fix(core): build pressure on distances real walks actually cover

[prabu-openclaw]

Half the world-event vocabulary has never been reachable on a real walk. This is the mechanism behind the recurring operator report that the companion "feels dead".

Evidence

Five of the ten event kinds are gated on pressure, and pressure came almost entirely from distance on a 3 km scale (min(0.55, distanceMeters / 3000)). Converted to walking distance, each kind needed:

Kind	Pressure	Distance before it could fire
companionObserves	0.04	120 m
pursuitFades	0.18	540 m
distantPresence	0.24	720 m
pursuitBegins	0.36	1 080 m
pursuitIntensifies	0.55	1 650 m

Field walks in the receipts run 17–134 m. Bounded pursuit is a binding product pillar and had never once been reachable on device — the companion could only ever draw near, move ahead, or sit quiet. Receipts show 2 world events in 255 s and 4 in 185 s, drawn from an effective vocabulary of three.

Change

Distance still leads, on a scale a real walk reaches, plus a slow duration component so a short stroll still develops texture:

let distancePressure = min(0.45, session.distanceMeters / 1200)
let durationPressure = min(0.20, session.activeTime / 900)

Both stay individually capped, so pressure remains bounded rather than a ramp toward punishment, and the pursuit rules keep their own maximumPressure gates — calm kinds recede as tension rises, which is the intended shape.

Measured through WorldState.derive:

50 m  / 60 s    pressure 0.11   5/10 kinds
150 m / 180 s   pressure 0.33   6/10   + distantPresence, pursuitFades
400 m / 420 s   pressure 0.53   6/10   + pursuitBegins
800 m / 720 s   pressure 0.65   5/10   + pursuitIntensifies; calm kinds recede

What this deliberately does not change

Event frequency. I first also reduced minimumTickSpacing 40s → 26s, which raised yield from 1.60 to 2.40 events/min — and broke the frequency bounds asserted in both WorldEventWeightTuneTests and SoloMVPVerticalSliceTests. Those encode intentional sparseness as product law, so I reverted it. This is a variety change, not a density one, and the guards remain green.

If the sparseness bound is worth revisiting that is a separate product decision, with the numbers above as input.

Validation

• 133 Swift package tests, 0 failures
• 415 app tests, 0 failures
• New WorldEventCadenceTests measure yield and vocabulary reach rather than asserting intent, so future tuning can be judged against numbers

Note on authority

Sources/WaykinCore/ is your codeowner lane and this touches gameplay feel, so it wants product sign-off rather than a routine review. Happy to adjust the curve — the thresholds are a judgement call; the finding that pursuit was unreachable is not.

🤖 Generated with Claude Code