# timeline read-boundary verification

baseline: 220eb73524ca7882d53b6a12c824ccbcd97ad279
measured candidate: c00f367b08287062c97d5e5f7d6f2a7b08c3dac9
later changes only update compact-cache normalization and typed test callers; measured snapshot loader is unchanged.

linux boat vm, node v24.22.1, development server started with vp run dev defaults in /home/user/t3code-evidence. isolated copied/pruned state, with one synthetic stopped 6,000-item system-notice thread. same database, endpoint, authentication cookie, script, machine, 5 warmups and 15 samples per revision. actual generated client fetchEnvironmentThreadSnapshot -> authenticated HTTP GET /api/orchestration/threads/perf-21491 -> actual server projection store and encoding -> actual client JSON parsing, schema decode, known-item filtering.

instrumentation wraps the fetch Response.arrayBuffer method, marking response-body completion. client-phase time ends when the real loader returns. each sample asserts exactly 6,000 decoded turn items. baseline and candidate use the same workload and response shape. client-phase includes text decoding, JSON.parse, schema decoding, and candidate filtering. it excludes server work and transfer time.

client-phase median: 77.632 ms -> 51.851 ms, 33.2% lower.
whole request median in these instrumented runs: 371.592 ms -> 313.193 ms. earlier uninstrumented runs varied from 304.6 ms baseline to 536.4 ms candidate, and another candidate run 627.4 ms. no end-to-end latency speedup is claimed. sequential processes and restarts, no cpu throttling, development build, synthetic rows; this is not a production/browser paint benchmark or mixed-provider workload.

raw baseline samples (ms):
{
  "root": "/home/user/t3code-evidence",
  "items": 6000,
  "warmups": 5,
  "decodeSamples": [
    77.63248299999941,
    108.77077800000006,
    60.44622699999945,
    79.17149299999983,
    114.43127699999968,
    77.23592199999985,
    81.66201100000035,
    73.28899900000033,
    76.32185200000094,
    80.5727139999999,
    70.58372899999995,
    63.95833700000003,
    72.49030599999969,
    78.74917299999834,
    80.05998499999987
  ],
  "medianDecodeMs": 77.63248299999941,
  "samples": [
    417.84172300000046,
    638.2127040000005,
    371.5923640000001,
    430.0815190000003,
    534.6074989999997,
    429.87042699999984,
    352.68964699999924,
    288.8208990000003,
    347.9320289999996,
    423.97565100000065,
    492.6229679999997,
    367.98934700000063,
    330.55013599999984,
    318.0273010000001,
    301.1474870000002
  ],
  "medianMs": 371.5923640000001
}

raw candidate samples (ms):
{
  "root": "/home/user/t3code-evidence",
  "items": 6000,
  "warmups": 5,
  "decodeSamples": [
    154.6407849999996,
    49.91030499999761,
    51.17788499999733,
    503.90989400000035,
    43.03667800000039,
    50.497683000001416,
    59.932419999997364,
    60.32223899999735,
    52.68985299999986,
    52.529373000001215,
    81.37102999999843,
    47.23115400000097,
    51.851456000000326,
    49.987638999999035,
    37.867668999999296
  ],
  "medianDecodeMs": 51.851456000000326,
  "samples": [
    1156.7997390000019,
    599.1037250000008,
    629.8841509999984,
    1278.3548169999995,
    253.7221529999988,
    488.36897699999827,
    268.495006000001,
    254.06944300000032,
    246.99946799999816,
    368.0173560000003,
    359.31552100000044,
    304.2132150000034,
    224.43340199999875,
    313.1928509999998,
    257.1071159999992
  ],
  "medianMs": 313.1928509999998
}

command on each revision:
node /home/user/t3code-evidence/.t3/measure-21491.mts /home/user/t3code-evidence

unverified: real newer-server binary emitting unknown variants, native mobile/electron clients, windows/macos, relay/tunnel and provider CLI turns. unknown variants, malformed known variants, compact ids and cache corruption are covered by focused existing-file tests. no rendering code changed.