Home How it works Contradiction Handling
Conflict resolution

AI memory should not choose winners silently.

When beliefs conflict, Tenure can flag the contradiction, preserve both sides, apply organization standards, and create a reviewable path to resolution.

Detect

Conflicts are identified explicitly.

Contradictions can come from extraction signals, confidence comparisons, compaction scans, or organization standards.

Preserve

Evidence is not erased.

Tenure keeps provenance and lineage so teams can see where conflicting memory came from.

Resolve

Policy can override preference.

Org standards can act as absolute constraints while user and team preferences remain reviewable.

The conflict path

Contradiction is a state-management problem.

Most memory systems retrieve whichever item scores highest. Tenure treats conflict as a lifecycle event: detect it, record it, keep the evidence, and resolve it without hiding the transition.

01

Compare beliefs

Tenure checks whether two beliefs assert incompatible things about the same subject.

02

Apply confidence

If content differs within the confidence margin, the conflict is flagged instead of overwritten.

03

Respect policy

Organization standards can mark a belief as conflicting with policy even without another belief.

04

Keep it reviewable

Contradictions remain pending until resolved, with belief ids, reason, scope, and timestamps.

Conflict states

A contradiction is not a failed memory. It is useful evidence.

When a team changes its mind, when a user preference conflicts with a team norm, or when a policy disagrees with a local convention, Tenure makes that visible.

Contradiction record
active Use REST for public API integrations. team:platform
pending conflict Use GraphQL for public API integrations. same subject · incompatible approach
resolved later Use REST for public API integrations unless the partner already requires GraphQL. supersedes prior versions
Belief conflicts

Same subject. Incompatible claims.

Tenure flags clear conflicts even when the beliefs do not share the same canonical name.

  • Preference conflicts stay pending instead of being collapsed.
  • Fact and decision conflicts include a one-sentence reason.
  • Existing pending pairs are deduplicated.
  • Resolution can supersede one or both earlier beliefs.
Org violations

Policy can be an absolute constraint.

When org standards are supplied, a belief can be flagged against the organization standard itself.

  • Contradictions can use the organization as the opposing side.
  • Merged content is not allowed to contradict org standards.
  • Non-user-edited violations can be auto-superseded.
  • Teams keep a traceable record of the policy conflict.

The goal is not to pretend memory is always consistent. The goal is to make inconsistency observable, reviewable, and resolvable.

Resolution model

Contradictions become decisions, not hidden prompt noise.

A resolved contradiction can produce a more precise belief while keeping older assertions available for audit and lineage.

// detected conflict
{ belief_ids: ["use_rest_public_api", "use_graphql_public_api"], status: "pending" }

// resolved state
{ canonical_name: "public_api_integration_style", content: "Use REST for public integrations unless a partner contract requires GraphQL.", supersedes: ["use_rest_public_api", "use_graphql_public_api"] }
Why teams need this

Shared memory creates shared disagreement.

Solo memory can survive with rough edges. Team memory cannot. As soon as multiple developers, repos, tools, and policies contribute context, contradiction handling becomes part of the infrastructure.

Without contradiction handling

The model receives whichever memory wins retrieval.

  • Old decisions can appear beside new decisions.
  • User preferences can override team conventions accidentally.
  • Policy conflicts become invisible prompt noise.
With Tenure

The system knows when memory disagrees.

  • Conflicts are recorded as first-class objects.
  • Org standards can constrain what gets merged and injected.
  • Resolved beliefs preserve a change history.
The point

Trust requires visible disagreement.

Tenure helps teams keep AI memory useful as facts, preferences, standards, and decisions evolve over time.