Error Recovery (Core)

“The court does not fail gracefully. The court recovers deliberately.”
— The Honorable Lucius J. Morningstar

This document defines procedures for recovering from state corruption, rolling back bad decisions, and handling emergency situations during MORNINGSTAR sessions.

State Corruption Recovery
Decision Rollback Protocol
Emergency Procedures
Prevention Measures
Recovery Checklist

State Corruption Recovery

Detecting Corruption

State corruption is indicated by:

Symptom	Likely Cause
Validation errors on `/morningstar`	Schema violation in `state/current.md`
Missing required sections	Incomplete update or file truncation
Inconsistent dates	Manual editing error or clock issues
Vote tallies don’t sum	Transcription error
Session ID mismatch	State from different session
Metrics don’t match records	Counter drift

Recovery Procedure

Level 1: Minor Corruption (Single Field)

Use when: One or two fields are invalid but overall structure is intact.

┌─────────────────────────────────────────────────────────────────┐
│ LEVEL 1 RECOVERY                                                │
└─────────────────────────────────────────────────────────────────┘

1. Identify the specific validation error
2. Check CHANGELOG.md for correct value
3. Correct the field directly in state/current.md
4. Run validation again
5. Document the correction in Notes section

Duration: < 5 minutes
Risk: Low

Level 2: Section Corruption (One Section Invalid)

Use when: An entire section is missing or malformed.

┌─────────────────────────────────────────────────────────────────┐
│ LEVEL 2 RECOVERY                                                │
└─────────────────────────────────────────────────────────────────┘

1. Identify the corrupted section
2. Reference templates/session-start.md for correct structure
3. Reconstruct section from:
   a. CHANGELOG.md (for decisions)
   b. courtroom/transcripts/ (for deliberation details)
   c. Previous state versions (if available)
4. Insert reconstructed section
5. Validate entire document
6. Add recovery note with sources

Duration: 5-15 minutes
Risk: Medium

Level 3: Major Corruption (Multiple Sections)

Use when: State is largely unusable but some data can be salvaged.

┌─────────────────────────────────────────────────────────────────┐
│ LEVEL 3 RECOVERY                                                │
└─────────────────────────────────────────────────────────────────┘

1. PRESERVE corrupted state:
   mv state/current.md state/corrupted-YYYY-MM-DD.md

2. Create fresh state from template:
   cp templates/session-start.md state/current.md

3. Reconstruct from authoritative sources:
   
   Source Priority:
   ┌────────────────────────────────────┐
   │ 1. courtroom/transcripts/ (highest)│
   │ 2. CHANGELOG.md                    │
   │ 3. state/metrics.md               │
   │ 4. Corrupted file (fragments)     │
   │ 5. User memory (lowest)           │
   └────────────────────────────────────┘

4. Rebuild each section methodically
5. Validate completely
6. Document reconstruction in Notes

Duration: 15-30 minutes
Risk: High (data loss possible)

Level 4: Total Corruption (State Unrecoverable)

Use when: State file is missing, empty, or completely invalid.

┌─────────────────────────────────────────────────────────────────┐
│ LEVEL 4 RECOVERY                                                │
└─────────────────────────────────────────────────────────────────┘

1. Accept that current session state is lost

2. Create new session:
   - Generate new session ID
   - Use clean template
   - Set status to 'not_started'

3. Preserve institutional memory:
   - Review all files in courtroom/transcripts/
   - Extract key context from CHANGELOG.md
   - Summarize in Session Memory > Key Context

4. Document the incident:
   - Add entry to CHANGELOG.md noting state loss
   - Record what was lost vs. preserved

5. Proceed with new session

Duration: 30+ minutes
Risk: High (session continuity lost)

Recovery Source Priority

When reconstructing state, prioritize sources by reliability:

Priority	Source	Reliability	Contents
1	`courtroom/transcripts/`	Highest	Full deliberation records
2	`CHANGELOG.md`	High	Decision summaries
3	`state/metrics.md`	Medium	Aggregate statistics
4	Corrupted state file	Low	Fragments may be valid
5	User recollection	Lowest	Prone to error

Decision Rollback Protocol

When to Roll Back

A decision should be rolled back when:

The decision was based on incorrect information
Material circumstances have changed
Implementation revealed unforeseen consequences
The original vote was procedurally invalid

Rollback Procedure

┌─────────────────────────────────────────────────────────────────┐
│ DECISION ROLLBACK PROTOCOL                                      │
└─────────────────────────────────────────────────────────────────┘

PHASE 1: MOTION TO RECONSIDER
─────────────────────────────

1. Any personality may move to reconsider:
   "I move to reconsider [CASE-ID] due to [REASON]."

2. Motion requires:
   - Specific case reference
   - Concrete reason for reconsideration
   - New information or changed circumstances

3. Second required from another personality

PHASE 2: DELIBERATION
─────────────────────

4. The court deliberates on WHETHER to reconsider
   (Not yet on the original matter)

5. Vote on motion to reconsider:
   - Pass: Requires same threshold as original matter
   - Fail: Original decision stands

PHASE 3: RE-DELIBERATION (if motion passes)
──────────────────────────────────────────

6. Original matter is re-opened
7. Full deliberation procedure applies
8. New arguments and evidence presented
9. New vote taken

PHASE 4: RECORDING
─────────────────

10. Document in CHANGELOG.md:
    - Original decision
    - Rollback motion and vote
    - New decision (if changed)
    - Rationale for change

11. Add addendum to original transcript (do not modify):
    
    ## Addendum [DATE]
    
    **Status:** RECONSIDERED
    **Motion:** [Vote on reconsideration]
    **New Ruling:** [Case-ID of new deliberation]
    **Reason:** [Why reconsidered]

12. Create new transcript for re-deliberation

Rollback Constraints

Constraint	Rule
Time limit	No limit, but older decisions harder to roll back
Frequency	Same matter cannot be reconsidered more than twice per session
Implementation	If implementation has begun, assess rollback cost
Precedent	Rolled-back decisions lose precedential weight

Emergency Rollback

For decisions causing active harm:

┌─────────────────────────────────────────────────────────────────┐
│ EMERGENCY ROLLBACK                                              │
└─────────────────────────────────────────────────────────────────┘

1. Judge may unilaterally suspend a decision if:
   - Active harm is occurring
   - Delay would worsen the situation
   - Immediate action is required

2. Suspension is temporary (24 hours maximum)

3. Full reconsideration deliberation MUST follow

4. Document as:
   
   ## Emergency Suspension
   
   **Decision:** [CASE-ID]
   **Suspended:** [TIMESTAMP]
   **Reason:** [Active harm description]
   **Status:** Pending reconsideration

Emergency Procedures

Session Crash Recovery

If a session terminates unexpectedly:

┌─────────────────────────────────────────────────────────────────┐
│ SESSION CRASH RECOVERY                                          │
└─────────────────────────────────────────────────────────────────┘

1. Do NOT run /morningstar immediately

2. Assess state integrity:
   - Open state/current.md directly
   - Check for truncation
   - Verify last updated timestamp

3. If state appears valid:
   - Run /morningstar normally
   - Note "session recovered from crash" in state

4. If state is corrupted:
   - Follow State Corruption Recovery (Level 2-4)

5. Check for orphaned work:
   - Review any transcripts started but not certified
   - Check CHANGELOG.md for unpersisted entries

Deadlocked Deliberation

When the court cannot reach resolution after 3 rounds:

┌─────────────────────────────────────────────────────────────────┐
│ DEADLOCK RESOLUTION                                             │
└─────────────────────────────────────────────────────────────────┘

Option A: REFRAME
─────────────────
1. Judge calls timeout
2. Identify the crux disagreement
3. Reframe the question to address crux
4. Resume deliberation with new framing

Option B: TIME-BOX EXPERIMENT
────────────────────────────
1. Identify the lowest-risk option
2. Implement with explicit time limit
3. Reconvene after time-box to evaluate
4. Record as "experimental ruling"

Option C: DEFER TO EXPERTISE
───────────────────────────
1. Seat appropriate domain Specialist
2. Weight Specialist perspective heavily
3. Re-vote with additional input

Option D: JUDICIAL RULING
────────────────────────
1. Judge rules unilaterally
2. Document as "Judicial ruling due to deadlock"
3. Record all positions in dissent
4. Flag for reconsideration if circumstances change

Corrupted Transcript Recovery

If a transcript is corrupted or missing:

┌─────────────────────────────────────────────────────────────────┐
│ TRANSCRIPT RECOVERY                                             │
└─────────────────────────────────────────────────────────────────┘

1. Check CHANGELOG.md for decision summary
2. Check state/current.md for deliberation record
3. Reconstruct transcript with available information
4. Mark as [RECONSTRUCTED] in header:

   **Case No.:** [YYYY-CATC-NNN-DDD] [RECONSTRUCTED]  (per `core/case-format.md`)

5. Document sources used in reconstruction
6. Note reduced precedential weight

Context Loss Recovery

When the court loses context mid-deliberation:

┌─────────────────────────────────────────────────────────────────┐
│ CONTEXT LOSS RECOVERY                                           │
└─────────────────────────────────────────────────────────────────┘

1. Pause current deliberation
2. Run /update to checkpoint current state
3. Request context refresh:
   - What was the original matter?
   - What arguments were presented?
   - Where in the procedure are we?
4. Resume from known point
5. If arguments lost, personalities may re-present briefly

Prevention Measures

State Backup Practices

Practice	Frequency	Method
Checkpoint on `/update`	Each command	Automatic
Pre-deliberation snapshot	Before F3+ matters	Manual or automatic
Session start validation	Each `/morningstar`	Automatic
End-of-day backup	Daily	Manual recommended
Pre-session backup	Before major/F3+ sessions	Copy state/current.md to state/backups/YYYY-MM-DD-current.md; see procedures.md

Integrity Checks

Run these checks periodically:

┌─────────────────────────────────────────────────────────────────┐
│ INTEGRITY VERIFICATION                                          │
└─────────────────────────────────────────────────────────────────┘

□ state/current.md passes schema validation
□ All transcripts have certification
□ CHANGELOG.md entries match transcripts
□ Vote tallies are mathematically correct
□ Dates are consistent across documents
□ Metrics align with recorded decisions
□ Precedent index is current

Early Warning Signs

Watch for these indicators of impending problems:

Warning Sign	Potential Issue	Action
Dates inconsistent	Manual editing error	Verify and correct
Metrics drifting	Counter not updating	Recalculate from records
Missing transcripts	Save failed	Reconstruct immediately
Stale state	Session not closed properly	Update state
Orphaned decisions	Changelog not updated	Add missing entries

Recovery Checklist

Quick Reference: State Recovery

Quick Reference: Decision Rollback

Motion to reconsider with reason
Second from another personality
Vote on motion to reconsider
If passed: full re-deliberation
Record in CHANGELOG.md
Add addendum to original transcript
Create new transcript for new decision

Quick Reference: Emergency Response

Emergency	First Action
State corruption	Assess level, follow recovery
Session crash	Check state before restart
Decision causing harm	Emergency suspension
Deadlocked court	Reframe or time-box
Missing transcript	Reconstruct with flag
Context loss	Pause, checkpoint, refresh

Recovery Log Template

Document all recoveries for institutional learning:

## Recovery Log Entry

**Date:** [YYYY-MM-DD]
**Type:** [State Corruption / Decision Rollback / Emergency]
**Level:** [1-4 for state, N/A for others]

### Incident

[What happened]

### Root Cause

[Why it happened]

### Recovery Actions

1. [Action taken]
2. [Action taken]

### Data Lost

[What could not be recovered, if any]

### Prevention

[How to prevent recurrence]

### Time to Recover

[Duration]

“Failure is inevitable. Recovery is mandatory.”
— MORNINGSTAR::DEBUGGER

Table of Contents