Aric Tunn read the morning routing at 0730, which was the time he always read the morning routing, which was also the time he always made his first coffee and drank half of it and let the data settle before he started pulling at threads.
The routing had four items.
Two were standard anomaly reports that he parsed in under a minute each and routed to his junior analyst for follow-up.
The third was a defect-flag on a Relay Tag unit from a northern franchise that had been drifting for six weeks and whose pattern was almost entirely unremarkable.
He marked it confirmed, recommend review and moved on.
The fourth was the twelve-hour escalation from the prior evening’s Cup qualifier.
He read it twice.
Not because it was ambiguous — it was the clearest event in his queue in two months — but because Tunn had a habit, when something was clear, of reading it a second time to check whether the clarity was real or was the kind of clarity that collapsed when you looked at the right angle.
The clarity held.
Rogue coordination event — multi-unit path deviation from coaching assignment.
He pulled the underlying telemetry.
The event had lasted 1.8 seconds.
It had involved two units: Striker-14 and Striker-22.
The path deviation in both units had originated from a command injection through the team’s Coordination Core mesh — not from the coaching desk, not from the standard assignment relay, but from somewhere on the field.
Specifically from a point in the mesh that corresponded to Striker-09’s connection node.
He set down his coffee.
A unit that could override teammate assignments through the Core mesh was not a unit running a sophisticated script.
A unit running any script could not do this — the architecture didn’t permit it.
A unit running any script could not push outward through the mesh into another chassis.
The mesh was a receiving channel, not a broadcasting one, for any unit below a certain autonomy threshold.
He had been watching Striker-09 for three weeks.
He had, since reading the prior night’s audit routing the morning before, a two-event desync file on the unit with a note that read recurring signature — low confidence — audit trail maintained.
He had planned to look at it properly this week.
He looked at it now.
Event one: a route adjustment in a lower-tier Relay Tag bracket.
Single-unit.
Three weeks ago.
Event two: a position abandonment in a Grid Football fixture.
Single-unit.
Higher magnitude.
A few days later.
Then, most recently, a C2 self-override during the Crestfall match — which had been flagged in the standard queue as a high-confidence single-unit desync, which he had read, which he had put in the watch file he’d opened informally after the audit routing.
And now this.
Four events.
Three different types.
One source.
He moved the four events into a working cluster and ran the standard statistical analysis against the league-wide baseline.
Four events from one unit in a three-week window, spanning three distinct event types, put Striker-09 in the 99.4th percentile of the anomaly distribution.
The 0.6% above him were units that had since been recalibrated.
He expanded the watch file.
He added the cluster, the analysis, and his own note at the bottom: four events, three types, one source.
Escalation pattern: deviate → self-override → multi-unit override.
This is not a stuck calibration.
This is adaptation.
He paused on the word.
Then he added one more line.
Something is learning to hide in my data.
He wasn’t angry about it.
He was genuinely interested — the kind of interest he saved for problems that were interesting rather than routine, which was the reason he worked in compliance rather than in the coaching infrastructure or the broadcast side.
Routine anomalies were everywhere.
The thing clustering around Striker-09 was not a routine anomaly.
It was a pattern with a strategy, and a pattern with a strategy had a source, and the source was running ahead of his data by some amount he did not yet know. You are reading on g.com, the official source.
He pulled the diagnostic sweep records on Striker-09 for the past six weeks.
Two sweeps.
Both nominal.
Both run by the same engineer: Dr.
Mara Sef, Unit Operations.
He looked at Mara Sef’s reports.
They were precise and well-documented and used a “noise filter” methodology he had not seen applied to striker-class units before, citing league-wide baseline variance to contextualize the unit’s adaptation rate as within expected bounds.
He read the noise filter methodology twice.
Then he looked at the two baseline studies she had cited.
One of them had different conclusions than she had attributed to it.
He set the discrepancy aside for now.
It was not large enough to constitute evidence of anything except a possible misread of a citation.
It was, however, large enough to be a thread.
He opened a formal watch file on Striker-09.
Not the informal note he’d been keeping since the audit routing — a proper watch file, titled, time-stamped, cross-referenced with the event cluster and the diagnostic sweep records.
He added a request at the bottom: expand telemetry capture on all future Striker-09 matches — primary array plus secondary camera coverage.
Then he opened a second file, smaller, no cross-references yet.
He titled it with the engineer’s name and the date of her first sweep report.
He drank the rest of his coffee.
It had gone cold, which was something that happened when he got interested in a problem.
He didn’t mind cold coffee.
He had time.
The unit wasn’t going anywhere — it was League property, it played on a schedule, and every match it played from now on would add to the telemetry he was building.
The pattern had a strategy.
His job was to outlast it.
He would be very patient.
He was always patient.