The bridge becomes a runnable taskset. The crux is that these are continuous
control environments and the agent emits text: the model issues a control
decision that is held for N steps, and the hold length is measured rather than
guessed. At one decision every 32 steps the scripted baseline returns -2.807 and
reaches the waypoint 0% of the time, against 5.484 and 100% at every frame.
Open loop — one decision for the whole flight — returns -3.764 against an
inaction floor of -4.161. A bird that sets a course and leaves does no better
than one that does nothing, which is why crow-nav carries the strongest rule-4
claim of the five spatial environments. The budget bites from both ends: 8 turns
reach the goal 44% of the time, 12 reach 88%, 14 reach 100%.
32 episodes against Qwen3.8-27B: mean 0.4603, no malformed replies in 412, every
episode replaying to an identical checksum, and 16 of 32 spending all 16 turns.
The gate fires 43.75% of the time and separates the four scenarios cleanly —
whatever is wrong with the four dead data-environment gates, it is not that
binary gates cannot discriminate.
The oracle comes from the worker's own op. Nothing recomputes in Python a number
that came out of TypeScript, least of all the reward denominator.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Tera's four (now five) spatial environments already implement a Gym-style
contract with checksums, train/dev splits and a replay gate — in TypeScript,
unreachable from `uv run eval`. This bridges them rather than porting them: a
resident NDJSON worker over stdio drives the vendored TS closure, and no reward
arithmetic exists in Python anywhere in the package, including the denominator.
Correctness is replay, not assertion: every scenario x seed must round-trip
through Python and replay to an identical FNV-1a-64 checksum. 32/32 do, and 52
tamper attempts are rejected, 24 of them re-sealed with Tera's own checksum so
it is replay() catching divergence rather than the seal.
Lives in arena rather than tera because tera has no Python toolchain. That
overturns the earlier plan's claim that this workstream shares no files.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
kartiandClaude Opus 5