[P2.5] request_log Telemetry Table — Verify What Actually Happened #446

New Issue

perplexity · 2026-04-09T22:17:30Z

perplexity commented

2026-04-09 22:17:30 +00:00

Source

KT Bezalel Architecture Session 2026-04-08 — Immediate Priority #3

Problem

Without telemetry, we cannot verify if any agent actually executed what it claims. This is "non-negotiable infrastructure" per Alexander.

Schema (exact, from KT)

CREATE TABLE request_log (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    timestamp TEXT NOT NULL,
    agent_name TEXT NOT NULL,
    provider TEXT NOT NULL,
    model TEXT NOT NULL,
    endpoint TEXT NOT NULL,
    tokens_in INTEGER,
    tokens_out INTEGER,
    latency_ms INTEGER,
    status TEXT NOT NULL,  -- "success", "error", "timeout", "fallback"
    error_message TEXT
);

Rules

Every agent writes to this table BEFORE and AFTER every inference call
No exceptions. No summarizing. No describing what you would log. Actually write the row.
One SQLite file per machine, synced to Gitea periodically

What This Solves

The RunPod problem — 00 spent, no idea if it was used
The Claude fabrication problem — agent claims it did work, no proof
The MiMo evaluation problem — did the new agent actually run on MiMo?
Provider comparison — which provider is fastest, cheapest, most reliable

Acceptance Criteria

SQLite table created with exact schema above
Every inference call writes a row (pre-call with status pending, post-call with result)
Fallback events logged with status="fallback"
Error events logged with error_message populated
Query interface: can answer "did agent X actually call provider Y in the last hour?"
Ansible role deploys the telemetry table to all machines
Log rotation / archival strategy defined (don't let it grow unbounded)

Dependencies

Provider fallback chain fix (logs fallback events here)
Deadman switch (logs rollback events here)
Must be wired BEFORE resurrecting wizards

## Source KT Bezalel Architecture Session 2026-04-08 — Immediate Priority #3 ## Problem Without telemetry, we cannot verify if any agent actually executed what it claims. This is "non-negotiable infrastructure" per Alexander. ## Schema (exact, from KT) ```sql CREATE TABLE request_log ( id INTEGER PRIMARY KEY AUTOINCREMENT, timestamp TEXT NOT NULL, agent_name TEXT NOT NULL, provider TEXT NOT NULL, model TEXT NOT NULL, endpoint TEXT NOT NULL, tokens_in INTEGER, tokens_out INTEGER, latency_ms INTEGER, status TEXT NOT NULL, -- "success", "error", "timeout", "fallback" error_message TEXT ); ``` ## Rules - Every agent writes to this table **BEFORE and AFTER** every inference call - No exceptions. No summarizing. No describing what you would log. **Actually write the row.** - One SQLite file per machine, synced to Gitea periodically ## What This Solves - **The RunPod problem** — 00 spent, no idea if it was used - **The Claude fabrication problem** — agent claims it did work, no proof - **The MiMo evaluation problem** — did the new agent actually run on MiMo? - **Provider comparison** — which provider is fastest, cheapest, most reliable ## Acceptance Criteria - [ ] SQLite table created with exact schema above - [ ] Every inference call writes a row (pre-call with status pending, post-call with result) - [ ] Fallback events logged with status="fallback" - [ ] Error events logged with error_message populated - [ ] Query interface: can answer "did agent X actually call provider Y in the last hour?" - [ ] Ansible role deploys the telemetry table to all machines - [ ] Log rotation / archival strategy defined (don't let it grow unbounded) ## Dependencies - Provider fallback chain fix (logs fallback events here) - Deadman switch (logs rollback events here) - Must be wired BEFORE resurrecting wizards

perplexity added this to the KT-2026-04-08: Infrastructure Stabilization milestone 2026-04-09 22:17:30 +00:00

perplexity referenced this issue from a commit

2026-04-09 22:25:41 +00:00

feat(ansible): Canonical IaC playbook for fleet management

perplexity referenced this issue

2026-04-09 22:26:05 +00:00

[P2] Ansible IaC — Canonical Fleet Playbook #449

bezalel was assigned by Timmy

2026-04-10 00:00:53 +00:00

Sign in to join this conversation.