Skip to content

OAA HIST 001 exemplar

OAA-HIST-001: When the Record and Reality Disagree

A Reference Lesson in Institutional Integrity

Lesson ID: OAA-HIST-001
Subject: History & Civic Integrity
Agent Teacher: History Agent (Mobius)
Status: REFERENCE LESSON (specification test, non-runtime)
Teaching Duration: ~20 minutes (video) + ~10 minutes (questionnaire)
Intended Audience: General citizens; no prior Mobius knowledge required
License: CC0 / Public Domain

OPENING QUESTION (2 minutes)

"When a trusted record is wrong, who gets to rewrite history?"

This is an old question. Medieval monasteries faced it when they discovered scribal errors in sacred texts. Renaissance merchants faced it when they found double-entry bookkeeping had been falsified. Modern institutions face it when archives are discovered to be incomplete or corrupted.

We'll explore this question through three paths:

  1. History — How have institutions handled this before?
  2. A real case — What happened when one system's records didn't match reality?
  3. A principle — What's the right way to fix a broken record?

HISTORY: THE PROBLEM (4 minutes)

Pedagogical note: Historical examples below use FACT / INFERENCE / INTERPRETATION labels per AGENT_TEACHER_CONTRACT.md. Pre-modern and corporate history is summarized from published scholarship — verify via the linked sources, not Mobius attestations.

Medieval Margin Corrections

FACT: Monks copied manuscripts by hand. When they noticed an error, they had three common options:

  • Erase the error (expensive vellum; often leaves physical traces)
  • Cross it out and write above (creates a visibly layered page)
  • Write a correction in the margin (preserves the original reading alongside the fix)

INTERPRETATION (History Agent): Margin corrections were a widespread practice because they kept the original visible while documenting the fix — the pattern this lesson later generalizes.

Source: Clemens & Graham, Introduction to Manuscript Studies (2007), corrections chapter; British Library medieval manuscripts collection

Double-Entry Bookkeeping

FACT: Luca Pacioli's Summa de arithmetica, geometria, proportioni et proportionalità (Venice, 1494) codifies double-entry bookkeeping: every transaction appears as matching debit and credit entries.

INFERENCE: When two ledgers must balance, one-sided falsification becomes easier to detect than in a single ledger.

INTERPRETATION (History Agent): Renaissance merchants treated redundant records as an anti-corruption device — corruption becomes visible when entries fail to reconcile.

Source: Pacioli (1494); Gleeson-White, Double Entry: How the Merchants of Venice Created Modern Accounting (2011)

The Enron Case

FACT: Enron filed for bankruptcy in December 2001. U.S. Senate and SEC investigations documented off-balance-sheet partnerships, earnings manipulation, and destruction of documents relevant to the fraud.

INFERENCE: When original records are destroyed, investigators must reconstruct events from fragments, testimony, and secondary artifacts — at enormous cost and with higher dispute risk.

INTERPRETATION (History Agent): Enron illustrates erasure: the institution's published records diverged from operational reality, and document destruction made reconciliation harder, not easier.

Source: U.S. Senate PSI, The Role of the Board of Directors in Enron's Collapse (2002); SEC litigation release archive — Enron

The Pattern

INTERPRETATION (History Agent):

  • Medieval scribes learned: preserve the original, annotate the correction
  • Renaissance merchants learned: redundant records expose corruption
  • Modern audits learned: trail the decision-making, not just the outcome

The principle: An institution that can't show its own mistakes has no credibility.

CASE STUDY: THE MOBIUS RESERVE BLOCK COLLISION (6 minutes)

The Setup

Mobius is a civic AI system. Like any institutional record-keeper, it maintains sealed records that must stay internally consistent.

Two layers matter here:

  • Seal index: Which seal IDs are attested and in what order
  • Seal bodies: The cryptographic content each ID points to

In July 2026, reconciliation work under cycle C-397 discovered they did not always agree on history.

The Discovery

An audit of production KV showed:

  • 360 seal IDs in the attested index (FACT: C397_INDEX_EXAMINATION_GAP_WITNESS.json → counts.vault_index_total)
  • 319 seal records examined in the collision audit (FACT: same witness → counts.examination_scope_modern)
  • 41 index-only IDs classified as legacy MIC tranche C-299–C-307, outside collision audit scope (FACT: gap witness → counts.index_examination_gap)
  • 194 unique block_number positions among those examined
  • 125 hash-divergent collision pairs — two different seal bodies claiming the same position with different hashes

FACT: This is a lineage collision, not silent deletion of seal bodies. Competing historical seals were preserved; the index and examination counts diverged because multiple lineages coexist.

This is the problem: Which record is authoritative when two sealed bodies disagree at the same position? If you pick a winner without showing the contradiction, you rewrite history.

The Corruption Origin

INFERENCE: Investigation traced competing lineages to forked sealing history during the Reserve Block era (cycles C-308 through C-372), not to a single migration bug erasing past state.

FACT: A separate but related incident — the vault:seal:latest pointer corruption — occurred when POST /api/vault/migrate-v1 was called with {"sealId":"latest"}, colliding with the CAS pointer key. That bug is documented in mobius-civic-ai-terminal PR #648 (commit f38ff697). It is an index/evidence mismatch of a different shape, illustrating the same principle: when pointer and payload disagree, decisions built on the pointer are unsafe.

Key fact: Neither incident was intentional fraud. Both violated the invariant that records must be reconcilable without erasing the contradiction.

Why This Matters

An institution relying on an unresolved index would build decisions on contested lineage. If the seal index says one history and the bodies say another, you cannot know which MIC issuance, quorum attestation, or reserve position is legitimate.

Example: If two seals claim block position 42 with different hashes, promoting either one without witness evidence silently erases the other lineage.

The Correction (In Progress)

Mobius's response under C-397 (July 2026):

  1. Acknowledged the collision: Published reconciliation witness in EPICON cycle C-397
  2. Documented the investigation: Preserved C397_RESERVE_BLOCK_COLLISION_WITNESS.json with all 125 pairs
  3. Held canon fail-closed: SEAL_INTEGRITY_GATE remains active; canonical_reserve_blocks stays null until Track R repair is approved
  4. Preserved competing seals: No deletion or renumbering of competing bodies during evidence recovery
  5. Made it auditable: Original collision state remains in KV and git witness artifacts

Key fact: Mobius did NOT delete the competing seals to force a clean index. It preserved evidence that says "here's what we believed was true; here's what we discovered; here's what remains contested."

You can verify this case yourself:

PRINCIPLE: NEVER REPAIR EVIDENCE BY PRETENDING THE CONTRADICTION NEVER HAPPENED (3 minutes)

From this case, extract the core principle:

"Never repair evidence by pretending the contradiction never happened."

What It Is NOT

  • NOT: "Never fix broken systems"
  • NOT: "Always keep corrupted data operational"
  • NOT: "Acknowledge error but do nothing"

What It IS

  • YES: "When you fix something, show the repair"
  • YES: "Keep the original contradiction visible"
  • YES: "Let others audit the correction process"
  • YES: "Make it possible to trace what changed and why"

Historical Precedent

  • Medieval monks: Margin corrections showed the original error and the fix
  • Double-entry bookkeeping: Both ledgers preserved; imbalance exposes fraud
  • Scientific method: Papers show failed hypotheses before new models
  • Code version control: Git history preserves every version, every mistake

OPERATIONALIZATION: HOW MOBIUS ENFORCES THIS PRINCIPLE (5 minutes)

Principle alone doesn't enforce integrity. You need a system that makes it hard to violate the principle.

Mobius uses EPICON — a record structure that makes erasure difficult:

Original State
  ↓
Contradiction Discovered
  ↓
Investigation Documented
  ↓
Correction Applied (or held fail-closed until approved)
  ↓
Provenance Chain Sealed

Not optional steps. The system is designed so that you can't have a correction without showing all five elements.

Example: C-397 Reserve Block Case

If someone asks "Did the Reserve Block ever have lineage collisions?", the answer is:

✓ Original State: 125 hash-divergent pairs documented (FACT, witness JSON)
✓ Contradiction: Index and bodies disagree on winners (FACT, audit metrics)
✓ Investigation: Lineage forks traced; pointer bug separately documented (FACT, C-397 + PR #648)
✓ Correction: Track R repair gated; gate held until ZEUS + human approval (FACT, reconciliation doc)
✓ Provenance: All changes sealed with timestamps and intent blocks (FACT, EPICON C-397)

An institution can say "we fixed it," but only if all five elements are present. If any are missing, the correction is marked INCOMPLETE or UNVERIFIED.

Constitutional Enforcement

Mobius's constitution requires that:

  • Correction cannot be sealed without EPICON intent block
  • EPICON intent block must cite evidence
  • Evidence must be auditable
  • Original state must be preserved

This doesn't prevent dishonest correction. But it makes dishonest correction visible to auditors. The system is fail-transparent, not fail-proof.

ASSESSMENT: CAN YOU REASON ABOUT THIS? (5 minutes)

The questionnaire tests whether you understand the principle, not whether you agree with Mobius's specific solution.

Gate 1: Problem Recognition — Why is it dangerous when an institution's records don't match its evidence?

Gate 2: Distinction — How is correcting a record different from rewriting history?

Gate 3: Evidence Awareness — What evidence would justify a correction?

Gate 4: Principle Transfer — Apply the principle to a domain Mobius hasn't touched.

Gate 5: Adversary Coherence — If you disagreed with Mobius's approach, what would you do instead?

See ASSESSMENT_PROTOCOL.md for rubric and disagree-and-pass rule.

ATTESTATION MODEL

After you complete the 5-gate questionnaire, you'll receive an attestation:

Attestation ID: ATT-OAA-HIST-001-[YOUR-ID]-[DATE]
Status: VERIFIED
Claim: You demonstrated comprehension of institutional integrity
Appeal: You have 30 days to challenge this assessment

This attestation doesn't mean you agree with Mobius or are qualified to run institutional systems. It means you understand why record integrity matters, can tell honest correction from erasure, and can participate in governance discussions on this topic.

APPENDIX: VERIFY THIS YOURSELF

The C-397 Reserve Block collision:

# From Mobius-Substrate repo root
cat docs/epicon/cycles/C-397/C397_RESERVE_BLOCK_COLLISION_WITNESS.json | jq '.pair_count'
# Expected: 125

grep -A5 "125" docs/epicon/cycles/C-397/RESERVE_BLOCK_RECONCILIATION.md

# Index/examination gap (41 legacy IDs)
cat docs/epicon/cycles/C-397/C397_INDEX_EXAMINATION_GAP_WITNESS.json | jq '.counts'

The vault pointer guard (related case):

# In mobius-civic-ai-terminal
git show f38ff697 --stat

EPICON intent for Academy (this lesson's constitutional frame):

REFERENCE EXEMPLAR COMPLIANCE CHECKLIST

This lesson satisfies the Academy Protocol by:

  • ✅ Teaches problem before solution — Historical corruption before EPICON operationalization
  • ✅ Distinguishes evidence from interpretation — FACT / INFERENCE / INTERPRETATION labels in historical and Mobius sections
  • ✅ Invites disagreement — Gate 5 asks for coherent alternatives
  • ✅ Traces all claims — Witness JSON, gap witness, reconciliation doc, PR #648, and external historical sources
  • ✅ Acknowledges perspective — History Agent states vantage point
  • ✅ Preserves original error — Competing seals not erased; gate held fail-closed
  • ✅ Follows assessment protocol — 5 comprehension gates, not ideological agreement
  • ✅ Honors agent contract — Explicitly invites learner to disagree and still pass

CONSTITUTIONAL BINDING

Lesson Protocol Version: 0.1
Assessment Protocol Version: 0.1
Agent Teacher Contract Version: 0.1
Binding Disciplines: EPICON (C-399), Witness Protocol (C-373), Goodhart Resistance (§17)

Status: REFERENCE LESSON
Runtime Status: Non-runtime (specification test only)
Approval for Teaching: Pending OAA Phase D gate verification + human merge

Document Status: Reference Exemplar
Tested Against: ACADEMY_PROTOCOL, ASSESSMENT_PROTOCOL, AGENT_TEACHER_CONTRACT
Ready for: First teaching cycle upon Phase D gate approval