Movistar Service Intelligence

SYNTHETIC DEMO 2401 homes, 41 access nodes, 6 regions, 576 intervals at 5 min cadence ◂ Operator view

A network node degrading (gradual access fault) · OLT-2800001-01

Madrid metro · layer access · SHI: critical for the affected element

Access node OLT-2800001-01 (Madrid metro / Las Tablas) degrading: 20 homes with rising freeze/re-tune over the past ~30 min, signature local access cluster. Recommended: inspect the access node and its aggregation before complaints escalate.

Stream

2401 homes across 6 regions stream four-field telemetry; the engine learns normal and watches the graph for structure.

Form

A cluster forms on the access layer: cluster signature on OLT-2800001-01.

Predict

Projected to widen (convergence widening); on the current trend it would broaden around interval 90 if untreated.

Prescribe

inspect the access node and its aggregation before complaints escalate. Flagged 110 min ahead of the scenario's modeled surfacing interval (a synthetic assumption, not an observed complaint stream); see the certified-decision receipt.

Certified-decision receipt

Claim: Access node OLT-2800001-01 degrading: 20 homes behind it with rising impairment

Evidence: 20 homes elevated and sustained beyond the dwell threshold

Evidence: mean departure z=7.5 relative to a baseline learned over 84 healthy intervals (a dispersion ratio, not a calibrated tail probability)

Evidence: 20 of 20 elevated homes sit behind OLT-2800001-01 by the network map

Check: the same homes' core-route and content-source edges did not fire together, so the convergence is specific to this access node rather than a shared core or content fault

Provenance: derived from the four-field magnitude stream over intervals 81 to 87; four-field magnitude stream only; baselines learned from history; signature spans 20 homes (e.g., STB-2800001-01-005, STB-2800001-01-009, STB-2800001-01-012)

Corroborating context (post-hoc, not used for detection)

- access node port utilization 61%, FEC errors 92 (corroborating context)

- example home STB-2800001-01-005 freeze 2.5s this interval

Confidence in the layer attribution: 0.98 (finite-baseline coverage bound; synthetic, not a calibrated field probability)

Score: detected 1/1, lead 22 intervals, localization 1/1 (provisional, n=1), layer 1/1 (provisional, n=1), action 1/1 (provisional, n=1)

An individual household degrading (isolated, peers healthy) · STB-2800002-02-010

Madrid metro · layer home · SHI: degraded for the household

Household STB-2800002-02-010 (Madrid metro) degrading in isolation: rising freeze with weak home connectivity over the past ~30 min, peers on the access node healthy, signature isolated home gateway. Recommended: proactive customer contact and gateway reconfiguration, no truck roll.

Stream

2401 homes across 6 regions stream four-field telemetry; the engine learns normal and watches the graph for structure.

Form

A single home forms on the home layer: single signature on STB-2800002-02-010.

Predict

Projected to widen (convergence widening); on the current trend it would broaden around interval 223 if untreated.

Prescribe

proactive customer contact and gateway reconfiguration, no truck roll. Flagged 255 min ahead of the scenario's modeled surfacing interval (a synthetic assumption, not an observed complaint stream); see the certified-decision receipt.

Certified-decision receipt

Claim: Household STB-2800002-02-010 degrading in isolation; its access-node peers are healthy

Evidence: 1 homes elevated and sustained beyond the dwell threshold

Evidence: mean departure z=6.9 relative to a baseline learned over 216 healthy intervals (a dispersion ratio, not a calibrated tail probability)

Evidence: 1 of 1 elevated homes sit behind OLT-2800002-02 by the network map

Check: only this home is elevated while its node peers remain at baseline, inconsistent with a shared network fault

Provenance: derived from the four-field magnitude stream over intervals 214 to 220; four-field magnitude stream only; baselines learned from history; signature spans 1 homes (e.g., STB-2800002-02-010)

Corroborating context (post-hoc, not used for detection)

- home Wi-Fi SNR 23.3 dB, WAN packet loss 0.837% (corroborating context)

- example home STB-2800002-02-010 freeze 2.7s this interval

Confidence in the layer attribution: 0.95 (finite-baseline coverage bound; synthetic, not a calibrated field probability)

Score: detected 1/1, lead 51 intervals, localization 1/1 (provisional, n=1), layer 1/1 (provisional, n=1), action 1/1 (provisional, n=1)

An invisible fault that survives a set-top-box swap · OLT-1500001-01

A Coruna · layer access · SHI: critical for the affected element

Access node OLT-1500001-01 (A Coruna / A Coruna centro) degrading: 5 homes with rising freeze/re-tune over the past ~30 min, signature local access cluster. Recommended: inspect the access node and its aggregation before complaints escalate.

Stream

2401 homes across 6 regions stream four-field telemetry; the engine learns normal and watches the graph for structure.

Form

A cluster forms on the access layer: cluster signature on OLT-1500001-01.

Predict

Projected to widen (convergence widening); on the current trend it would broaden around interval 368 if untreated.

Prescribe

inspect the access node and its aggregation before complaints escalate. Flagged 75 min ahead of the scenario's modeled surfacing interval (a synthetic assumption, not an observed complaint stream); see the certified-decision receipt.

Certified-decision receipt

Claim: Access node OLT-1500001-01 degrading: 5 homes behind it with rising impairment

Evidence: 5 homes elevated and sustained beyond the dwell threshold

Evidence: mean departure z=8.8 relative to a baseline learned over 362 healthy intervals (a dispersion ratio, not a calibrated tail probability)

Evidence: 5 of 5 elevated homes sit behind OLT-1500001-01 by the network map

Check: the same homes' core-route and content-source edges did not fire together, so the convergence is specific to this access node rather than a shared core or content fault

Provenance: derived from the four-field magnitude stream over intervals 359 to 365; four-field magnitude stream only; baselines learned from history; signature spans 5 homes (e.g., STB-1500001-01-002, STB-1500001-01-005, STB-1500001-01-006)

Corroborating context (post-hoc, not used for detection)

- access node port utilization 62%, FEC errors 97 (corroborating context)

- example home STB-1500001-01-002 freeze 3.0s this interval

Confidence in the layer attribution: 1.00 (finite-baseline coverage bound; synthetic, not a calibrated field probability)

Score: detected 1/1, lead 15 intervals, localization 1/1 (provisional, n=1), layer 1/1 (provisional, n=1), action 1/1 (provisional, n=1), box-swap avoided by construction (designed demonstration, not a measured rate)

Run scorecard

Measured on this run · illustrative euro/impact figures excluded · no Movistar / Telefónica data

3/3

faults detected (provisional, small n)

0 of 173 no-fault intervals

false positives (single-run count)

by construction

box swap avoided (designed demo)

Honest instruments

Competence boundary: Competence boundary (ramp-edge, pre-detection deferral): a disturbance is forming at the ramp edge but the evidence will not yet resolve to a layer. A cluster may be forming on this access node, watch and confirm.

Decoys: 0 of 3 benign anomalies fired (a fire would be a false alarm)

False-positive rate: 0 of 173 no-fault intervals fired a false positive (single-run count, not a population rate; ~2% upper bound at 95% by the rule of three)

Scorecard

Structural Intelligence demo scorecard (measured values, no preset thresholds)
('surfacing' below is the scenario's modeled surfacing interval - a synthetic assumption, not an observed customer complaint stream)

uc1_network_node:
  detected: 1/1
  detection lead time: 22 intervals (110 min) before the modeled surfacing interval
  localization accuracy: 1/1 (provisional, n=1)
  layer attribution: 1/1 (provisional, n=1)
  action correctness: 1/1 (provisional, n=1)
  note: detected at interval 86, named OLT-2800001-01 (access/cluster), lead 22 intervals to the modeled surfacing interval at 108

uc2_individual:
  detected: 1/1
  detection lead time: 51 intervals (255 min) before the modeled surfacing interval
  localization accuracy: 1/1 (provisional, n=1)
  layer attribution: 1/1 (provisional, n=1)
  action correctness: 1/1 (provisional, n=1)
  note: detected at interval 219, named STB-2800002-02-010 (home/single), lead 51 intervals to the modeled surfacing interval at 270

uc3_invisible:
  detected: 1/1
  detection lead time: 15 intervals (75 min) before the modeled surfacing interval
  localization accuracy: 1/1 (provisional, n=1)
  layer attribution: 1/1 (provisional, n=1)
  box-swap discrimination: by construction the invisible fault is defined to survive a set-top-box swap, so the swap is avoided here - a designed demonstration of segment-vs-CPE separation, not a measured discrimination rate
  action correctness: 1/1 (provisional, n=1)
  note: detected at interval 365, named OLT-1500001-01 (access/cluster), lead 15 intervals to the modeled surfacing interval at 380

false-positive rate: 0 of 173 no-fault intervals fired a false positive (single-run count, not a population rate; ~2% upper bound at 95% by the rule of three)
decoys that fired: 0/3

Map of Spain · illustrative footprint (synthetic)

Markers = the run's detected elements at their real ids (red = access-node cluster; amber = isolated household, whose access node stays healthy) · click a marker for its certified-decision receipt. Synthetic demo · illustrative geography — not real network data.

Static run report generated from the verdicts only (four fields per home). Synthetic, fault-injected demo - figures illustrate the method, not a real network.