Upload 6 files

CHANGELOG.md

@@ -1,5 +1,58 @@
+
+
+## v121 latency/display cleanup
+
+- Reduced perceived latency when re-opening already inspected runs by reusing cached progress/detail payloads immediately and avoiding duplicate polling starts.
+- Added DOM render signatures for run documents, Agent recovery and Gradio endpoint panels to prevent flicker during polling.
+- Preserved heavy hydration in the background for full reports/artifacts while keeping the central panels responsive.
+
+
+## v120 — Agent recovery UI for blockage protocol
+
+- Added an **Agent recovery** card to the Active Run panel so the v118/v119 core recovery protocol is visible to users.
+- The card summarizes Pi diagnosis decisions, selected action, logs quality/failure category/confidence, evidence, next step, and the diagnose → decide → act → revalidate ladder.
+- Exposed `repair/REPAIR_DECISION.json` and `repair/BLOCKAGE.json` through the backend bundle, progress payload, and Run traces dock.
+- Kept the existing unified `agent_trace.jsonl` as the canonical raw/redacted narrative across initial build, diagnosis, and repair.
+- Added regression tests for the recovery panel, decision/blockage artifact links, progress payload exposure, and docs.
+
 # Changelog
 
+## v119 — unified agentic Pi traces
+
+- Added a run-level `traces/raw/agent_trace.jsonl` and `traces/redacted/agent_trace.jsonl` journal that ties together the initial Pi build, Pi blockage diagnosis, repair decision, and optional repair patch into one auditable agent narrative.
+- Kept specialized logs (`logs/pi_output.txt`, `logs/pi_diagnosis_output.txt`, `logs/pi_repair_output.txt`) as detailed artifacts, but made the curated `Pi RAW` / `Pi redacted` dock point to the unified trace when available.
+- Diagnosis and repair phases now add explicit trace records for command start/end, output artifacts, repair decisions, and repair plan/summary artifacts so they no longer look like disconnected conversations.
+
+
+## v118 — Pi diagnosis blockage protocol
+
+- Added a Pi **blockage diagnosis** step before any code repair. Failures now produce `repair/INCIDENT_BRIEF.md`, `repair/PI_DIAGNOSIS_GOAL.md`, `repair/REPAIR_DECISION.json`, and `logs/pi_diagnosis_output.txt`.
+- Pi/Qwen now chooses a bounded action instead of being asked to patch immediately: `wait_for_logs`, `inspect_more_logs`, `factory_rebuild_same_code`, `patch_code`, `request_manual_hardware`, or `declare_technical_blocker`.
+- Added Factory guardrails: no code patch on empty/no-reason logs, no patch for auth/quota/hardware blockers, and strict one-action budgets for wait, same-code rebuild, and patch repair.
+- Added same-code factory rebuild recovery for HF build/runtime flakes where logs are empty or non-actionable.
+- Patch repair now runs only after Pi diagnosis chooses `patch_code` and the Factory validates the decision.
+- Added blockage artifacts and timeline events: `failure_detected`, `failure_diagnosis`, `pi_diagnosis`, `repair_decision`, `wait_for_logs`, `factory_rebuild`, `technical_blocker`, and `manual_hardware_required`.
+- Updated tests to protect the diagnosis-before-patch contract and the failure timeline mapping.
+
+## v117 — Structured repair mode and repair timeline
+
+- Reworked Pi repair from a loose log-based retry into a structured repair method.
+- Added failure classification before repair (`dependency_error`, `import_error`, `gradio_api_mismatch`, `cuda_oom`, `model_loading_error`, `hf_auth_error`, `space_boot_timeout`, `wrong_output_type`, `zero_gpu_duration_error`, or `unknown_runtime_error`).
+- The worker now writes `repair/REPAIR_BRIEF.md`, requests `REPAIR_PLAN.md` before patching, and stores `REPAIR_SUMMARY.md` after patching.
+- Added strict repair constraints: patch the current workspace, do not rebuild unnecessarily, preserve real inference, do not fake outputs, keep the health endpoint, and preserve the validation contract.
+- Added repair sanity checks before redeploying a patched workspace.
+- Added repair-specific events: `repair_diagnosis`, `repair_plan`, `repair_patch`, `repair_upload`, and `repair_validation`.
+- Updated the progress timeline so repair phases are visible and terminal failures always have a red failure point.
+- Added a Repair document shortcut in the Active Run traces dock when repair artifacts exist.
+
+
+## v116 — Compact Run Explorer and billing note relocation
+
+- Reworked the Run Explorer from large cards into a compact premium navigation list with one primary line, one context line, status pill, relative time and a discreet delete action.
+- Kept build and validation rows distinct while making the business entity more prominent than the raw run ID.
+- Moved the Inference Providers / paid credits / restricted A100-H200 fallback note out of New Build and into Billing & compute.
+- Added regression tests for the compact list and billing-note placement.
+
 ## v114 — Premium vector activity feed icons
 
 - Replaced text-coded activity markers such as `PI`, `API`, and `GPU` with inline SVG icons rendered by a dedicated `renderEventIcon(...)` helper.

README.md

@@ -46,7 +46,7 @@ The UI is organized into three main areas:
 
 - **Run storage** — check or create the signed-in user's private Bucket.
 - **New Build** — starts from `Tongyi-MAI/Z-Image-Turbo` by default, fast pre-scans the model card, then launches a private Space build only after an acceptable scan verdict.
-- **Billing & compute** — show payment readiness when available and link to HF Billing, Jobs pricing, and Inference Providers pricing.
+- **Billing & compute** — show payment readiness, link to HF Billing, and centralize the compute-cost note for Jobs, Inference Providers, and safe hardware fallbacks.
 
 ### Center workspace
 
@@ -55,7 +55,7 @@ Two tabs share the main detail area:
 - **Active Run** — build progress, run actions, Space/Settings/Job/Artifacts links, manual hardware guidance, inference latency, a vector-icon latest-events feed, and curated trace documents.
 - **Space Test** — validation run details, actionnable Gradio endpoint picker, endpoint params, validation status, latency, point-based progress, and validation actions.
 
-Selecting a build run opens `Active Run`. Selecting a validation run opens `Space Test`.
+Selecting a build run opens `Active Run`. Selecting a validation run opens `Space Test`. The Runs Explorer is intentionally compact: each row prioritizes the model or Space, status, relative freshness, and a secondary context line while keeping the raw run ID available in the row.
 
 ### Right column
 
@@ -63,6 +63,13 @@ Selecting a build run opens `Active Run`. Selecting a validation run opens `Spac
 - **Run Stats** — aggregate metrics across indexed runs: totals, success/pass rates, validation latency median/p95, and issue counts.
 
 
+
+### Agent recovery visibility
+
+When a build or runtime blockage appears, the Active Run panel shows an **Agent recovery** card. It summarizes Pi's diagnosis decision, the selected recovery action (`wait_for_logs`, `factory_rebuild_same_code`, `patch_code`, manual hardware, or blocker), key evidence, and the next step. This makes the blockage protocol visible without turning the UI into a raw log viewer.
+
+The Run traces dock also links to the unified `agent_trace.jsonl`, the diagnosis `REPAIR_DECISION.json`, repair artifacts, and blockage details when they exist.
+
 ## Model pre-scan
 
 New builds are gated by a fast model-card pre-scan. The scan is metadata-only and does not download model weights. It checks:
@@ -102,7 +109,9 @@ runs/<run_id>/generated/
 runs/<run_id>/tests/
 runs/<run_id>/artifacts/
 runs/<run_id>/traces/raw/
+runs/<run_id>/traces/raw/agent_trace.jsonl
 runs/<run_id>/traces/redacted/
+runs/<run_id>/traces/redacted/agent_trace.jsonl
 runs/<run_id>/logs/
 ```
 
@@ -154,6 +163,11 @@ The validation job:
 - Hardware assignment is best-effort and can fall back to manual user action.
 - Raw and redacted Pi traces are kept separate.
 
+### Unified agent trace
+
+Pi can be invoked more than once during a run: the initial build, a blockage diagnosis pass, and an optional structured repair patch. The Factory now writes a canonical run-level agent journal to `traces/raw/agent_trace.jsonl` and `traces/redacted/agent_trace.jsonl` so those phases read as one auditable agentic loop instead of separate conversations. Specialized stdout logs remain available under `logs/`, but the Active Run `Pi RAW` / `Pi redacted` buttons prefer the unified trace when it exists.
+
+
 ## Non-goals and limits
 
 The app does not guarantee that every model card becomes a working Space. Some models require multi-GPU execution, Docker, unavailable CUDA extensions, external services, accepted licenses, very large hardware, or manual fixes.
@@ -182,3 +196,18 @@ See:
 
 
 Default example model: `Tongyi-MAI/Z-Image-Turbo`. It is treated as a known-good Diffusers text-to-image model for the fast pre-scan calibration and public-preview smoke tests.
+
+### Pi diagnosis and structured recovery
+
+When the generated Space fails live validation, the factory now enters a diagnosis-first blockage protocol instead of asking Pi to patch immediately. The worker collects HF build/runtime logs, Space status and validation errors, writes `repair/INCIDENT_BRIEF.md`, then asks Pi/Qwen Coder to choose one bounded action in `repair/REPAIR_DECISION.json`: wait for delayed logs, inspect again, force a same-code factory rebuild, patch code, request manual hardware, or declare a technical blocker.
+
+A code repair is allowed only when Pi selects `patch_code` and the Factory accepts the decision. Empty/no-reason HF failures are routed through wait/rebuild/blocker paths rather than speculative code edits. Patch repair still writes `REPAIR_BRIEF.md`, `REPAIR_PLAN.md`, and `REPAIR_SUMMARY.md`, uploads once, then revalidates the live Space with the normal API checks.
+
+Repair mode is intentionally strict: it must preserve the original model and real-inference contract, avoid fake/static outputs, keep a cheap health endpoint, and report blockers instead of hiding runtime errors.
+
+## v121 latency/display cleanup
+
+- Reduced perceived latency when re-opening already inspected runs by reusing cached progress/detail payloads immediately and avoiding duplicate polling starts.
+- Added DOM render signatures for run documents, Agent recovery and Gradio endpoint panels to prevent flicker during polling.
+- Preserved heavy hydration in the background for full reports/artifacts while keeping the central panels responsive.
+

app.py

@@ -597,6 +597,8 @@ def register_custom_routes(fastapi_app: FastAPI) -> None:
                 "api_schema": bundle.get("api_schema") or {},
                 "hardware_strategy": bundle.get("hardware_strategy") or {},
                 "technical_blockers": bundle.get("technical_blockers") or {},
+                "repair_decision": bundle.get("repair_decision") or {},
+                "blockage": bundle.get("blockage") or {},
                 "run_documents": bundle.get("run_documents") or [],
                 "events": events[-20:],
                 "view": view,