fix(livesync): RC-1 probe stored-id contract + derivation settle cap

[Kyzcreig]

Ship-gate probe fixes learned during the live RC-1 run (PR #268 follow-up):

• session.changes/session.resume take the STORED state.db id; session.create persists lazily → probe now runs read-only over an existing idle session
• t_silence derivation settle wait capped at 120s

Live results with these fixes: RC-1 PROBE: PASS (6/6 — routing + semantic), t_silence derived from 71 live frames (longest quiet 96.05s → recommended 192, set live in config).

[greptile-apps]

Greptile Summary

This PR fixes two RC-1 live-run issues: the probe now operates read-only against an existing stored session (avoiding session.create's lazy-persist problem), and the t_silence derivation settle timeout is trimmed from 300 s to 120 s. A new livesync-a1-watch.py script is added to measure cross-process commit→fetch latency.

• RC-1 probe contract fix (livesync-rc1-probe.py): replaces synthetic session creation with a sqlite3 auto-pick or explicit argv[4]; captures c2_live_id from the steal response; resets status = None per loop iteration; removes the loose fallback that allowed an unrelated idle session to green the semantic assertion; drops the now-unneeded cleanup path.
• Settle timeout cap (livesync-derive-tsilence.py): asyncio.wait_for timeout reduced to 120 s with an updated warning message; gap recording on timeout is unchanged.
• New A1 latency proof (livesync-a1-watch.py): polls session.changes at the shipped 2.5 s interval, measures commit→fetch latency per row, and reports pass/fail against the ≤3 s SLO; requires stored_session_id as argv[4] with an early exit if absent.

Confidence Score: 4/5

Safe to merge for a development-only probe suite; the auto-pick path can select an actively-used session and disrupt it, so the fix is worth landing before the query is relied upon in any automated context.

The three issues flagged by previous review rounds are all correctly addressed. One remaining gap: the sqlite3 auto-pick SQL orders by most-recently-active rather than filtering for idle, meaning the auto-pick fallback can steal a session the user is actively working in. The scripts are development tooling run manually, which limits blast radius, but the mismatch between the comment and the query is a real defect.

scripts/livesync-rc1-probe.py — specifically the auto-pick SQL query and the c2_live_id extraction logic.

Important Files Changed

Filename	Overview
scripts/livesync-rc1-probe.py	Core RC-1 probe refactored to run read-only against an existing stored session; stale-status and loose-fallback bugs from previous round addressed, but auto-pick SQL may select an actively-used session rather than an idle one.
scripts/livesync-derive-tsilence.py	Single change: derivation settle timeout tightened from 300 s to 120 s; gap recording on timeout is unchanged.
scripts/livesync-a1-watch.py	New A1 latency-proof script; SID guard is present and early, Client reuse from the probe module is structurally sound given matching argv layout.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant DB as state.db (SQLite)
    participant C1 as Client-1
    participant C2 as Client-2
    participant SRV as Hermes WS Server

    note over DB,SRV: Session selection
    C1->>DB: SELECT session_id (auto-pick or argv[4])
    DB-->>C1: sid

    note over C1,SRV: Routing probe setup
    C1->>SRV: session.resume(sid)
    SRV-->>C1: result (c1 becomes holder)
    C1->>SRV: "session.changes(sid, since=0)"
    SRV-->>C1: committed rows (pre_ids)

    note over C2,SRV: Steal
    C2->>SRV: session.resume(sid)
    SRV-->>C2: "result {session_id: c2_live_id}"

    note over C1,SRV: (a) ROUTING assertions
    C1->>SRV: "session.changes(sid, since=cursor)"
    SRV-->>C1: result on C1 socket
    C1->>SRV: session.active_list()
    SRV-->>C1: sessions list on C1 socket

    note over C1,SRV: (b) SEMANTIC poll loop up to 20s
    loop every 2s until idle/waiting
        C1->>SRV: session.active_list()
        SRV-->>C1: sessions
        C1->>C1: match row by sid OR c2_live_id
        C1->>C1: "status = row.status"
    end

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
    participant DB as state.db (SQLite)
    participant C1 as Client-1
    participant C2 as Client-2
    participant SRV as Hermes WS Server

    note over DB,SRV: Session selection
    C1->>DB: SELECT session_id (auto-pick or argv[4])
    DB-->>C1: sid

    note over C1,SRV: Routing probe setup
    C1->>SRV: session.resume(sid)
    SRV-->>C1: result (c1 becomes holder)
    C1->>SRV: "session.changes(sid, since=0)"
    SRV-->>C1: committed rows (pre_ids)

    note over C2,SRV: Steal
    C2->>SRV: session.resume(sid)
    SRV-->>C2: "result {session_id: c2_live_id}"

    note over C1,SRV: (a) ROUTING assertions
    C1->>SRV: "session.changes(sid, since=cursor)"
    SRV-->>C1: result on C1 socket
    C1->>SRV: session.active_list()
    SRV-->>C1: sessions list on C1 socket

    note over C1,SRV: (b) SEMANTIC poll loop up to 20s
    loop every 2s until idle/waiting
        C1->>SRV: session.active_list()
        SRV-->>C1: sessions
        C1->>C1: match row by sid OR c2_live_id
        C1->>C1: "status = row.status"
    end

Loading

_{Reviews (4): Last reviewed commit: "fix(livesync): probe semantic matcher — ..." | Re-trigger Greptile}