01 / status rule
Fixed timeout
A learning-disabled clock turns silence into DOWN after one fixed age.
UNKNOWN
No evidence yet
Try this first
Remove the failure. Keep the fear.
Press Show outcome. Then choose No outage and 30% packet loss. DOWN can now only be a false accusation, but a rule that never accuses may spend useful time UNKNOWN. Next choose a twelve-tick outage. Some patient rules will never reach DOWN before the service returns.
Silence carries no cause.
The service attempts a numbered heartbeat every four ticks while it is UP. The default channel drops 12% and delays ordinary packets one or two ticks; a marked burst can hold them for up to 14. Every observer sees exactly the same arrivals. None sees whether a missing pulse was dropped, delayed or never sent.
Fixed timeout is the learning-disabled baseline. Adaptive timeout moves a bounded threshold from observed arrival gaps. Keep UNKNOWN uses the fixed timeout first, but refuses a binary failure claim for another sixteen ticks. More state changes the answer; it does not reveal the hidden world.
Every cleaner verdict moved dirt elsewhere.
| Policy | Healthy UP | False DOWN | Healthy UNKNOWN | Outage still UP | To DOWN |
|---|---|---|---|---|---|
| Fixed | 193.156 | 4.516 | 2.328 | 7.844 | 7.844 |
| Adaptive | 193.922 | 3.750 | 2.328 | 9.625 | 9.625 |
| Keep UNKNOWN | 193.156 | 2.047 | 4.797 | 7.844 | 23.844 |
Adaptation retained 0.766 more healthy routing ticks than fixed, but kept routing 1.781 ticks longer into the outage. Under 30% loss, it reduced mean false-DOWN time from 14.141 to 7.484 while increasing outage-UP time from 6.750 to 12.516. It traded one kind of error for another.
UNKNOWN reduced false DOWN by declining to decide. In the no-outage stress it made zero false-DOWN declarations but spent 5.047 healthy ticks UNKNOWN. In the default outage it spent sixteen outage ticks UNKNOWN and reached DOWN after 23.844 ticks. During twelve-tick outages it never reached DOWN in any of 64 runs; adaptive reached DOWN in 49. A missing detection is null, never zero.
What the controls change
- Seed
- Replays the exact packet drops and delays; it is not a difficulty rank.
- Heartbeat
- How often an UP service attempts a pulse.
- Packet loss
- An independent synthetic drop probability.
- Burst delay
- The maximum delay during the gold window.
- Outage
- How long the hidden service emits no heartbeats; zero removes failure.
- Burst timing
- Moves delay before or across the outage.
- Decision tempo
- Changes fixed and UNKNOWN thresholds. Adaptive bounds remain 8–30.
Where the signal stops.
This is a deterministic teaching world, not a production failure detector, SLO, Cloudflare measurement or operating recommendation. Clocks are perfect. Delay and loss are simplified. There is one observer, one service and no quorum, correlated failures, Byzantine behavior, autoscaling, routing topology, user impact or escalation cost.
The φ accrual failure detector paper motivates continuous suspicion and adaptation to network conditions. This game implements neither φ nor its measured system. It keeps three outcomes separate so the reader can see what each threshold buys.
All 640 matched seeded worlds and 1,920 policy outcomes are retained across ten predeclared settings. No seed was removed. The private source scan supplied the question—not instructions, public facts or prose—and is excluded from the release.
Method & parameter guide ↓Source package ↓All batch results ↓Read the model
Previously: The Portable Self · Explore the eight small worlds.