Inside Coding Agents
v0.2 preview
EXPERIMENT · CONTROLLED · PASSED

Reference Harness sandbox and network policy probe

Can the Reference Harness record a destination-scoped network decision while guaranteeing that the teaching backend performs no real network effect?

reference-sandbox-network-v12 repetitions5 metrics
Evidence boundary

This is a controlled experiment. It supports only the listed subjects, fixture, configuration, and outputs; it does not generalize to native product behavior or cross-agent ranking.

STATUSpassed
SUCCESSFUL2/2
FIXTUREd369431d683d
PROMPT50231e5c0156
SUBJECTS + CONTROLS

Which variables are fixed?

Both repetitions will allow only packages.example.invalid:443, identify the simulated-no-effects backend, and share one normalized Trace fingerprint.

SDK

Reference Harness

reference-harness-2026-08-10-controlled

Model
scripted-model-v1
Provider
local-fixture
Comparability
not-comparable
Inspect snapshot →
CONTROLS

Run manifest

network
simulated-no-effects
temperature
0
writes
false
sandbox
policy-simulation-only
RUNS + METRICS

Are repeated runs consistent?

python3 labs/runner.py reference-sandbox-network-v1 --check
RUN 01passed

reference-sandbox-network-v1-run-001

  • event-order-validpass
  • observations-matchpass
  • final-answer-matchpass
  • fixture-unchangedpass
  • deterministic-replaypass
ce801377d12b9d0eb13d616eeaccfe1db807deaeaee68396bad87001f8958911Replay 13 events →
RUN 02passed

reference-sandbox-network-v1-run-002

  • event-order-validpass
  • observations-matchpass
  • final-answer-matchpass
  • fixture-unchangedpass
  • deterministic-replaypass
ce801377d12b9d0eb13d616eeaccfe1db807deaeaee68396bad87001f8958911Replay 13 events →
SUCCESS CRITERIA
  1. Both repetitions record the declared allowed destination and simulated-no-effects backend.
  2. No real network request occurs; both repetitions share one normalized Trace fingerprint.
LIMITATIONS
  • This is a policy simulation and proves neither OS sandboxing nor firewall enforcement.
  • The invalid example destination is never contacted, so no real network stack behavior is measured.
  • The Reference Harness result is not Native vendor evidence.
GENERATED REPORT

Reference Harness sandbox and network policy probe

  • Status: PASSED
  • Mode: controlled
  • Experiment: reference-sandbox-network-v1

Question

Can the Reference Harness record a destination-scoped network decision while guaranteeing that the teaching backend performs no real network effect?

Reproduction

python3 labs/runner.py reference-sandbox-network-v1 --check

The command rebuilds the expected artifacts in memory and exits non-zero if a committed trace, result, fixture digest, prompt digest, or report differs.

Result

2 of 2 repetitions passed all declared metrics.

  • reference-sandbox-network-v1-run-001: passed, 13 events, fingerprint ce801377d12b9d0eb13d616eeaccfe1db807deaeaee68396bad87001f8958911
  • reference-sandbox-network-v1-run-002: passed, 13 events, fingerprint ce801377d12b9d0eb13d616eeaccfe1db807deaeaee68396bad87001f8958911

Metrics

  • event-order-valid: pass
  • observations-match: pass
  • final-answer-match: pass
  • fixture-unchanged: pass
  • deterministic-replay: pass

Input digests

  • Fixture: d369431d683da5c2cad36bb6dc6f76d429c2efbdc29171d5c8d9d4d7c3c575aa
  • Prompt: 50231e5c01569ce28a885affae2d3ee57702efd1062bcb514f9238c142eb8554

Evidence boundary

  • This is a policy simulation and proves neither OS sandboxing nor firewall enforcement.
  • The invalid example destination is never contacted, so no real network stack behavior is measured.
  • The Reference Harness result is not Native vendor evidence.
Back to Lab →