Case 02 / Proofline

Meeting Decision Memory

Turning meetings, calls, and working sessions into source-backed operational memory.

Why

Meetings create decisions. Most systems only keep recordings.

Important decisions, rationale, risks, open questions, agreements, and ownership often happen inside calls and working sessions. After the meeting, the recording exists, but the knowledge is hard to retrieve, verify, or reuse.

The real problem is not missing summaries. The real problem is missing decision memory: a governed source-of-truth layer that knows what was decided, why it was decided, what evidence supports it, and what remains unresolved.

Before

Recordings became dead storage.

Teams had recordings, but no practical way to recover decisions without replaying long calls.

Summaries lost the reason behind decisions.

Notes captured surface points, not rationale, tradeoffs, dissent, risk, or source evidence.

Context had no reliable owner.

People remembered different versions of what was agreed, why it mattered, and what remained open.

AI had no governed memory to search.

A chatbot over raw transcripts would retrieve fragments, not trusted operational knowledge.

Operating model

SCAN

Detect new recordings, transcripts, or meeting artifacts.

FETCH

Collect source files and metadata.

PRE-CLEAN

Remove noise, normalize text, preserve speaker, time, and source references.

SEGMENT

Split long conversations into useful topic blocks.

CLASSIFY

Identify decision, rationale, risk, action, open question, definition, or context.

EXTRACT

Create structured memory records with source references.

REVIEW

Mark what is confirmed, unresolved, risky, or needs human approval.

STORE

Save reviewed records into a searchable source-of-truth layer.

RAG

Answer future questions using retrieved, source-backed memory.

Knowledge record

A meeting becomes a source-backed decision record.

Proofline does not treat a meeting as one blob of transcript. It turns useful moments into structured records that remain tied to their source, review state, and unresolved edges.

Decision

Rationale

Owner

Date

Source meeting

Timestamp / excerpt

Risks

Open questions

Follow-up actions

Review status

Confidence / unresolved flag

Trust layer

The hard part is not transcription. It is trust.

RAG only works if the knowledge layer is clean. Raw transcripts alone create confident confusion: answers can sound precise while mixing fragments, unresolved discussion, and unreviewed assumptions.

Proofline adds source links, review status, confidence flags, access and privacy boundaries, and unresolved markers before knowledge becomes reusable. The system is designed around evidence handling first, then retrieval.

What changed

The memory layer becomes usable because it is governed.

The value is not another meeting summary. The value is a source-backed operational memory that separates confirmed decisions from loose transcript fragments.

Teams can ask what was decided and why.
Decisions stay connected to source evidence.
Open questions do not disappear after the meeting.
AI answers from reviewed memory, not loose transcript fragments.
Repeated meetings become a growing operational memory layer.

What this points to

AI workflows fail when decisions, evidence, review, and source of truth are unclear.

Proofline shows one version of the broader problem. Before automation is built on top of a workflow, the workflow needs clear context boundaries, evidence routes, review states, and source-of-truth ownership.

Diagnose one workflow

Back to cases

See the source-of-truth case.

LifeOS shows the larger context architecture pattern: source files, routing rules, evidence handling, durable memory, and repair loops.

Read LifeOS