From 0da655979d3f801118d2f20a97f1f0c7abe3f11a Mon Sep 17 00:00:00 2001 From: tegwick Date: Thu, 2 Jul 2026 11:54:04 +0200 Subject: [PATCH] ACTIVITY-WP-0016: T05 done via live railiance01 proof, T01 cancelled (evidence unrecoverable) Co-Authored-By: Claude Fable 5 --- ...16-llm-output-robustness-trust-boundary.md | 21 +++++++++++++++++-- 1 file changed, 19 insertions(+), 2 deletions(-) diff --git a/workplans/ACTIVITY-WP-0016-llm-output-robustness-trust-boundary.md b/workplans/ACTIVITY-WP-0016-llm-output-robustness-trust-boundary.md index f601013..67dc056 100644 --- a/workplans/ACTIVITY-WP-0016-llm-output-robustness-trust-boundary.md +++ b/workplans/ACTIVITY-WP-0016-llm-output-robustness-trust-boundary.md @@ -90,7 +90,7 @@ there are two sensible strategies — and they compose: ```task id: ACTIVITY-WP-0016-T01 -status: wait +status: cancel priority: high state_hub_task_id: "74fd16a5-4ea5-4dfe-8526-dfa27cf76138" ``` @@ -353,7 +353,7 @@ Done when: ```task id: ACTIVITY-WP-0016-T05 -status: progress +status: done priority: high state_hub_task_id: "c881500b-5459-4620-81c0-b176971e989f" ``` @@ -415,3 +415,20 @@ for the next 07:20 Berlin fire, then confirm State Hub `daily_triage` evidence i - **Boundary discipline:** keeps activity-core inside its SCOPE — this hardens the instruction-executor output contract; it does not move provider credentials, cluster reconciliation, or task lifecycle into this repo. + + +## Closure 2026-07-02 (RAIL-BS-WP-0008 live deploy) + +- T05 done: the robustness bundle (strict per-item schema + T03 quarantine + parser + bounded top-7/NDJSON runtime prompt, activity-core `7612112`) was + deployed to railiance01 and live-proven. A manually triggered daily-triage + run produced a clean schema-valid report with exactly 7 ranked + recommendations: State Hub event `24d2d321-c761-47f7-bf9e-7950a6253c21`, + `output_validated=true`, working memory written. Calibration re-entry: the + three-clean-run streak (WP-0006-T03 / WP-0010-T04) restarts from this run. +- T01 cancelled: the raw 2026-06-26 llm-connect response is unrecoverable + (stateless pod, no response store, log stream holds only 2 startup lines + since 2026-06-19). Root cause stands on the retained 4000-char preview and + break-at-char-5268 evidence: output exceeded the old ~1200-token budget and + truncated mid-JSON. The deployed mitigation (1800-token headroom, bounded + top-7, per-item recovery) addresses exactly that failure mode.