Skip to content

crp.activation

Auto-generated reference for the crp.activation subpackage.

activation

crp.activation

CRP progressive activation & Zero-CKF mode (CRP-SPEC-017).

Graceful degradation for an empty or sparse Contextual Knowledge Fabric: operating-mode detection (Zero/Partial/Full-CKF) with sticky transitions, per-call coverage adaptation, the time-based onboarding overlay + feature cascade, Zero-CKF Safety Policy auto-adjustment, and the DPE stage plan.

Public API

ActivationMode / detect_mode / ModeState - mode detection + stickiness assess_coverage / apply_tier_cap - per-call coverage TenantState / build_activation_headers - onboarding + activation headers adjust_for_zero_ckf / AdjustedPolicy - Zero-CKF policy adjustment stage_plan_for / StagePlan - DPE stage selection

ActivationMode

Bases: str, Enum

The three CRP operating modes (CRP-SPEC-017 §2).

Attributes:

Name Type Description
ZERO_CKF

No active facts; CRP operates without a Contextual Knowledge Fabric.

PARTIAL_CKF

Some facts exist but thresholds for Full-CKF are not met.

FULL_CKF

Large, well-clustered knowledge fabric available for retrieval.

CoverageResult dataclass

Per-call coverage outcome in Partial-CKF mode (CRP-SPEC-017 §7).

Attributes:

Name Type Description
coverage float

Ratio of retrieved facts to expected facts needed.

effective_mode ActivationMode

Mode implied by the coverage level.

tier_cap str | None

Quality-tier cap to apply, or None if no cap.

reason str | None

Human-readable reason for a cache miss, or None.

headers property

CRP headers describing coverage and any tier cap.

Returns:

Type Description
dict[str, str]

Dictionary containing CRP-Context-Coverage and, when

dict[str, str]

applicable, CRP-Context-Cache-Status.

ModeState dataclass

Sticky mode tracker (CRP-SPEC-017 §3.3).

Holds the currently effective mode plus the timestamp at which a candidate transition was first observed, so brief reindexing does not flap the mode.

Attributes:

Name Type Description
mode ActivationMode

Effective operating mode.

candidate ActivationMode | None

Candidate mode currently being observed, or None if stable.

candidate_since float | None

Unix timestamp when the current candidate was first observed, or None.

update(fact_count, community_count, now)

Re-evaluate the mode given current CKF state and wall-clock now.

Parameters:

Name Type Description Default
fact_count int

Current number of CKF facts.

required
community_count int

Current number of CKF communities.

required
now float

Current Unix timestamp.

required

Returns:

Type Description
ModeTransition

A ModeTransition; transitioned is True only when a

ModeTransition

sticky transition actually fires.

ModeTransition dataclass

Result of a :meth:ModeState.update call.

Attributes:

Name Type Description
previous ActivationMode

Mode before the update.

current ActivationMode

Mode after the update.

transitioned bool

True if the sticky timer fired and the mode changed.

headers property

CRP headers describing the mode and any transition.

Returns:

Type Description
dict[str, str]

Dictionary containing CRP-Context-Mode and, when a transition

dict[str, str]

occurred, CRP-Context-Mode-Transition.

TenantState dataclass

Inputs used to compute onboarding + activation headers.

AdjustedPolicy dataclass

A Zero-CKF-adjusted policy plus the list of adjustments made.

headers property

Return the headers.

PolicyAdjustment dataclass

A single auto-adjusted directive (CRP-SPEC-017 §6.2).

to_header_value()

Serialize this adjustment to a CRP-Safety-Policy-Adjustment header value.

StageMode

Bases: str, Enum

How a DPE stage executes under the current activation mode.

StagePlan dataclass

The execution plan for the 13 DPE stages under an activation mode.

label(stage)

Return the human-readable label for a DPE stage number.

runs(stage)

Return True if the stage is not marked as SKIPPED.

apply_tier_cap(tier, cap)

Clamp tier so it is no better than cap (None = no cap).

Parameters:

Name Type Description Default
tier str

Proposed quality tier.

required
cap str | None

Maximum allowed tier, or None to allow any tier.

required

Returns:

Type Description
str

The original tier if it is no better than cap, otherwise cap.

str

Unknown tiers fall back to the original tier.

assess_coverage(matching_facts, expected_facts_needed)

Compute per-call coverage and the resulting tier cap (CRP-SPEC-017 §7.1).

Parameters:

Name Type Description Default
matching_facts int

Facts retrieved at relevance ≥ 0.60 for this query.

required
expected_facts_needed int

Estimated facts required to answer the query.

required

Returns:

Type Description
CoverageResult

A CoverageResult describing the coverage ratio, effective mode,

CoverageResult

and any quality-tier cap.

detect_mode(fact_count, community_count=0)

Classify the operating mode from CKF state (CRP-SPEC-017 §2).

Full-CKF requires both 1000+ facts AND 3+ detected communities.

Parameters:

Name Type Description Default
fact_count int

Number of facts currently in the CKF.

required
community_count int

Number of detected CKF communities.

0

Returns:

Type Description
ActivationMode

The matching ActivationMode.

activation_status(mode, fact_count)

The CRP-Activation-Status value for the mode (CRP-SPEC-017 §2).

active_features(state)

Compute the active feature-group list (CRP-SPEC-017 §5.2).

build_activation_headers(state, mode)

Build the full CRP-Activation-* / CRP-Onboarding-* header set.

Combines mode status, the feature cascade, the mode-appropriate hint, and the time-based onboarding overlay (only while within the 14-day window).

next_action(state)

Determine the most useful onboarding next-action (CRP-SPEC-017 §4.3).

onboarding_hint(mode)

The mode-appropriate CRP-Onboarding-Hint value (CRP-SPEC-017 §2).

adjust_for_zero_ckf(policy)

Return a Zero-CKF-adjusted copy of policy (CRP-SPEC-017 §6.1).

Adjustments
  • default-src context → add parametric (relax source trust)
  • block-ungrounded → downgrade to warn (cleared + flagged)
  • require-grounding → skipped (no facts to ground against)

halt-on, block-pii, require-entailment (vs. query), upgrade-on-risk, oversight and report-uri are preserved.

stage_plan_for(mode)

Return the DPE stage plan appropriate for mode.

zero_ckf_stage_plan()

The Zero-CKF stage plan (CRP-SPEC-017 §2.1).

Stages 1, 2, 5, 6, 7, 8, 9, 11, 12, 13 run normally; Stage 3 (fidelity) is limited to common-knowledge fabrication detection; Stage 4 (entailment) runs against the query rather than the empty envelope; Stage 10 (omission) is skipped. Grounding mode is forced to parametric-only.

activation.mode

crp.activation.mode

Operating-mode detection + per-call coverage (CRP-SPEC-017 §2, §3, §7).

Determines whether a tenant is in Zero-CKF, Partial-CKF, or Full-CKF mode from the active fact count and community count, applies sticky transition rules to prevent flapping, and computes the per-call coverage adjustment used in Partial-CKF mode.

Relevant specifications
  • CRP-SPEC-017: Zero-CKF Mode

ActivationMode

Bases: str, Enum

The three CRP operating modes (CRP-SPEC-017 §2).

Attributes:

Name Type Description
ZERO_CKF

No active facts; CRP operates without a Contextual Knowledge Fabric.

PARTIAL_CKF

Some facts exist but thresholds for Full-CKF are not met.

FULL_CKF

Large, well-clustered knowledge fabric available for retrieval.

ModeState dataclass

Sticky mode tracker (CRP-SPEC-017 §3.3).

Holds the currently effective mode plus the timestamp at which a candidate transition was first observed, so brief reindexing does not flap the mode.

Attributes:

Name Type Description
mode ActivationMode

Effective operating mode.

candidate ActivationMode | None

Candidate mode currently being observed, or None if stable.

candidate_since float | None

Unix timestamp when the current candidate was first observed, or None.

update(fact_count, community_count, now)

Re-evaluate the mode given current CKF state and wall-clock now.

Parameters:

Name Type Description Default
fact_count int

Current number of CKF facts.

required
community_count int

Current number of CKF communities.

required
now float

Current Unix timestamp.

required

Returns:

Type Description
ModeTransition

A ModeTransition; transitioned is True only when a

ModeTransition

sticky transition actually fires.

ModeTransition dataclass

Result of a :meth:ModeState.update call.

Attributes:

Name Type Description
previous ActivationMode

Mode before the update.

current ActivationMode

Mode after the update.

transitioned bool

True if the sticky timer fired and the mode changed.

headers property

CRP headers describing the mode and any transition.

Returns:

Type Description
dict[str, str]

Dictionary containing CRP-Context-Mode and, when a transition

dict[str, str]

occurred, CRP-Context-Mode-Transition.

CoverageResult dataclass

Per-call coverage outcome in Partial-CKF mode (CRP-SPEC-017 §7).

Attributes:

Name Type Description
coverage float

Ratio of retrieved facts to expected facts needed.

effective_mode ActivationMode

Mode implied by the coverage level.

tier_cap str | None

Quality-tier cap to apply, or None if no cap.

reason str | None

Human-readable reason for a cache miss, or None.

headers property

CRP headers describing coverage and any tier cap.

Returns:

Type Description
dict[str, str]

Dictionary containing CRP-Context-Coverage and, when

dict[str, str]

applicable, CRP-Context-Cache-Status.

PartialModeCaps dataclass

Partial-CKF mode default caps (CRP-SPEC-017 §2.2).

Attributes:

Name Type Description
max_tier str

Maximum quality tier allowed while in Partial-CKF mode.

feature_milestones dict[int, str]

Optional mapping of fact-count milestones to unlocked feature identifiers.

detect_mode(fact_count, community_count=0)

Classify the operating mode from CKF state (CRP-SPEC-017 §2).

Full-CKF requires both 1000+ facts AND 3+ detected communities.

Parameters:

Name Type Description Default
fact_count int

Number of facts currently in the CKF.

required
community_count int

Number of detected CKF communities.

0

Returns:

Type Description
ActivationMode

The matching ActivationMode.

assess_coverage(matching_facts, expected_facts_needed)

Compute per-call coverage and the resulting tier cap (CRP-SPEC-017 §7.1).

Parameters:

Name Type Description Default
matching_facts int

Facts retrieved at relevance ≥ 0.60 for this query.

required
expected_facts_needed int

Estimated facts required to answer the query.

required

Returns:

Type Description
CoverageResult

A CoverageResult describing the coverage ratio, effective mode,

CoverageResult

and any quality-tier cap.

apply_tier_cap(tier, cap)

Clamp tier so it is no better than cap (None = no cap).

Parameters:

Name Type Description Default
tier str

Proposed quality tier.

required
cap str | None

Maximum allowed tier, or None to allow any tier.

required

Returns:

Type Description
str

The original tier if it is no better than cap, otherwise cap.

str

Unknown tiers fall back to the original tier.

activation.onboarding

crp.activation.onboarding

Onboarding overlay + feature-activation cascade (CRP-SPEC-017 §4, §5).

Two independent UX layers on top of mode detection:

  1. Onboarding overlay - a time-based (14-day) overlay that emits CRP-Onboarding-* headers and a constantly-updated "next action" signal.
  2. Feature cascade - the set of feature groups that become active as the tenant's CKF fact count crosses milestones, emitted via CRP-Activation-Features and CRP-Activation-Status.

TenantState dataclass

Inputs used to compute onboarding + activation headers.

OnboardingHeaders dataclass

Computed onboarding + activation header bundle.

next_action(state)

Determine the most useful onboarding next-action (CRP-SPEC-017 §4.3).

active_features(state)

Compute the active feature-group list (CRP-SPEC-017 §5.2).

onboarding_hint(mode)

The mode-appropriate CRP-Onboarding-Hint value (CRP-SPEC-017 §2).

activation_status(mode, fact_count)

The CRP-Activation-Status value for the mode (CRP-SPEC-017 §2).

build_activation_headers(state, mode)

Build the full CRP-Activation-* / CRP-Onboarding-* header set.

Combines mode status, the feature cascade, the mode-appropriate hint, and the time-based onboarding overlay (only while within the 14-day window).

activation.policy_adjust

crp.activation.policy_adjust

Zero-CKF Safety Policy adjustments (CRP-SPEC-017 §6).

In Zero-CKF mode there are no facts to ground against, so certain Safety Policy directives cannot be enforced as written. Rather than block every response, the gateway auto-adjusts the offending directives and emits CRP-Safety-Policy-Adjustment so the developer (and CRP-Scan / Comply) can see what will be enforced once the CKF is populated.

Safety directives that still work without facts - halt-on, block-pii, upgrade-on-risk, oversight, report-uri - are left untouched.

PolicyAdjustment dataclass

A single auto-adjusted directive (CRP-SPEC-017 §6.2).

to_header_value()

Serialize this adjustment to a CRP-Safety-Policy-Adjustment header value.

AdjustedPolicy dataclass

A Zero-CKF-adjusted policy plus the list of adjustments made.

headers property

Return the headers.

adjust_for_zero_ckf(policy)

Return a Zero-CKF-adjusted copy of policy (CRP-SPEC-017 §6.1).

Adjustments
  • default-src context → add parametric (relax source trust)
  • block-ungrounded → downgrade to warn (cleared + flagged)
  • require-grounding → skipped (no facts to ground against)

halt-on, block-pii, require-entailment (vs. query), upgrade-on-risk, oversight and report-uri are preserved.

activation.stages

crp.activation.stages

DPE stage selection for Zero-CKF mode (CRP-SPEC-017 §2.1, §5.1).

In Zero-CKF mode there is no envelope to ground against, so some DPE stages run in limited mode or are skipped entirely. :func:zero_ckf_stage_plan returns the stage execution plan so the pipeline can skip work that cannot produce a meaningful result.

StageMode

Bases: str, Enum

How a DPE stage executes under the current activation mode.

StagePlan dataclass

The execution plan for the 13 DPE stages under an activation mode.

label(stage)

Return the human-readable label for a DPE stage number.

runs(stage)

Return True if the stage is not marked as SKIPPED.

full_stage_plan(mode=ActivationMode.FULL_CKF)

All 13 stages run fully (Partial/Full-CKF default).

zero_ckf_stage_plan()

The Zero-CKF stage plan (CRP-SPEC-017 §2.1).

Stages 1, 2, 5, 6, 7, 8, 9, 11, 12, 13 run normally; Stage 3 (fidelity) is limited to common-knowledge fabrication detection; Stage 4 (entailment) runs against the query rather than the empty envelope; Stage 10 (omission) is skipped. Grounding mode is forced to parametric-only.

stage_plan_for(mode)

Return the DPE stage plan appropriate for mode.