This is especially useful after:

• Initial client onboarding (is the foundation solid?)
• A large ingestion (did the right things land?)
• A few weeks of use (have gaps filled in organically?)

Running a Knowledge Health Check#

Switch to the client you want to assess, then ask:

"How's my knowledge coverage for Acme?"

Or be more direct:

"Show me knowledge health for Acme"

Behind the scenes, this calls memory, which evaluates eight dimensions of client knowledge.

What Gets Evaluated#

Each dimension receives one of four scores:

Reading the Report#

A typical report looks like this:

Knowledge Health: Acme Corp

  Brand Voice:      rich      1,200 words with tone and values configured
  Stakeholders:     adequate  4 found: Sarah, Marcus, Lisa, Tom
  Campaign History: sparse    1 campaign tracked
  Meeting History:  rich      8 meetings, latest March 20
  EOS Config:       adequate  Q2 2026 active
  Content Themes:   none      No content themes configured
  Memory Depth:     rich      156 memories across 5 types

Overall: Good coverage with some gaps to fill.

The overall assessment is based on the balance of scores:

• All adequate or rich -- "Coppermind knows this client well."
• More adequate/rich than sparse/none -- "Good coverage with some gaps to fill."
• More sparse/none than adequate/rich -- "Early stage -- fill the gaps below to unlock better briefings."

Acting on Suggestions#

Every dimension scored "none" or "sparse" includes a concrete suggestion. These are not vague -- they tell you exactly what to share or say.

Brand Voice Gaps#

Suggestion: Set brand voice with set_brand_voice -- include tone, values, and examples.

Fix it:

"Pull the brand voice from acme.com"

Or share brand guidelines directly in the conversation.

Campaign History Gaps#

Suggestion: Share past campaign reports or mention results:
"Q1 email campaign had 34% open rate."

Fix it:

"Remember: Q4 webinar series drove 200 registrations, 45% attendance rate,
12 SQLs. LinkedIn promotion outperformed email by 3x on registrations."

Stakeholder Gaps#

Suggestion: Mention key people: "Sarah Chen is VP Marketing, she owns the rebrand."

Fix it by mentioning people in conversation or ingesting org charts and meeting notes where people are discussed in context.

Content Theme Gaps#

Suggestion: Mention content topics:
"Our content pillars are thought leadership and product launches."

Fix it:

"Set Acme's content themes to developer education, API tutorials, and
customer success stories"

EOS Configuration Gaps#

Suggestion: Configure the quarter with mind to enable pacing and sprint tracking.

Fix it:

"Set up Acme for EOS -- Q2 2026, 6 two-week sprints"

When to Run Knowledge Health#

Memory Type Distribution#

The report also includes a breakdown of memories by type:

Memory types: fact (80), stakeholder (25), decision (20),
commitment (15), preference (10), campaign_outcome (1), question (5)

This helps identify imbalances. A client with 80 facts but only 1 campaign outcome suggests you have been storing general knowledge but not tracking campaign results. A client with many commitments but few decisions might indicate you are capturing action items but not the strategic choices behind them.

Factbase Breakdown#

When fact memories exist, the health report includes an additional factbase analysis that goes deeper than the overall "facts" dimension:

Factbase:
  Total facts: 12 (10 fresh, 2 stale)
  Categories: brand (3), strategy (4), metrics (3), team_org (2)
  Coverage: 4/5 — missing: tactical

This tells you:

• Staleness -- facts that have aged past their category-specific time-to-live. A stale fact about last quarter's revenue target is worth refreshing.
• Category coverage -- the five factbase categories are brand, team_org, strategy, metrics, and tactical. Missing categories are called out explicitly so you know what to fill.
• Coverage score -- how many of the 5 categories have at least one fact.

If a client's factbase shows 2 stale facts and is missing the tactical category, you know exactly what to store next:

"Remember: Acme's current blog publishing cadence is twice per week, Tuesday and Thursday"
"Remember: Acme's Q2 revenue target is $2.4M"

Key Details#

• Knowledge health measures coverage, not quality. A "rich" brand voice with wrong information still scores rich. Use get_brand_voice to review the actual content.
• High memory count does not guarantee good coverage. A client mind with 200 meeting transcript extractions but no brand voice is still missing key dimensions.
• A brand new client mind will show all "none" scores. This is expected -- the report becomes your onboarding checklist.
• The report is read-only. It does not change any data. Run it as often as you want.