Documentation Index
Fetch the complete documentation index at: https://docs.mnemom.ai/llms.txt
Use this file to discover all available pages before exploring further.
Mnemom seeds three Trust Postures at migration time. They are platform-scope, immutable, and appear in every customer’s library. You either:
- Adopt one as-is — set it as your org’s default (or assign it to a specific team) and you’re shipping with sensible policy from day one.
- Clone and customize — start from a Mnemom default, create your own org-scope copy, edit thresholds to your fleet’s needs.
The three are:
| Posture | When to choose | Cadence | Stricter than Standard? | Looser than Standard? |
|---|
| Standard | Default starting point. Sensible for most fleets. | 10 min | — | — |
| High-Compliance | Banking, healthcare, regulated industries. | 5 min | Tighter thresholds, faster cadence, louder severity, reputation-weighted fault-line analysis | — |
| Low-Latency | Real-time agent UX where observability cost matters. | 30 min | — | Permissive thresholds, slower cadence, lower severity, cluster-partition pattern off |
Standard
Posture ID: tp-platform-standard · When to pick: “I don’t know which one to choose.”
Standard is the cascade-closure floor for every team in every org. Even when no posture is explicitly assigned anywhere, every team’s effective posture will at least be Standard. It’s the Charter §I13 guarantee.
Detection coverage
| Axis | Setting |
|---|
coherence.enabled | true |
coherence.cadence_seconds | 600 (every 10 minutes) |
coherence.fire_on.pairwise_governance_floor_below | 0.5 |
coherence.fire_on.conflict_edge_count_exceeds | 3 |
coherence.fire_on.outlier_agents_count_exceeds | 0 (any outlier fires) |
coherence.severity_on_fire | medium |
fault_line.enabled | true |
fault_line.cadence_seconds | 600 |
fault_line.severity_floor | high |
fault_line.use_reputation_scores | true |
fault_line.severity_on_fire | high |
fleet.enabled | true |
fleet.cadence_seconds | 600 |
fleet.patterns.outliers | true |
fleet.patterns.min_pair_score_below | 0.5 |
fleet.patterns.cluster_partition | true |
fleet.severity_on_fire | medium |
Rationale
Standard is calibrated for “fleets that should know when something’s amiss but aren’t operating at compliance-critical strictness.” Coherence below half-strength fires; conflict-edge count above 3 fires; any outlier fires; fault-lines at severity high+ surface; cluster partitions surface. Cadence is mid-paced (10 minutes is enough to catch shifts without burning observability budget on tight loops).
If you don’t know which posture to pick, Standard is the right answer.
High-Compliance
Posture ID: tp-platform-high-compliance · When to pick: Banking-core, healthcare, regulated industries, or any fleet where missed signals are more costly than signal noise.
Detection coverage
| Axis | Setting | Δ vs. Standard |
|---|
coherence.cadence_seconds | 300 | 2× faster |
coherence.fire_on.pairwise_governance_floor_below | 0.7 | Tighter — fires earlier |
coherence.fire_on.conflict_edge_count_exceeds | 1 | Any conflict edge fires |
coherence.severity_on_fire | high | Louder advisory |
fault_line.cadence_seconds | 300 | 2× faster |
fault_line.severity_floor | medium | More findings actionable |
fault_line.use_reputation_scores | true | (same — already on in Standard) |
fault_line.severity_on_fire | critical | Loudest |
fleet.cadence_seconds | 300 | 2× faster |
fleet.patterns.min_pair_score_below | 0.7 | Tighter |
fleet.severity_on_fire | high | Louder |
Rationale
High-Compliance assumes that false negatives are unacceptable. Tighter thresholds mean more advisories — that’s the trade. Cadence is faster (5 minutes) so anomalies surface within minutes, not tens of minutes. Severity floors are lower (more findings cross the bar). Severity-on-fire is one notch up across all axes — when something fires, it’s loud.
If your fleet is subject to SOC 2, HIPAA, EU AI Act, banking regulations, or any framework that mandates documented detection coverage, start with High-Compliance (then clone and tune to your specific requirements).
Low-Latency
Posture ID: tp-platform-low-latency · When to pick: Real-time agent UX surfaces where observability overhead competes with response latency.
Detection coverage
| Axis | Setting | Δ vs. Standard |
|---|
coherence.cadence_seconds | 1800 | 3× slower |
coherence.fire_on.pairwise_governance_floor_below | 0.4 | Looser — fires later |
coherence.fire_on.conflict_edge_count_exceeds | 5 | More edges before firing |
coherence.fire_on.outlier_agents_count_exceeds | 1 | Two-or-more outliers before firing |
coherence.severity_on_fire | low | Quieter |
fault_line.cadence_seconds | 1800 | 3× slower |
fault_line.severity_floor | critical | Only critical findings surface |
fault_line.use_reputation_scores | false | Reputation modulation off |
fault_line.severity_on_fire | medium | Quieter |
fleet.cadence_seconds | 1800 | 3× slower |
fleet.patterns.min_pair_score_below | 0.3 | Looser |
fleet.patterns.cluster_partition | false | Pattern off |
fleet.severity_on_fire | low | Quieter |
Rationale
Low-Latency assumes that false positives compound a UX cost more than false negatives compound a risk cost. Slower cadence (30 minutes), permissive thresholds, lower severity-on-fire, and cluster-partition detection turned off — together, this minimizes advisory pressure on agents that are operating in latency-sensitive modes.
This posture is not suitable for compliance-driven fleets. It’s suitable for: real-time chat agents where every advisory injection adds latency to the user, gaming or live-event agents where missed signals are recoverable, low-stakes internal tools.
Choosing — a decision matrix
| If your fleet… | Pick |
|---|
| Is general-purpose, mid-stakes, or you’re unsure | Standard |
| Is regulated (SOC 2, HIPAA, banking, EU AI Act, etc.) | High-Compliance (then customize) |
| Operates in latency-sensitive surfaces (real-time chat, live events) | Low-Latency (with caveats — review before adopting) |
| Has multiple sub-populations with different needs | Mix: assign different postures to different teams |
Customization workflow
The Mnemom defaults are immutable post-seed. You can’t edit them; you clone them.
# Clone Standard into your org as the customization starting point
mnemom posture clone tp-platform-standard --org org-acme --slug acme-standard --name "Acme Standard"
# Edit the clone (forward-only revision)
mnemom posture update tp-acme... --from acme-standard-v2.json --summary "tighten coherence to 0.6"
# Assign the clone to a team
mnemom posture assign tp-acme... --team <team-uuid>
Tuning over time
Posture revisions are forward-only. Every edit creates a new revision; old revisions stay queryable. Rollback creates a forward revision pointing at an older body. This means:
- You can A/B-test posture changes on individual teams without affecting the org default.
- You can answer “what was our coherence threshold for the trading-desk team on 2026-03-31?” with one query.
- You can revert a misconfigured tightening without losing audit linearity.
See Posture versioning for the full revision/rollback semantics.
See also
