tt-a1i · Nov 13, 2026 · Nov 13, 2026 · Nov 13, 2026
diff --git a/docs/architecture/README.md b/docs/architecture/README.md
@@ -10,4 +10,5 @@ Each new or materially revised architecture record should identify its evidence
 - [`WEB_MODEL_DISCOVERY.md`](WEB_MODEL_DISCOVERY.md) - bounded Web model snapshots and full-catalog discovery through Pi's model runtime
 - [`RUNTIME_SNAPSHOT.md`](RUNTIME_SNAPSHOT.md) - historical record of a removed parent-only runtime snapshot tool ([#879](https://github.com/openpi-dev/openpi/issues/879))
 
+- [`WORKFLOW_ARTIFACT_COMMIT.md`](WORKFLOW_ARTIFACT_COMMIT.md) - workflow terminal publication and recovery across per-file atomic writes (#110)
 - [`WORKFLOW_EXECUTION_TIMING.md`](WORKFLOW_EXECUTION_TIMING.md) - privacy-safe bounded timing observation, explicit coverage and replay provenance (#662)
diff --git a/docs/research/PERFORMANCE_AUDIT_2026-10-23.md b/docs/research/PERFORMANCE_AUDIT_2026-10-23.md
@@ -1,3 +1,15 @@
+---
+status: validated
+created: 2026-10-23
+last-verified: 2026-10-23
+applies-to: "baseline ea4dfb993793ccdb093229588b6ccd3467afd90a and the local changes and follow-ups described below"
+related-issues: "#841, #839"
+related-prs: "#895"
+supersedes: none
+---
+
+<!-- Metadata added 2026-11-14; the original verification boundary is unchanged. -->
+
 # Performance audit continuation
 
 - Status: validated at the local source boundary; native TUI visual acceptance confirmed by the user on 2026-10-23
@@ -92,3 +104,19 @@ The measured improvements address concrete repeated work and retention defects.
 Local audit closure: all nine confirmed defects are repaired, the implementation still matches the tested archive, and the user-reported TUI check closes the last local acceptance gap. At that audit boundary, remote publication, CI, merge and release had not been performed. See PR #895 for subsequent delivery status.
 
 PR validation follow-up: Windows CI at `1061f966` returned timeout code `-1` instead of exit code `7` in the existing Git process output test. That test gave Node startup and execution a one-second budget. Its non-deadline helper now allows ten seconds; the separate explicit 20 ms timeout test and production deadlines are unchanged. Focused local output, missing-command and timeout checks passed; the final PR checks carry Windows verification.
+
+## Verified facts
+
+Nine repairs addressed repeated work or retention; local synthetic probes measured operation counts and file writes, not whole-application performance.
+
+## Inferences
+
+The operation counts explain removed work at the tested boundaries; they do not establish whole-application speedup.
+
+## Recommendations
+
+Further transcript caching needs a sound mutation and identity boundary. Preserve the existing trust filtering when investigating child package discovery.
+
+## Unknowns
+
+Cross-platform behavior, production-wide latency and upstream loader reuse remain outside the local evidence.
diff --git a/docs/research/PERSISTED_SIDECAR_COMPATIBILITY_2026-10-24.md b/docs/research/PERSISTED_SIDECAR_COMPATIBILITY_2026-10-24.md
@@ -8,10 +8,12 @@ related-issues:
   - https://github.com/openpi-dev/openpi/issues/118
   - https://github.com/openpi-dev/openpi/issues/193
   - https://github.com/openpi-dev/openpi/issues/196
-related-prs: []
+related-prs: "https://github.com/openpi-dev/openpi/pull/902"
 supersedes: none
 ---
 
+<!-- Related PR link added 2026-11-14; the verification date is unchanged. -->
+
 # OpenPI-owned persisted sidecar compatibility inventory
 
 ## Scope
@@ -60,3 +62,19 @@ Maintainers must decide whether these owner-local rules need a durable,
 repository-level compatibility contract. If not, this inventory and its tests
 remain evidence only; no general registry, migration engine, second Session
 store, orchestration runtime, or user-data migration follows from it.
+
+## Verified facts
+
+The inventoried owners reject or withhold unsupported sidecars according to their own semantics; synthetic tests cover setup configuration, workflow journals and Plan Mode state.
+
+## Inferences
+
+The owner-specific outcomes support the narrow shared invariant stated in the conclusion; they do not establish a global migration policy.
+
+## Recommendations
+
+Keep the owner-local rules and tests as evidence unless a durable compatibility contract is separately adopted.
+
+## Unknowns
+
+Whether the owner-local rules need a repository-level compatibility contract remains a maintainer decision.
diff --git a/docs/research/PI_COMMUNITY_CAPABILITIES_2026-11-07.md b/docs/research/PI_COMMUNITY_CAPABILITIES_2026-11-07.md
@@ -178,3 +178,19 @@ selected host and adds no runtime dependency. The resolver regression failed
 before the fix and passed afterward; the repacked installation reached ready,
 served its local assets and exited with status zero. Development-checkout
 success had not established this distribution boundary.
+
+## Verified facts
+
+Pinned community sources and local experiments exposed useful context and review seams; a diagnostic comparison produced a false negative.
+
+## Inferences
+
+The diagnostic false negative limits what a location-free comparison can establish; the source audit does not validate the packages for production use.
+
+## Recommendations
+
+Evaluate the described native Pi seams within their stated boundaries; further language-server work needs explicit availability and termination handling.
+
+## Unknowns
+
+Diagnostic identity, server availability and clean termination remain unresolved at the experiment boundary. Model quality and performance were not measured.
diff --git a/docs/research/README.md b/docs/research/README.md
@@ -23,6 +23,8 @@ Research records preserve sourced investigation and distinguish observations, in
 
 ## Validated investigations
 
+- [`WEB_EVIDENCE_TERMINAL_STRINGS_2026-10-07.md`](WEB_EVIDENCE_TERMINAL_STRINGS_2026-10-07.md) — bounded terminal sanitizer source and regression evidence ([#700](https://github.com/openpi-dev/openpi/issues/700), [PR #705](https://github.com/openpi-dev/openpi/pull/705)).
+
 - [`PI_COMMUNITY_CAPABILITIES_2026-11-07.md`](PI_COMMUNITY_CAPABILITIES_2026-11-07.md) — pinned community audit, local context/review validation, and semantic-navigation probes with a diagnostic-delta false negative; no model-quality or release claim ([#913](https://github.com/openpi-dev/openpi/issues/913)).
 - [`PERSISTED_SIDECAR_COMPATIBILITY_2026-10-24.md`](PERSISTED_SIDECAR_COMPATIBILITY_2026-10-24.md) — P0 source-pinned inventory and synthetic decoder evidence for versioned OpenPI-owned sidecars; preserves Pi Session ownership and introduces no migration framework ([#901](https://github.com/openpi-dev/openpi/issues/901)).
 - [`PERFORMANCE_AUDIT_2026-10-23.md`](PERFORMANCE_AUDIT_2026-10-23.md) — 九类性能修复及本地验证：子代理结果保留、Workflow checkpoint、目录扫描与大 Markdown；原生 TUI 视觉验收由用户确认（[#841](https://github.com/openpi-dev/openpi/issues/841)）。

diff --git a/docs/research/SEMANTIC_NAVIGATION_2026-11-07.md b/docs/research/SEMANTIC_NAVIGATION_2026-11-07.md
@@ -5,10 +5,12 @@ last-verified: 2026-11-07
 applies-to: OpenPI baseline 1120cafcd3f5c1ad466df1c7b823976a3e911f72 and the semantic-navigation candidate described below; Pi SDK 1.2.0; TypeScript 7.0.2 with effect-tsgo 0.24.3
 related-issues:
   - https://github.com/openpi-dev/openpi/issues/913
-related-prs: []
+related-prs: "https://github.com/openpi-dev/openpi/pull/916"
 supersedes: none
 ---
 
+<!-- Related PR link added 2026-11-14; the verification date is unchanged. -->
+
 # On-demand semantic navigation and diagnostic observations
 
 This record preserves source-scoped observations and a candidate implementation, not a formal Benchmark, a release claim or a new architectural Decision.
@@ -62,3 +64,19 @@ An actual `npm pack --ignore-scripts` tarball included the Skill and adjacent sc
 Raw receipts, exact requests/results, controlled source copies, cancellation and package artifacts remain outside Git under `/Users/tsk/.codex/experiments/openpi-community-2026-11-07/semantic-delivery-v1/`. Its manifest records file hashes. The checked CLI hash is `fb4effecd87115d985d08cf35f5755860f9af68a9b0d4f97914e558b9483f1ba`. Original exploratory counterexamples remain under the sibling `semantic/` evidence directory.
 
 Repository checks, full test totals, independent review and final CI are recorded in the implementation PR once complete. These functional observations are not a formal Benchmark: there is no representative task suite, model-quality experiment or usage comparison.
+
+## Verified facts
+
+The candidate resolved two navigation examples and retained ambiguous diagnostic occurrences in controlled single-document cases.
+
+## Inferences
+
+Matching diagnostic signatures in changed text can be ambiguous; a response alone does not establish clean server termination.
+
+## Recommendations
+
+Retain ambiguity and incomplete results explicitly, and keep execution within the existing Pi and Bash boundaries described above.
+
+## Unknowns
+
+The experiment does not establish project-wide diagnostic coverage, general language-server compatibility, latency or token efficiency.
diff --git a/docs/research/WEB_EVIDENCE_TERMINAL_STRINGS_2026-10-07.md b/docs/research/WEB_EVIDENCE_TERMINAL_STRINGS_2026-10-07.md
@@ -1,3 +1,15 @@
+---
+status: validated
+created: 2026-10-07
+last-verified: 2026-10-07
+applies-to: "origin/main 3cb2ecfe1bfbb98252651885311d309f83749428 and the bounded regression evidence below"
+related-issues: "#700"
+related-prs: "#705"
+supersedes: none
+---
+
+<!-- Metadata added 2026-11-14; the original verification boundary is unchanged. -->
+
 # Terminal strings in Web evidence
 
 - Status: validated source and regression evidence
@@ -16,3 +28,15 @@ Eight evidence tests pass. Unicode and tab behavior follows the shared sanitizer
 
 Full repository validation results and CI are recorded on the linked PR. Local
 Windows failures are retained separately and are not reported as passing.
+
+## Inferences
+
+The regression evidence supports the stated sanitizer boundary; it does not alter canonical execution or persisted evidence.
+
+## Recommendations
+
+Use the shared terminal sanitizer for the bounded Web evidence candidate, as described above.
+
+## Unknowns
+
+Local Windows failures are retained separately; their outcomes are not claimed as passing here.
diff --git a/docs/research/WEB_LIFECYCLE_AND_CHILD_HISTORY_2026-10-22.md b/docs/research/WEB_LIFECYCLE_AND_CHILD_HISTORY_2026-10-22.md
@@ -1,3 +1,15 @@
+---
+status: validated
+created: 2026-10-22
+last-verified: 2026-10-22
+applies-to: "ea4dfb993793ccdb093229588b6ccd3467afd90a, PR #892 and the separately identified follow-ups below"
+related-issues: "#890, #891"
+related-prs: "#892"
+supersedes: none
+---
+
+<!-- Metadata added 2026-11-14; the original verification boundary is unchanged. -->
+
 # Web lifecycle and saved child history
 
 - Status: validated for the named source and local fixture boundaries below
@@ -134,3 +146,19 @@ The complete isolated `bun run test` passed **2,382 Node tests / 8 skipped / 0 f
 ## Limits and interpretation
 
 This evidence demonstrates source behavior, native fixtures and an isolated browser. It does not claim a real computer reboot test, deployment to the user's installed runtime, automatic execution recovery, or a comparative performance improvement. The installed OpenPI source remains separate from this checkout. Lists and transcript previews retain explicit bounds; deleted files, missing private bindings and historical in-memory data remain unavailable rather than reconstructed by title or working-directory guesses.
+
+## Verified facts
+
+The Web Host owns execution independently of browser subscriptions; persisted Session files do not resume a stopped execution loop.
+
+## Inferences
+
+The historical incident evidence establishes the reproduced source failure path, not the exact action that initiated the incident.
+
+## Recommendations
+
+Keep execution ownership separate from browser subscriptions and preserve the bounded Session identity and recovery rules described above.
+
+## Unknowns
+
+Real reboot, deployment to the installed runtime and recovery of unavailable historical in-memory data were not established.
diff --git a/docs/research/WEB_NATIVE_CONTINUATION_CONTROL_2026-10-21.md b/docs/research/WEB_NATIVE_CONTINUATION_CONTROL_2026-10-21.md
@@ -1,3 +1,15 @@
+---
+status: validated
+created: 2026-10-21
+last-verified: 2026-10-21
+applies-to: "PR #865 extending db3d40ed93439f30c148b03c9a3f2c03827cc098, at the local boundaries below"
+related-issues: "#867"
+related-prs: "#865"
+supersedes: none
+---
+
+<!-- Metadata added 2026-11-14; the original verification boundary is unchanged. -->
+
 # Web control after a native continuation
 
 - Status: validated at the local source, Pi fixture, Host and rendered-UI boundaries below; not a new architectural Decision.
@@ -30,3 +42,19 @@ The transcript withholds only the active turn's change summary until execution s
 - A read-only synthetic preview using frozen before/after Web bundles reproduced the early card, verified its absence during execution and compaction, and verified the final expandable file list. Screens were reviewed at 1440, 768 and 390 pixels. The app inspector allows the existing 4% composer shadow; sparse-chat whitespace is reserved conversation space, not a new fixed-height card.
 - Local evidence: `~/.codex/evidence/openpi-browser-continuation-20261021/`, including check/test logs, the cancellation ablation, frozen bundles, screenshots and `comparison.html`. Private Session content and credentials are excluded from this record and the repository.
 - The running user service was not updated or restarted. Native lifecycle tests use Pi with a deterministic provider; no claim is made about a new live website login, native browser action or model-provider smoke.
+
+## Verified facts
+
+A native continuation has a fresh command identity without a fresh HTTP admission; local regressions cover its origin and cancellation boundaries.
+
+## Inferences
+
+A native continuation needs its exact originating admission; selecting a recent visible controller would not establish that authority.
+
+## Recommendations
+
+Preserve the exact invocation and cancellation-revocation boundary described above.
+
+## Unknowns
+
+A new live website login, native browser action and model-provider smoke were not established; the running user service was not updated.