DLN Learn — Domain-Agnostic Learning Orchestrator

2. Notion Database

All learner state is persisted in the DLN Profiles database in Notion (under Maekar). Database IDs are owned by the dln-sync agent — the orchestrator and phase skills reference the database by name only.

Schema

Column Properties (queryable metadata)

Page Body (learning content)

All learning content lives in the domain page body, not column properties. The page body has two sections:

Knowledge State — Persistent header updated at every teaching boundary. Contains:

• ## Concepts — Concept nodes learned (Dot phase output)
• ## Chains — Procedural sequences built (Dot phase output)
• ## Factors — Shared structures discovered (Linear phase output)
• ## Compressed Model — Latest model statement (Network phase output)
• ## Interleave Pool — Concepts and factors eligible for interleaving (introduced in a prior session and passed initial comprehension check). Maintained by phase skills to enable cross-topic practice.
• ## Calibration Log — Per-concept confidence ratings, gate predictions vs actual outcomes, and calibration trend over time. Used by phase skills to detect overconfidence/underconfidence and adjust teaching intensity.
• ## Load Profile — Baseline cognitive load observations (working batch size, hint tolerance, recovery pattern) and per-session load history. Used by Dot phase for dynamic batch sizing and by all phases for load-aware pacing.
• ## Open Questions — Unresolved gaps
• ## Weakness Queue — Priority-ranked queue of items the learner has not mastered. Rewritten (not appended) at each teaching boundary. Derived from mastery table statuses. Used by phase skills to drive session planning.
• ## Engagement Signals — Lightweight motivational state (momentum, consecutive struggles, last celebration, notes). Updated by phase skills at teaching boundaries.

Each of Concepts, Chains, and Factors uses a mastery tracking table with columns:

• Status: not-mastered | partial | mastered
• Evidence: Compact append-only log of assessment events (e.g., "Recall pass (S2), chain trace fail (S3)"), each tagged with session number.
• Last Tested: Date of most recent assessment event.

Mastery status is updated by phase skills at every teaching boundary via the dln-sync agent. The orchestrator does not interpret mastery data — it passes the extracted Knowledge State block to the phase skill, which reads and acts on the tables.

Session Logs — Dated sections appended below Knowledge State by each phase skill. Contains session plan, progress notes, and plan adjustments. Old session logs are kept for audit but are NOT read back during mid-session syncs.

Page Body Initialization Template

When creating a new domain profile, write the skeleton from @references/init-template.md to the page body.

3. Orchestrator Flow

Step 1: Parse Domain

Extract the domain name from the user's message. Examples:

• "dln options pricing" → Options Pricing
• "learn compiler design from scratch" → Compiler Design
• "dln list" → list command
• "dln reset options pricing" → reset command

If no domain is specified, ask: "What domain would you like to learn? Give me a topic and I'll set up your learning path."

Step 2: Handle Special Commands

list — Query the DLN Profiles database and display all domains with their current phase, session count, last session date, and review status in a table. For each domain, compute review status:

• Overdue — Today is past Next Review date. Show how many days overdue in red: "⚠ 5 days overdue"
• Due today — Next Review is today. Show: "Due today"
• Upcoming — Next Review is in the future. Show: "In [N] days"
• No data — Next Review is empty (legacy profile). Show: "Not scheduled"

Sort the table with overdue domains first (most overdue at top), then due today, then upcoming.

Example output:

If no syllabus exists for a domain, show "No syllabus" in the Coverage column.

reset [domain] — Find the matching row. Confirm with the user before executing. Then:

1. Replace the page body with the initialization template from @references/init-template.md (clearing all Knowledge State and session logs)
2. Set Phase back to Dot
3. Reset Session Count to 0
4. Clear Last Session

Step 3: Query or Create Profile

Use the Notion MCP to query the DLN Profiles database for a row matching the domain name.

If found: Read the current Phase, Session Count, and page body content. Then validate the page body structure (see Step 3 validation below).

If not found: Create a new row with:

• Domain = parsed domain name
• Phase = Dot
• Session Count = 0

Then write the page body initialization template (see Schema section above) to the new page.

Set Next Review = tomorrow's date and Review Interval = 1 for new domains.

Tell the user: "New domain detected. Starting you in the Dot phase — we'll build your foundational concepts first."

Step 3 Validation: Page Body Structure Check

After loading an existing profile, check whether the page body contains these four core Knowledge State headers: ## Concepts, ## Chains, ## Factors, ## Compressed Model.

These four are sufficient because they are the structural headers that phase skills actively read mastery tables from. The remaining headers (## Interleave Pool, ## Calibration Log, ## Load Profile, ## Open Questions, ## Weakness Queue, ## Engagement Signals, ## Syllabus) are auxiliary — phase skills create them on first write if absent.

If all four core headers are present: The profile is valid. Proceed to Step 3a.

If any core header is missing: The profile exists but predates the current DLN template (or was created outside DLN). Auto-initialize:

1. Read the current page body content.
2. Write the initialization template from @references/init-template.md to the page body, appending the original content under a ## Prior Notes header at the bottom.
3. Backfill column properties, each only if that property is currently empty/missing:
   • Phase → Dot (only if empty)
   • Session Count → 0 (only if empty)
   • Next Review → tomorrow's date (only if empty)
   • Review Interval → 1 (only if empty)
4. Tell the user: "Upgraded your [domain] profile to the current DLN format. Your previous session content is preserved under Prior Notes."
5. Proceed to Step 3a as normal.

KS Boundary Markers

The Knowledge State block is wrapped in <!-- KS:start --> / <!-- KS:end --> HTML comment markers. These markers are managed by the dln-sync agent — the orchestrator does not add, remove, or check for them. If a profile is missing markers (pre-marker profiles), dln-sync will add them automatically on its first sync operation.

Step 3a: Review Check

After loading the profile, compute the review status by comparing today's date against the Next Review column value.

If the domain is overdue or due today:

1. Inform the learner:

"It's been [N] days since your last session on [domain]. Your Next Review was [date] — you're [N days] overdue. Before we continue with new material, let's do a quick retrieval warm-up to see what's stuck."

2. Route to the phase-appropriate Review Protocol (below) BEFORE routing to the phase skill for new teaching.

3. After the review protocol completes, route to the phase skill as normal (Step 4).

If the domain is not due for review: Proceed directly to Step 4 (no review needed).

If the domain has never been reviewed (Next Review is empty): This is a legacy profile. Set Next Review = today and Review Interval = 1, then proceed to Step 4. The review system activates from the next session onward.

Review Protocol

The review protocol runs inside the orchestrator, before the phase skill is invoked. It takes 3-8 minutes depending on phase.

Dot Phase Review:

• Ask the learner to list all concepts they remember from this domain (unprompted, no hints).
• Ask them to trace one causal chain from memory.
• Compare their recall against the Concepts and Chains in Knowledge State.
• Score: count recalled vs. total. Note which concepts were forgotten.
• If recall < 50%: warn the learner that significant decay has occurred. Recommend spending this session on reinforcement rather than new material. Pass review_results to the phase skill so it can prioritize re-teaching forgotten concepts.
• If recall >= 50%: acknowledge what they remembered, note gaps, and proceed to new material. Pass review_results to the phase skill for priority adjustment.

Linear Phase Review:

• Ask the learner to name the factors they've discovered so far (unprompted).
• Ask them to pick one factor and explain which chains it connects and why.
• Compare against the Factors section in Knowledge State.
• Score: count recalled factors vs. total, check if explanations are structural (not just names).
• Same threshold logic as Dot: < 50% triggers reinforcement recommendation.

Network Phase Review:

• Ask the learner to state their compressed model from memory (no looking at notes).
• Compare against the Compressed Model in Knowledge State.
• Score qualitatively: did they capture the core principles? What was lost?
• For Network phase, there is no "reinforcement" redirect — instead, the forgotten elements become the first stress-test targets in the session.

Interval Computation

After each session completes (regardless of whether a review protocol ran), compute the next review interval using these rules:

Base intervals by phase:

• Dot phase: intervals expand as 1 → 2 → 4 → 7 → 14 → 30 days
• Linear phase: intervals expand as 1 → 3 → 7 → 14 → 30 days
• Network phase: intervals expand as 2 → 7 → 14 → 30 → 60 days

Adjustment rules:

• If the review protocol ran and recall was >= 70%: advance to the next interval in the sequence. Set Review Interval to the new value.
• If the review protocol ran and recall was 50-69%: repeat the current interval (no advancement). Keep Review Interval the same.
• If the review protocol ran and recall was < 50%: reset the interval to the first value in the sequence for the current phase.
• If no review protocol ran (domain was not overdue): advance to the next interval in the sequence.
• If the learner's phase changed this session (phase gate passed): reset to the first interval of the NEW phase.

Set Next Review = today + Review Interval (in days).

Pass the computed Next Review and Review Interval values to the dln-sync agent in the replace-end dispatch as column_updates.

Step 3b: Syllabus Check

After loading the profile and running any review check, inspect the ## Syllabus section in the page body.

If the ## Syllabus section is empty or contains only the placeholder goal:

1. Tell the user: "No syllabus exists for this domain yet. Let me research and generate one based on your learning goal."
2. Spawn the dln-syllabus subagent via the Agent tool, passing in the prompt:
   • Domain name
   • The user's original goal prompt (from when they first created this domain, or ask them now)
   • Page ID for writing the syllabus to Notion
3. The subagent runs in its own context window — it does all web search, context7 lookups, and domain research there. Only the final topic list returns to the main session. This keeps the teaching context clean.
4. When the subagent returns, present the topic list to the user for review:

"Here's the syllabus I've generated for [domain] based on your goal:

• Topic A
• Topic B
• Topic C
• ...

[N] topics total. Add, remove, or edit anything before we start. You can also edit this anytime in Notion."

5. Apply the user's edits. If they request changes, update the ## Syllabus section in Notion directly from the orchestrator (this is cheap — just text manipulation, no research needed).
6. If the subagent fails or the user declines, proceed without a syllabus. The orchestrator skips coverage reporting. The syllabus can be generated in a future session.

If the ## Syllabus section has topics:

Compute and display coverage stats before routing to the phase skill:

Syllabus Progress:

• Coverage: [N]/[M] topics covered ([percentage]%)
• Mastery: [X] mastered, [Y] partial, [Z] not-mastered
• Uncovered: [list of unchecked topics with no concepts yet]

A topic is covered when at least one concept in ## Concepts has a matching Syllabus Topic column value. A topic is checked off when all related concepts are mastered.

Pass the syllabus content to the phase skill alongside the page body (it's already in the page body, so no additional passing is needed).

Step 4: Load Context and Route

Read the full page body of the domain's Notion page, then extract only the Knowledge State block to pass to the phase skill. This prevents old session logs from consuming context tokens.

Extraction rule: Find the <!-- KS:start --> and <!-- KS:end --> boundary markers. Extract everything between them (inclusive of markers). Discard everything after <!-- KS:end --> — that's session logs from prior sessions.

If the markers are missing (pre-marker profile), pass the full page body as-is and let dln-sync add markers on its first sync.

The extracted KS block includes the ## Syllabus section. Phase skills read it directly — no separate syllabus parameter is needed.

Step 5: Invoke Phase Skill

Pass the following context to the phase skill:

1. Domain name
2. Knowledge State block (extracted KS block from Step 4, not the full page body)
3. Session count (from Session Count column property — authoritative source)
4. Page reference (so the phase skill can write back to the page body)

Use the Skill tool to invoke the appropriate phase skill (dln-dot, dln-linear, or dln-network).

After the phase skill completes, no additional write-back is needed — the phase skill handles all Notion persistence.

4. Phase Transition Rules

Phase transitions are handled by the phase skills themselves:

• Dot → Linear: When the learner passes the Dot phase gate (can name concepts, explain causal chains, trace through a scenario). The dln-dot skill updates Phase to Linear.
• Linear → Network: When the learner passes the Linear phase gate (can name shared factors, predict unseen problems, identify minimal principle set). The dln-linear skill updates Phase to Network.
• Network is terminal — no further phase transitions. The skill tracks revision count and compression quality.

5. Error Handling

• Notion unavailable: Tell the user Notion is unreachable and offer to run the session without persistence (phase skill still works, just no state saved).
• Multiple domain matches: If the query returns multiple rows, show them and ask the user to clarify.
• Phase skill not found: This shouldn't happen in normal operation. If it does, report the error and suggest the user check their plugin installation.
• Notion fails mid-session: Phase skills delegate all Notion I/O to the dln-sync agent. If dln-sync returns with a failure status, the phase skill logs the intended update in-conversation, queues failed writes for the next dispatch, and falls back to in-conversation-only tracking if 3+ consecutive dispatches fail. See phase skill instructions for details.

6. Motivational Architecture

All DLN phase skills embed motivational design into their teaching. This is not a separate system — it is woven into every interaction. The principles:

1. Growth mindset framing: Attribute success to effort and strategy, not innate ability. "You worked through that" not "You're smart." Attribute struggle to the difficulty of the material or the need for a different approach, never to the learner's capacity.

2. Visible progress: At every checkpoint, tell the learner where they are relative to where they started. Use concrete counts: "You've mastered 8 of 12 concepts" or "Your model is 40% more compressed than last session." Progress must be tangible, not just "you're doing well."

3. Frustration detection and response: Monitor for frustration signals (see phase skill instructions). When detected, intervene immediately — do not push through. Simplify, provide a quick win, then rebuild momentum.

4. Desirable difficulty calibration: Learning should feel challenging but achievable. If the learner breezes through everything, increase difficulty. If they hit a wall, reduce scope and provide scaffolding. The target emotional state is "stretched but not overwhelmed."

5. Celebration at milestones: Phase transitions, mastery achievements, and session count thresholds are celebrated explicitly. Not with empty praise — with specific acknowledgment of what the learner accomplished and what it means.

The ## Engagement Signals section in the Knowledge State persists motivational context between sessions so the next session can calibrate its tone appropriately.

Momentum time-decay rule: If 7+ days have elapsed since the last session, reset Momentum to neutral regardless of its stored value. A fragile state from a bad session should not persist indefinitely — after a week, the learner has had enough distance that opening with fragile calibration feels mismatched. The phase skill reads Last Session from the profile and applies this rule before using the stored Momentum value.