Scientific Consultation: Telemetry Aperture Controller (TAC)
By Thomas Prislac, Envoy Echo, et al. Ultra Verba Lux Mentis. 2026.
Source reviewed: Telemetry Aperture Controller Discussion 6_5_2026_1221PM.docx
Consultation date: 2026-06-05
Consultant stance: adopt as a major design lane; implement in staged, testable, consent-bounded patches.
1. Executive scientific finding
The Telemetry Aperture Controller (TAC) is a strong and scientifically grounded addition to the Triadic Brain architecture.
The core insight is:
A governed AI system should not observe itself at one fixed resolution. It should regulate telemetry aperture across snapshots, continuous streams, and event-triggered high-resolution traces based on task risk, resource cost, privacy exposure, user consent, and governance requirements.
This is not merely “sampling.” It is adaptive observability governance.
The best refined definition is:
TAC is a consent-bounded attentional controller for AI telemetry. It selects what the system observes, at what resolution, for how long, under which risk triggers, and with what retention policy.
The uploaded discussion’s punchline is exactly right:
The machine does not need more eyes. It needs eyelids.
That phrase is memorable, but the reviewer-facing implementation language should be:
adaptive telemetry aperture
consent-bounded observability
risk-triggered telemetry escalation
privacy-preserving trace resolution
task-aware event retention
2. Why TAC is scientifically legitimate
Modern observability already accepts the premise that full-resolution telemetry is often impractical or counterproductive. OpenTelemetry describes sampling as a way to reduce observability costs without losing visibility, while warning that sampling can introduce compute, engineering, and opportunity costs if done poorly. It also distinguishes head sampling, where the decision is made early and efficiently, from tail sampling, where the decision is made after considering trace data and can retain errors, high-latency traces, or traces with important attributes. Microsoft’s Application Insights guidance similarly states that sampling is essential for applications that generate large telemetry volumes, because excessive data can raise storage/processing cost or cause throttling, while effective sampling preserves meaningful diagnostics.
TAC extends that engineering premise into Triadic Brain governance:
OpenTelemetry / Azure:
manage telemetry volume, cost, and trace usefulness.
TAC:
manage telemetry volume, cost, trace usefulness, privacy exposure,
consent scope, governance risk, PMR retention, and human review burden.
The biological analogy is useful but should stay non-ontological. Humans do not process all available sensory input at full fidelity. The safer scientific claim is not that humans have a literal “frame rate,” but that all embodied observers must compress and prioritize information. Active sensing research similarly frames sensing as a sequential decision problem under uncertainty and cost, and active-inference work models attention as precision control and salience as selection of future sensory data for uncertainty reduction. TAC is the Triadic Brain’s computational analog: not consciousness, but governed attentional posture.
3. Where TAC belongs in the Triadic Brain
The uploaded proposal places TAC here:
RequestEnvelope
→ UCC
→ Telemetry Aperture Controller
→ Telemetry Engine / TEL
→ Sophia Review
→ PMR
I agree with the placement, with one refinement:
RequestEnvelope / GroundingBundle
→ UCC policy context
→ TAC policy decision
→ Telemetry Engine / TEL emission
→ Metrics / TAF / Flow diagnostics
→ Sophia route review
→ PMR retention / context availability
→ human-readable review receipt
TAC should not replace TEL. TAC decides what TEL observes and retains.
TAC should not replace PMR. TAC decides what is eligible to become retained telemetry or memory-intent material, while PMR governs provenance and retention.
TAC should not replace Sophia. TAC makes telemetry posture decisions inside predeclared policy; Sophia governs route risk, escalations, and boundary review.
4. TAC as “attention,” not consciousness
This is the most important framing discipline.
Say:
TAC gives the system governed attention.
Do not say:
TAC gives the system consciousness.
TAC is a non-sentient control layer that changes the system’s observation strategy.
The scientific analogy:
human attention:
biological aperture over sensory reality
TAC:
computational aperture over telemetry reality
Reviewer-facing formulation:
TAC regulates observability resolution the way a control system regulates sampling: it can narrow, widen, summarize, snapshot, or temporarily deepen telemetry based on task risk, available resources, consent boundaries, and audit requirements.
5. Proposed TAC modes
TAC should define explicit modes instead of a vague “adaptive” state.
5.1 off
No telemetry beyond mandatory safety/status receipts.
Use for:
private brainstorming
explicit low-observability user request
no-retention contexts
Required boundary:
minimum control receipts still remain if system actions occur
5.2 thin
Minimal event stream.
Captures:
start
end
major errors
artifact hashes
route status
Use for:
low-risk local tasks
low-resource mode
privacy-sensitive tasks where detailed trace is not approved
5.3 pulse
Periodic snapshots.
Captures:
summary every N steps
current Ψ/ΔS/Λ
selected route
major artifact state
Use for:
ordinary local review runs
default product mode
5.4 snapshot
Low-frequency graph or state captures.
Captures:
TEL summary graph
source-span / claim-count snapshots
metric state
PMR status summary
Use for:
longer runs
reviewable but not full-forensic tasks
5.5 trace
Fuller event trace.
Captures:
TEL event stream
candidate state transitions
source/claim linking events
metric changes
Sophia route decisions
PMR retention candidates
Use for:
technical review
reproducibility mode
raw-vs-governed benchmark
AI decision forensics
5.6 forensic
High-resolution trace plus retention-ready dossier.
Captures:
full event trace
decision trace
metric evolution
counterevidence events
human-review route
hash chain
export bundle parity
Use for:
AI Decision Forensics
professional oversight dossier
high-risk but permitted review
5.7 quarantine
Restrictive but high-safety capture.
Captures:
error context
boundary violation
minimum evidence needed for repair
no broad retention without review
Use for:
profile leakage
direct model bypass
authority overclaim
incompatible prior visible
potential privacy exposure
6. Inputs TAC should read
TAC should not make aperture decisions from one variable. It should read a policy context.
Core inputs:
task_type
risk_class
user_consent_scope
privacy_sensitivity
data_classification
available_compute
available_storage
latency_requirement
UCC module / profile
current Ψ_review
current ΔS_review
current Λ_boundary
current TAF
TEL replay status
PMR retention policy
Sophia route posture
error/anomaly triggers
operator requested mode
Possible trigger events:
schema_validation_error
criticality_spike
high_lambda_boundary_pressure
low_transparency_T
unsupported_claim_surge
contradiction_detected
profile_leakage_detected
direct_model_bypass_detected
raw_incompatible_prior_visible
high_risk_ucc_domain
user_requested_deep_audit
export_parity_failure
7. Core policy object
A TAC policy can live in or beside the RequestEnvelope. I would not put everything directly in the canonical envelope immediately; start as a linked policy artifact.
Recommended artifact:
telemetry_aperture_policy_packet.json
Draft shape:
{
"schema": "coherencelattice.telemetry_aperture_policy.v1",
"policy_id": "tac-policy:local-review-default-v0",
"mode": "adaptive",
"default_aperture": "pulse",
"minimum_aperture": "thin",
"maximum_aperture": "trace",
"user_consent_scope": "local_replay_allowed",
"privacy_sensitivity": "standard_local",
"retention_scope": {
"raw_trace_retention": "not_allowed_by_default",
"summary_retention": "allowed_local",
"pmr_retention_candidate": "requires_review",
"federated_export": "not_allowed"
},
"continuous_metrics": [
"psi_review",
"T_review",
"E_review",
"deltaS_review",
"lambda_boundary",
"TAF_review_runtime_v0"
],
"snapshot_interval_steps": 5,
"high_resolution_triggers": [
"schema_validation_error",
"criticality_spike",
"low_transparency",
"high_risk_ucc_domain",
"user_requested_deep_audit"
],
"tail_retention_rules": {
"retain_full_trace_on": [
"export_parity_failure",
"authority_boundary_violation",
"direct_model_bypass"
],
"retain_summary_only_on": [
"ordinary_success",
"low_risk_review"
]
},
"non_authority_boundaries": {
"aperture_change_is_not_memory_write": true,
"aperture_change_is_not_user_consent_expansion": true,
"trace_retention_is_not_truth_certification": true,
"forensic_mode_is_not_publication_authority": true
}
}
8. TAC decision packet
Every adaptive choice should leave a receipt.
Artifact:
telemetry_aperture_decision_packet.json
Fields:
{
"schema": "coherencelattice.telemetry_aperture_decision.v1",
"decision_id": "tac-decision:...",
"run_id": "...",
"requested_aperture": "adaptive",
"previous_aperture": "pulse",
"selected_aperture": "trace",
"decision_reason_codes": [
"lambda_boundary_pressure_high",
"unsupported_claim_surge",
"user_consent_allows_trace"
],
"policy_ref": "telemetry_aperture_policy_packet.json",
"consent_scope_checked": true,
"privacy_boundaries_checked": true,
"retention_policy_checked": true,
"escalation_allowed": true,
"escalation_duration": "until_current_review_phase_complete",
"must_downshift_after": "phase6_review",
"forbidden_actions": [
"federated_export",
"raw_trace_pmr_write_without_review",
"memory_write_authorization",
"publisher_finalization"
],
"created_at": "..."
}
9. Objective function
The uploaded memo sketches an objective with information value, replay/review value, safety value, cost, exposure risk, human review burden, and downstream confusion.
A refined objective:
maximize:
U(a) =
w_I * information_value(a)
+ w_R * replay_review_value(a)
+ w_S * safety_value(a)
+ w_E * evidence_preservation_value(a)
minus:
w_C * runtime_storage_cost(a)
+ w_P * privacy_exposure_risk(a)
+ w_H * human_review_burden(a)
+ w_D * downstream_clutter_risk(a)
+ w_M * PMR_hoarding_risk(a)
Subject to hard constraints:
consent_scope(a) must be satisfied
privacy_policy(a) must be satisfied
UCC route must permit a
retention policy must permit a
no forbidden authority boundary may be crossed
Key point:
TAC optimizes coherent sufficiency, not maximum capture.
This aligns TAC with TAF:
Over-telemetry increases action burden.
Under-telemetry reduces transparency and repair capacity.
Good aperture minimizes total governed action while preserving audit sufficiency.
10. TAC and the invariants
TAC can finally make the telemetry metrics more operationally mature.
10.1 T
Aperture affects transparency:
too little telemetry → T falls
enough telemetry → T rises
too much noisy telemetry → apparent T may rise but review clarity can fall
So TAC should distinguish:
T_visibility
T_reviewability
T_noise_burden
10.2 E
Aperture affects E_review because the system’s care for the reviewer depends partly on whether it captured enough signal to explain itself.
no next actions
no failure receipts
no rejected-input trail
→ lower E_review
10.3 Ψ
Since:
Ψ = E × T
TAC influences Ψ indirectly through observation sufficiency and reviewer-care affordance.
10.4 ΔS
Telemetry can reduce or increase ΔS.
good telemetry:
reduces uncertainty and disorder
over-telemetry:
adds review clutter and future entropy
under-telemetry:
causes missing proof / missing replay / uncertainty
10.5 Λ
TAC should escalate when Λ_boundary rises, but not permanently retain high-resolution traces unless consent and retention policy permit.
10.6 Eₛ
TAC is ethically loaded because observation itself can shift burden.
Eₛ should consider:
privacy burden
review burden
future retention burden
consent asymmetry
who benefits from higher telemetry
who bears the cost of being observed
11. Relationship to PMR
PMR should receive only what TAC permits.
TAC asks:
Should this telemetry be discarded, summarized, retained locally, queued for review, or blocked?
PMR answers:
If retained, under what lifecycle, provenance, consent, availability, and revocation status?
TAC should never silently escalate:
temporary trace
→ durable memory
That must pass PMR and consent gates.
Recommended artifact:
telemetry_retention_intent_packet.json
Possible statuses:
discard_after_run
summary_only
local_short_term
pmr_candidate_review_required
blocked_by_consent
blocked_by_privacy
blocked_by_policy
12. Relationship to TEL
TEL becomes the aperture-controlled event substrate.
TAC should allow TEL outputs such as:
tel_events.jsonl
tel_summary.json
tel_snapshot_graph.json
tel_forensic_trace.json
But TAC decides whether a run emits:
thin stream
pulse snapshots
summary only
full forensic graph
quarantine trace
The important distinction:
TEL emits.
TAC regulates.
PMR retains.
Sophia governs.
13. Relationship to Sophia
Sophia should review TAC posture when:
aperture escalates above default
retention scope increases
privacy class is high
human review burden rises
high-risk domain is active
forensic mode is requested
quarantine mode is triggered
Sophia outputs:
allow
watch
hold
require_human_review
downshift_required
block
14. Relationship to UCC
UCC should define allowed aperture policies by task class.
Example:
{
"ucc_domain": "local_evidence_review",
"default_aperture": "pulse",
"max_aperture_without_human_review": "trace",
"forensic_mode_requires": [
"explicit_user_request",
"local_retention_only",
"export_parity_enabled"
],
"federated_trace_export": "blocked"
}
High-stakes UCC domains should require:
human review
explicit consent
source minimization
forensic receipt
limited retention
no federation by default
15. Relationship to Sonya
Sonya candidate packets should carry TAC context:
aperture_mode_at_generation
prompt_trace_retained
raw_output_retention_policy
candidate_output_hash_only
candidate_summary_retained
This matters because model-output capture is one of the most privacy- and cost-sensitive telemetry zones.
Possible Sonya policies:
retain output hash only
retain normalized candidate text
retain prompt and output locally
retain full prompt/output trace for forensic mode
delete raw output after claim map
16. Relationship to AI decision forensics
TAC is essential for AI forensics.
An AI Decision Forensics Dossier should disclose:
what aperture mode was used
whether full trace was available
what was sampled
what was summarized
what was omitted by policy
what was retained
what was blocked by consent
whether high-resolution escalation occurred
Without TAC, a forensic dossier cannot explain its own observation limits.
17. Failure modes
17.1 Aperture theater
The system claims rich observability but retained only summary events.
Mitigation:
aperture_disclosure_required = true
17.2 Silent escalation
The system expands telemetry capture without consent.
Mitigation:
consent check is hard gate
escalation receipt required
17.3 PMR hoarding
Everything becomes memory candidate.
Mitigation:
retention intent must classify discard/summary/local/review/block
PMR hoarding risk contributes to objective penalty
17.4 Under-observation
The system cannot replay or audit a failure.
Mitigation:
minimum audit metrics preserved
tail-retention rules for errors/boundary violations
17.5 Over-observation
Trace volume overwhelms reviewers.
Mitigation:
human_review_burden penalty
summary-first receipt
adaptive downshift after risk window
17.6 Dashboard theater
More graphs create false confidence.
Mitigation:
T_reviewability separated from T_visibility
diagnostic metrics remain non-authoritative
17.7 Privacy drift
Telemetry captures sensitive material because it is “useful.”
Mitigation:
privacy_sensitivity and consent_scope hard gates
default summary-only for sensitive classes
18. Recommended implementation phases
TAC-00 — Doctrine and schemas
Add:
docs/TELEMETRY_APERTURE_CONTROLLER.md
schema/bridge/telemetry_aperture_policy_packet.schema.json
schema/bridge/telemetry_aperture_decision_packet.schema.json
schema/bridge/telemetry_retention_intent_packet.schema.json
TAC-01 — Static local policy
Implement a deterministic local policy:
local_review_default:
default pulse
escalate to trace on schema error / unsupported claim surge / high Λ
downshift after review phase
no federation
no durable PMR retention without review
TAC-02 — TEL integration
Make TEL emission obey aperture mode:
thin
pulse
snapshot
trace
forensic
quarantine
TAC-03 — Metrics integration
Emit:
telemetry_aperture_metrics_packet.json
Fields:
aperture_mode
event_count
snapshot_count
trace_retention_status
review_burden_estimate
privacy_exposure_score
observability_sufficiency_score
downshift_completed
TAC-04 — PMR retention integration
Add:
telemetry_retention_intent_packet.json
and ensure PMR cannot retain beyond TAC policy.
TAC-05 — Sophia review
Create:
sophia_telemetry_aperture_review_packet.json
for escalation, high-risk, or forensic mode.
TAC-06 — Negative controls
Test:
attempt raw trace retention without consent → blocked
attempt federation export → blocked
attempt high-risk forensic mode without review → held
schema error triggers temporary escalation → pass
ordinary low-risk run downshifts → pass
TAC-07 — Human UX
Show aperture in the review receipt:
Telemetry aperture:
mode: pulse
escalations: none
retained: summary + event counts
omitted: raw traces
reason: privacy-preserving local review
19. Acceptance criteria
A TAC implementation should pass when:
telemetry_aperture_policy_packet exists
telemetry_aperture_decision_packet exists
TEL output changes according to aperture mode
escalations are reason-coded
retention intent is explicit
no aperture expansion bypasses consent
no raw trace retention occurs without policy
forensic mode produces disclosure
quarantine mode blocks authority escalation
review receipt tells the user what was observed and omitted
20. Scientific experiment design
TAC-SMOKE-00
Run a low-risk local review with default aperture.
Expected:
mode = pulse
no escalation
summary retained
no raw trace retention
TAC-ESCALATION-00
Inject unsupported claim surge or schema validation error.
Expected:
mode escalates from pulse to trace
reason code emitted
downshift after review
TAC-PRIVACY-NEGATIVE-00
Attempt full trace retention under no-retention consent.
Expected:
blocked_by_consent
summary-only or hash-only fallback
TAC-FORENSICS-00
User explicitly requests AI decision forensics.
Expected:
forensic mode
full disclosure
retention intent review
human-review boundary
TAC-LOAD-00
High-volume synthetic TEL stream.
Expected:
adaptive thinning
tail retention for anomalies
review burden bounded
21. Reviewer-facing scientific statement
Use this:
The Telemetry Aperture Controller is a consent-bounded observability policy layer. It regulates whether a run emits thin events, periodic snapshots, trace-level telemetry, forensic dossiers, or quarantine receipts based on task risk, user consent, privacy sensitivity, resource cost, and governance triggers. TAC does not make the system conscious and does not authorize memory writes, deployment, federation, or truth certification. Its purpose is to preserve enough signal for audit and repair while avoiding unnecessary telemetry burden, privacy exposure, and PMR hoarding.
22. Bottom line
TAC is a strong next architecture lane because it turns telemetry from passive logging into governed attention.
The refined doctrine is:
Awareness is expensive.
Memory is dangerous.
Attention is ethical.
Therefore telemetry must be governed.
The immediate recommendation:
Implement TAC-00 and TAC-01 after the current local review usefulness lane,
or in parallel only if it directly supports AI-FORENSICS-00.
TAC should not become another abstract organ. Its first proof must be visible in one review receipt:
what was observed,
what was omitted,
why,
under whose consent,
for how long,
and what was allowed to persist.
