Skip to content

PR7 Adversarial History-Integrity Review v1

Status: Normative adversarial supplement for PR7

This review attacks the PR7 history layer after the first exact-head implementation gate. It does not add a mutable history workflow, global registry, policy-authentication service, or correction/retraction engine.

1. Historical integrity is not current-state integrity

PR7 preserves a separate history layer:

CURRENT STATE != HISTORY
CURRENT READINESS LOSS != ACHIEVEMENT ERASURE
ACHIEVEMENT != CURRENT READINESS

An achievement records an accomplishment under an exact family ref and declared qualification context. It does not assert permanent readiness.

2. Retroactive family redefinition

AchievementInstance.family_ref is exact and revisioned. Validation against an AchievementFamilyCatalog fails closed when the supplied catalog contains the same stable family id at another revision.

FAMILY @1 != FAMILY @2
EXACT FAMILY REF != LATEST FAMILY REF
CURRENT FAMILY CATALOG != HISTORICAL REVISION ARCHIVE

PR7 v1 deliberately does not claim that an exact family-ref string is a content hash or globally authenticated semantic identity. A durable system must retain/import the exact semantic snapshot needed to interpret historical refs and must not reuse the same exact ref for materially different content.

EXACT REF != CONTENT HASH
SAME REF ACROSS INDEPENDENT STORES != PROOF OF SAME CONTENT
HISTORICAL INTERPRETATION REQUIRES RETAINED EXACT SEMANTICS

A material change to what an accomplishment means should not be disguised as a same-revision edit. Depending on semantic continuity, it requires a new revision or a new stable family identity.

3. Qualification-time boundary

PR7 v1 does not introduce a separate mutable qualification-decision object or a second qualified_at timestamp. The immutable AchievementInstance.recorded_at boundary is the time by which the declared qualification record and all internal qualification basis must exist.

Historical backfill remains valid:

achieved_at < recorded_at

A later claim/evaluation may help qualify an earlier accomplishment if it exists by recorded_at. A basis created after recorded_at is rejected.

ACHIEVEMENT TIME != QUALIFICATION RECORD TIME
LATE REVIEW != RETROACTIVE EVENT TIME
BASIS AFTER recorded_at != VALID QUALIFICATION BASIS

PR7 does not claim to preserve a separate earlier moment at which a human or model privately reached the decision before the immutable record was written. A future workflow may add such an event if that distinction becomes operationally necessary.

4. Blocker repair — achievement replay

The first implementation allowed two different AchievementInstanceId records to reuse the same event-bearing EvidenceRecord or EXTERNAL_ARTIFACT basis for the same stable AchievementFamilyId.

That could inflate history without a second distinguishable accomplishment.

Repair: within one PersonalHistoryRecordSet, the same exact event-bearing basis ref cannot be reused by multiple achievement instances of one stable family identity, even across family revisions.

NEW ACHIEVEMENT ID != NEW ACCOMPLISHMENT
SAME EVENT BASIS + SAME STABLE FAMILY != TWO ACHIEVEMENTS
REPLAYED BASIS != REPEATED ACCOMPLISHMENT

The same event may still legitimately support different achievement-family identities. If one evidence record describes several distinct repeated events, PR7 v1 has no per-event disambiguator and therefore fails closed rather than manufacturing countable instances.

This guarantee is snapshot-local. Replaying the same event in two independently assembled snapshots cannot be detected without persistence/import reconciliation.

5. Milestone significance is attributed content, not authority

PersonalMilestoneEvent.significance_note is part of the immutable record attributed to its MilestoneRecorderRef and recording policy.

It is not automatically a statement endorsed by the subject and is not a scalar importance measurement.

SIGNIFICANCE NOTE != SUBJECT ENDORSEMENT
RECORDER != AUTHORITY
MODEL-RECORDED SIGNIFICANCE != SUBJECT-ASSERTED SIGNIFICANCE
MILESTONE SIGNIFICANCE != GLOBAL IMPORTANCE

PR7 intentionally exposes no importance_score, prestige tier, milestone points, or canonical significance rank.

6. Blocker repair — future-record Legend laundering

The first implementation enforced:

source event time <= legend.as_of

but did not require the cited history record to exist by the time the Legend was generated. That allowed a Legend with generated_at=2026 to cite a milestone recorded in 2027 as long as the milestone described an earlier event.

Repair: a Legend source must satisfy both:

source achieved_at/occurred_at <= legend.as_of
source recorded_at <= legend.generated_at

This still permits honest historical reconstruction. A milestone that happened in 2024, was backfilled in 2026, and is used by a Legend generated in 2027 with as_of=2024 is valid.

AS_OF != KNOWLEDGE-AVAILABLE-AT
HISTORICAL BACKFILL != FUTURE-RECORD LAUNDERING

7. Selective-history distortion

A PersonalLegend is intentionally selective. Different valid Legends may choose different subsets and orderings from the same underlying history.

LEGEND != COMPLETE HISTORY
LEGEND OMISSION != HISTORY DELETION
LEGEND SELECTION != GLOBAL IMPORTANCE
LEGEND ORDER != GLOBAL IMPORTANCE
PERSONAL LEGEND != OFFICIAL SELF-NARRATIVE

PR7 can validate that every Legend entry cites real history and respects time boundaries. It cannot prove that natural-language narrative is fair, complete, emotionally appropriate, or free of framing bias.

Those are projection/review/UI concerns, not facts derivable from local source-reference validation.

No canonical_legend, official_legend, latest-wins rule, or completeness flag is inferred by core PR7.

8. Blocker repair — history-to-evidence feedback

AchievementInstance, PersonalMilestoneEvent, and PersonalLegend are not PR2 evidence records.

The first implementation nevertheless permitted an opaque history id to be copied into a generic PR2 provenance/payload ref. That allowed a cycle such as:

AchievementInstance
        ↓ relabeled as ARTIFACT
EvidenceRecord
        ↓ achievement basis
AchievementInstance

Repair: when PR7 history is cross-validated against a supplied EpistemicRecordSet, exact achievement/milestone ids may not appear as evidence or claim provenance source refs, and may not appear as evidence payload refs.

HISTORY RECORD != EVIDENCE SOURCE BY RELABELING
HISTORY ID != GENERIC ARTIFACT ESCAPE HATCH
HISTORY -> EVIDENCE -> HISTORY != VALID PR7 CYCLE

This is intentionally snapshot-local. PR7 cannot detect someone exporting history text to a new external artifact id and later importing that new artifact as evidence without preserved lineage. Such provenance laundering belongs to future persistence/import governance.

9. Cross-snapshot ID reuse

Achievement, milestone, and Legend ids are opaque identifiers with uniqueness enforced inside their corresponding snapshot collections.

They are not content hashes and PR7 has no global registry.

SNAPSHOT-LOCAL UNIQUENESS != GLOBAL UNIQUENESS
OPAQUE HISTORY ID != CONTENT HASH
SAME ID ACROSS SNAPSHOTS != SAME MATERIAL RECORD

A future durable store/import/sync layer must prevent historical id reuse for materially different content and reconcile collisions explicitly.

10. Correction and retraction pressure

PR7 history records are immutable, but immutability does not mean that a false or fraudulently qualified record can never be corrected.

IMMUTABLE HISTORY != IRRETRACTABLE FALSEHOOD
READINESS DECAY != ACHIEVEMENT RETRACTION
QUALIFICATION CORRECTION != CURRENT-STATE CHANGE

PR7 v1 deliberately does not define HistoryCorrection, AchievementRetraction, effective-history precedence, deletion, or in-place mutation.

A future correction/retraction layer should be append-only and provenance-preserving. It must distinguish at least:

  • the historical event actually did not happen;
  • the event happened but did not satisfy the declared family/policy;
  • later evidence changed current readiness only;
  • the record payload itself was entered incorrectly;
  • a privacy/removal policy affects visibility without pretending the event never occurred.

Until that governance exists, core PR7 exposes no revoke_achievement, delete_achievement, correct_history_in_place, or equivalent shortcut.

11. Frozen adversarial invariants

NEW ID != NEW ACCOMPLISHMENT
REPLAYED EVENT BASIS != REPEATED ACCOMPLISHMENT

EXACT FAMILY REF != LATEST FAMILY REF
EXACT FAMILY REF != CONTENT HASH
CURRENT FAMILY CATALOG != HISTORICAL REVISION ARCHIVE

recorded_at IS THE PR7 v1 QUALIFICATION-RECORD BOUNDARY
BASIS AFTER recorded_at != VALID QUALIFICATION BASIS

SIGNIFICANCE NOTE != SUBJECT ENDORSEMENT
MODEL-RECORDED SIGNIFICANCE != SUBJECT-ASSERTED SIGNIFICANCE

LEGEND != COMPLETE HISTORY
LEGEND OMISSION != HISTORY DELETION
LEGEND SELECTION != GLOBAL IMPORTANCE

SOURCE EVENT TIME <= LEGEND as_of
SOURCE RECORD TIME <= LEGEND generated_at

HISTORY RECORD != EVIDENCE RECORD
HISTORY ID != EVIDENCE/PROVENANCE ESCAPE HATCH
HISTORY -> EVIDENCE -> HISTORY != VALID CYCLE

SNAPSHOT-LOCAL UNIQUENESS != GLOBAL UNIQUENESS
OPAQUE HISTORY ID != CONTENT HASH

IMMUTABLE HISTORY != IRRETRACTABLE FALSEHOOD
CORRECTION/RETRACTION != IN-PLACE MUTATION

12. Non-goals of this repair pass

This adversarial pass does not add:

  • a historical semantic archive service;
  • family-ref content hashing or signatures;
  • a qualification workflow engine;
  • global event identity/fingerprinting;
  • a cross-snapshot ID registry;
  • semantic duplicate detection across differently named external artifacts;
  • natural-language Legend truth/fairness scoring;
  • subject-endorsement workflow;
  • correction/retraction precedence;
  • persistence transactions;
  • sync reconciliation;
  • publication or public-profile governance.