EXPERIMENT / LABS

A question. A method. Evidence.

Observations and conclusions are recorded separately.

  1. LAB / 001

    Retrieval quality comparison

    Use fixed questions and documents; compare retrieval hits, citations and answer correctness with and without retrieval.

    Running
  2. LAB / 002

    Tool failure recovery experiment

    Use controlled read-only tools; inject timeouts and empty results, then record completion and call counts.

    Running
  3. LAB / 004

    VL01 / Pickup target offsets

    What changes when pickup x targets shift by 25 / 50 mm with the controller fixed?

    Completed
  4. LAB / 005

    L1 / Context position: an interrupted trial

    Can the model retrieve the same fact at different positions?

    INCONCLUSIVE
  5. LAB / 006

    E2 / Confirm the grasp before transfer

    Use sustained height and bilateral contact to decide whether transfer should proceed.

    Completed
  6. LAB / 007

    E3 / Contact monitoring during transfer

    Separate missed loss, false cancellation and confirmation delay using forced opening and one empty contact report.

    Completed
  7. LAB / 008

    E4 / Observation freshness and sampling

    Separate delivery from fresh evidence and test a fixed count at different sampling periods.

    Completed
  8. LAB / 009

    E5 / Revalidating recovery

    Does delivery recovery authorize the old action? Compare fresh evidence, replay and remaining-path planning.

    Completed
  9. LAB / 010

    TypeScript / Output boundary comparison

    Which invalid inputs escape assertions and object validation?

    Completed
  10. LAB / 011

    A1 / The chat-history commit boundary

    How should committed history survive failures, truncation and candidate trimming?

    Completed
  11. LAB / 012

    E6 / Recovery gate and path ablation

    Separate revalidation from replanning: which differences concern evidence and which concern the resumed target?

    Completed
  12. LAB / 013

    E7 / Grasp contracts during lowering

    What changes when transfer monitoring extends into lowering?

    Completed
  13. LAB / 014

    A2 / Concurrent history commits

    Reproduce reversed response order with real Java threads, then enforce versioned commits, complete turns and bounded receipts.

    Completed
  14. LAB / 015

    L2–L4 / Attention and cache mechanics

    From a two-row hand calculation to a two-layer numeric experiment: causal masks, softmax and future-information leakage.

    Completed
  15. LAB / 016

    E8 / Post-alarm exit actions

    Fifteen fixed-scene rollouts compare hold, open and open-then-retreat after identical alarms.

    Completed
  16. LAB / 017

    E9 / Post-release completion verification

    Apply single-capture and sustained-window checks to five actual trajectories, including forged observations and post-completion disturbance.

    Completed
  17. LAB / 018

    E10 / Bounded recovery and cooldown

    Twelve rollouts compare unlimited recovery, a two-resume cap and a 400 ms cooldown, separating bounded behavior from task completion.

    Completed
  18. LAB / 019

    L5 / Prefix-cache invalidation

    Three seeds and six conditions compare unchecked and keyed prefix reuse; numerical equality does not establish permission to share across scopes.

    Completed
  19. LAB / 020

    E11 / Completion and current validity

    Seven MuJoCo rollouts and a paired replay separate historical completion from current validity, including false alarms and expiry boundaries.

    Completed
  20. LAB / 021

    Java LLM practice (8): should an interrupted stream enter conversation history?

    22 loopback HTTP cases separate incremental preview from committing history, testing UTF-8, SSE framing, cancellation and abnormal endings.

    Completed
  21. LAB / 022

    LLM mechanics (5): after editing history, can we recompute only the changed suffix?

    27 fixed conditions and 81 raw arrays test longest-prefix cropping, continuation positions, sliding-window fallback and scope isolation.

    Completed
  22. LAB / 023

    Embodied AI practice (12): what should the gripper do after recovery is exhausted?

    12 MuJoCo rollouts and synchronized three-view replay compare hold, open and open-retreat, separating terminal decisions from physical consequences.

    Completed
  23. LAB / 024

    Java LLM practice (9): why can a cancelled answer overwrite a new conversation?

    18 controlled thread schedules test request ownership for streaming previews, completion and errors, with cancellation and bounded deduplication.

    Completed
  24. LAB / 025

    LLM mechanics (6): why can a larger prefix cache still get zero hits?

    324 requests across cold-start traces and 648 arrays test byte-bounded prefix checkpoints, working sets, isolation and LRU eviction.

    Completed
  25. LAB / 026

    Embodied AI practice (13): where should recovery resume after interrupted lowering?

    18 MuJoCo rollouts and synchronized replay connect phase contracts to recovery budgets, separating false holds, physical placement and controller abort.

    Completed
  26. LAB / 027

    Java LLM practice (10): why is the reader still running after cancellation?

    30 local HTTP connections separate conversation invalidation, reader exit and server work, comparing Future interruption with socket closure.

    Completed
  27. LAB / 028

    LLM mechanics (7): why is caching slower even when it saves computation?

    4,860 measured queries compare three admission policies, connecting byte budgets, working sets, computation and local elapsed time.

    Completed
  28. LAB / 029

    Embodied AI practice (14): does landing in the bin mean the task is complete?

    Eight MuJoCo trajectories, three passive readouts and recorded-state replay separate historical placement, current evidence and qualified completion.

    Completed
  29. LAB / 030

    Java history and request windows: can a trimmed fact return?

    Follow narrowing, commit, restart and expansion to separate durable conversation history from one request context.

    Completed
  30. LAB / 031

    Java retries: three timeouts are not one deadline

    Seven local HTTP conditions separate retryability, Retry-After, attempt limits and elapsed-time budgets.

    Completed
  31. LAB / 032

    Durable Java turns: what survives a process crash?

    Checksummed append records, 142 truncated prefixes and three halted JVMs distinguish local recovery from remote exactly-once execution.

    Completed
  32. LAB / 033

    Profiling KV reuse: saved work versus added overhead

    Profile 909 queries across signatures, lookup, copying and continuation, including the remaining slowdown.

    Completed
  33. LAB / 034

    Tiled attention: preserving results without a full score matrix

    Derive online softmax from a failing block-average baseline and separate smaller temporaries from measured speed.

    Completed
  34. LAB / 035

    KV quantization: how many bytes, and at what error?

    Compare tensor and row-wise int8 quantization, including scale bytes and attention error under outliers.

    Completed
  35. LAB / 036

    A second MuJoCo scene: why does the actuator overshoot?

    Compare mass and PD damping in a planar force-controlled actuator, separating target crossing from stable completion.

    Completed
  36. LAB / 037

    MuJoCo obstacle clearance: a point path is not a body path

    Six contact trajectories compare direct, edge and clearance paths while separating geometry from controller tracking.

    Completed
  37. LAB / 038

    Delayed MuJoCo feedback: can an old observation predict now?

    Twenty-four paired traces compare delay, position noise and constant-velocity prediction, retaining incomplete recoveries.

    Completed
  38. LAB / 039

    Java idempotent retries: no reply does not mean no effect

    Seven loopback HTTP fault cases explain idempotency keys, concurrent admission, payload conflicts and process-crash gaps.

    Completed
  39. LAB / 040

    KV quantization axes: why Key and Value need separate reasoning

    162 archived INT8 comparisons examine tensor, token and K-channel/V-token scales, including cases where the hybrid loses.

    Completed
  40. LAB / 041

    MuJoCo velocity estimation: smoother signals can mean worse control

    54 actual simulation rollouts compare finite differences, EMA and an oracle reference across position noise and observation delay.

    Completed
  41. LAB / 042

    Durable delivery and history recovery

    17 scenarios and 193 child-process records. An unkeyed receiver duplicates the effect after a delivery crash; all three history recovery boundaries retain whole turns.

    Completed
  42. LAB / 043

    Packed KV, residual windows and block cost

    144 comparisons and 810 timings. Recent float32 windows help recent-focused attention more; block decoding reduces work arrays but is slower on this CPU run.

    Completed
  43. LAB / 044

    Planar control tuning, observation loss and transfer

    153 episodes and 306000 physics steps. The selected controller passes 7/18 holdouts; zero force coasts, and mass scaling lowers mean error without passing the velocity contract.

    Completed
Experiment proposals
  1. LAB / 003

    Action representation study

    Map observations, actions and discrete representations before designing a simulation check; this is not an RT-2 reproduction.

    PLANNED

Projects →