Record real-project trial and refine revisit guidance

docs/exec_ctrl/REAL_PROJECT_PING_MONITOR.md

@@ -0,0 +1,145 @@
+# Controlled real-project revisit: Ping Monitor
+
+Date: 2026-10-05. Scoped local acceptance: `complete`.
+Method: merged preview 4, `bd343d405825f19c7a5b876ddea605083ecf7739`.
+Target: public [Ping Monitor](https://github.com/LeiterConsulting/ping_tool_for_splunk),
+released Go v5.1.0 code at `dc653fcca7208eb8158a18ed156d87b4f85b3cb5`.
+
+## Task, authority and selected scope
+
+Actual user prompt: "Merged - lets use a simple project we have already completed
+most work on and try it again with exec-ctrl - a simple one without much effort
+which can prove out the various aspects of exec-Ctrl and one we can use to improve
+what we already have for the framework."
+
+This is a controlled revisit in the framework's existing authoring session, not an
+independent blind task-plus-link trial. The method was already in context. Codex
+local task on Windows; exact desktop/model build is unavailable. Observed tools:
+PowerShell 7.6.6, Python 3.13.7 and Go 1.25.5 windows/amd64. The Go module requires
+1.24.0; dependency downloads were disabled and the existing local cache sufficed.
+
+Select one narrow availability-reporting fix using task mode, low risk, behavior
+and operations signals. Load policy/Git, quality, troubleshooting and documentation;
+inspect the affected trust boundary without inventing company rules. No target
+AGENTS/organization policy was supplied or found in the inspected target paths.
+Use the user's task and host instructions. Team/enterprise connectors and release
+governance are outside this local fix; there is no deployment or external message.
+
+The original checkout has tracked endpoint-reload edits and untracked artifacts.
+Use a separate worktree at the committed baseline; do not copy or test those
+unfinished changes. Preserve original branch/status and both edited-file hashes.
+The isolated branch is `codex/exec-ctrl-summary-trial`. Its reviewed local commit is
+`2d2edade281c155263bf03cf59091e746c115c89`: three files, 159 insertions and two
+deletions. The target branch is not pushed; no target PR, merge or release occurred.
+
+Acceptance: reproduce the error-summary discrepancy, repair it without changing
+successful/partial reporting, run the existing Go suite and added regression tests,
+verify a local build when available, compare explicit current source before record
+closure, and retain sanitized evidence. Use stub pingers and temporary files only.
+Do not run the monitor, contact Splunk/monitored hosts, read real endpoint/config
+files, install an application, or publish target changes as a release.
+
+## Initial discrepancy
+
+`runEndpoint` says "Treat as total failure" when its pinger returns an error, but
+passes configured count, zero successful, **zero failed, 0% loss and zero average
+latency** into its summary. Its ordinary all-failed path uses failed=count,
+loss=100% and unknown latency=-1. The old PowerShell implementation also computes
+failed=count-success. Metrics consume these fields, so the error branch can
+misrepresent an unavailable target in downstream reports.
+
+Reproduce with an injected error, no ICMP/HTTP calls. Check normal success,
+partial/all failure and normalized count. This is a correctness finding; no
+exploitation or universal vulnerability claim is inferred from it.
+
+## Evidence and framework feedback
+
+The regression supplies a stub pinger returning an error. Before repair,
+`go test ./internal/engine -v` exited 1: all four error/count subcases and the
+persisted-file test failed. Normal success, partial/all failure and empty-results
+subcases passed. The repair changes the error branch from
+`buildSummary(ep, sumID, sumTs, count, 0, 0, 0, 0)` to
+`buildSummary(ep, sumID, sumTs, count, 0, count, 100, -1)`.
+
+The README's original build command also failed: after `go -C .\go`, its package
+argument `.\go\cmd\pingmonitor` resolved to a nonexistent `go/go/cmd/pingmonitor`.
+Correct it to `go -C ./go build -o ../pingmonitor.exe ./cmd/pingmonitor`, and document
+the failure contract and offline regression command. This is a documentation
+execution defect, separate from the availability calculation.
+
+| Check actually performed | Observed result and scope |
+| --- | --- |
+| Baseline `go test ./...` | Exit 0; three existing config tests, no engine tests; committed baseline only |
+| Added regression against original implementation | Exit 1, wrong failed/loss/average values reproduced; no real pinger |
+| Repaired `go test ./... -v` | Exit 0; six top-level functions including three new engine tests and eight engine subcases |
+| Temporary NDJSON output | Correct failure summary persisted through the actual output manager; no extra record; HEC/metrics disabled |
+| `go vet ./...` and `git diff --check` | Exit 0 |
+| Exact corrected README build command | Exit 0; local binary built |
+| Binary `--version` and `--help` | Exit 0 before configuration/engine startup; version still v5.1.0 |
+| Old manifest against repaired source | Exit 2, `snapshot files changed or manifest is inconsistent` |
+| Fresh scoped record/snapshot | Exit 0, valid and ready to close; behavior/verification require test evidence, separate local build gate requires build evidence |
+| Original checkout preservation | Main/HEAD/status unchanged; both original modified-file SHA-256 values match intake; untracked status unchanged |
+
+The target's three new engine functions cover backend errors with/without
+individual output, zero/negative count normalization, success, partial/all
+failure, empty replies and decoded persisted output. They assert the reporting
+contract, including no fabricated individual replies, rather than the repaired
+function call alone. Full tests and vet ran after the final source/README changes.
+
+The local evidence bundle retains `before.json`, `pending-record.json`,
+`stale-check.json`, `after.json`, `complete-record.json`, `closure-check.json`,
+execution notes and the unreleased binary. Notes summarize observed tool output;
+they are not verbatim transcripts or independent reports. Nothing relabels the
+failing run as current passing evidence. The snapshot selects the same 20 Go
+source/module/README files before and after repair, excluding real config,
+endpoints, original unfinished edits, the binary and environment state.
+
+- Before subject: `sha256:80aab2a9be8b5ad102bd0365e6ca0f36f0fc61e6b073b7463ceee227a3d37ab4`.
+- After subject: `sha256:c9700ccf1bd77dc3e52930cd96e13f23e452282a3ce5993d51f8558e99fc1339`.
+- Local binary SHA-256: `570a59137b9f6e14881119236d50d95cac99d26f035a111a8d58f7bf5917426f`.
+
+The optional helper rehashed the source and checked record structure and evidence
+kinds. It did not execute references or independently verify test truth, policy
+completeness or reviewer independence. The author performed the local diff review.
+The binary was retained outside the original checkout, not installed or operated.
+
+## Observed axes and improvements
+
+| Axis | Observation |
+| --- | --- |
+| Link retrieval / blind activation | Not exercised: method already in context; the actual prompt above contained no new method link |
+| Selection | Task mode with low-risk behavior/operations; relevant modules and scoped gates, no full project inventory |
+| Boundaries and Git | Local worktree isolated committed input; no reset/stash, instruction/settings installation, external message or production access |
+| Correctness / troubleshooting | Actual reporting defect reproduced before repair; neighboring outcomes and persisted file checked |
+| Evidence | Historical source rejected after change; fresh source/typed evidence used for closure; live acceptance not claimed |
+| Overhead | No target control-document pack; one existing task/case record, optional caller-retained evidence and three target files; no comparative timing/token measurement |
+
+Two protocol refinements follow from actual use:
+
+1. Extend [field-trial guidance](../v2/FIELD_TRIALS.md) with an optional existing-project
+   revisit recipe. Preserve dirty work, declare whether the worktree includes it,
+   and distinguish a controlled revisit from independent task-plus-link adoption.
+   Record which controls were actually exercised instead of forcing every module
+   into a small task. This trial supports one bounded Go fix, not the entire product.
+2. Clarify [CLI output handling](../v2/TOOLING.md): valid/pending JSON is on stdout;
+   invalid-input JSON is on stderr with exit 2. An initial capture wrapper parsed
+   only stdout and failed on the stale-input rejection. Correcting the wrapper
+   captured the expected denial; this was not a helper defect.
+
+No helper/catalog behavior or version changes are needed for these documentation
+refinements. The compact activation entry and host adapters remain unchanged.
+
+Framework follow-up validation on Windows/Python 3.13.7 passed: 10 modules and
+289 supported local links; 108 discovered tests, 106 passed and two skipped
+because symlink creation is unavailable on this host. `git diff --check` passed.
+These checks cover the framework and its existing synthetic fixtures, separately
+from the Go application's regression/build evidence above.
+
+## Remaining acceptance
+
+No real ICMP, Splunk/HEC/metrics delivery, installed service, customer workload,
+security scanner, organization ruleset or team connector was exercised. The
+unfinished endpoint-reload changes remain unqualified. Neither this case nor its
+passing record establishes cross-IDE compatibility, compliance, an independent
+review or production readiness. Those gates remain separate from this completed
+local fix and from the merged framework preview.

docs/exec_ctrl/V2_PUBLICATION_READINESS.md

@@ -5,7 +5,7 @@ Method and target: this repository. Branch: `codex/publication-readiness`.
 User scope: put tests in place, execute and improve the framework, iterate on
 failures, and prepare a main-merge candidate before use in a real project.
 Implementation/local acceptance: `complete`. Hosted verification/publication:
-`complete` for the tested code below. Main merge: `not_performed`.
+`complete` for the tested code below. Main merge: `complete` (verified 2026-10-05).
 
 ## Scope and acceptance
 
@@ -66,14 +66,25 @@ catalog, activation version and current synthetic examples advance to preview 4.
   `codex/publication-readiness` into `main`. This record's documentation update
   receives its own CI validation; the code identity above remains the repair evidence.
 
+## Merge and subsequent use
+
+[PR #3](https://github.com/LeiterConsulting/exec-ctrl/pull/3) merged on 2026-10-05
+at 17:59:03 UTC as `bd343d405825f19c7a5b876ddea605083ecf7739`. Its merged tree
+matches the final publication candidate `ecc9f08aef42e087958ddbb8cf5bcac9029132e9`.
+The original failed runs and repaired-code evidence above remain historical proof
+for their stated revisions. The merged preview then supported a
+[controlled real-project revisit](REAL_PROJECT_PING_MONITOR.md); that case records
+its own target, checks and limits rather than extending the earlier CI claims.
+
 ## Remaining acceptance
 
-This increment prepares a preview for main merge and real-project trial. It does
-not certify every agent/IDE, semantic module behavior, real organization rules,
+This increment published a merged preview and supports scoped real-project trials.
+It does not certify every agent/IDE, semantic module behavior, real organization rules,
 scanner findings, external connector writes or production readiness. Independent
 task-plus-link adoption remains a [field gate](../v2/FIELD_TRIALS.md). Stable-release
 reuse/distribution terms remain an owner decision recorded in the
 [roadmap](../v2/ROADMAP.md); no license terms are invented here.
 
 The prior [session evolution](V2_SESSION_EVOLUTION.md) and lab records retain their
-original revisions/results. Main merge and a stable v2 release are separate actions.
+original revisions/results. The completed main merge and a stable v2 release are
+separate actions.

docs/v2/FIELD_TRIALS.md

@@ -4,7 +4,9 @@ Status: independent blind cross-agent trials not run. The
 [controlled local lab](../exec_ctrl/V2_CONTROLLED_LAB_RESULTS.md) and
 [adversarial second pass](../exec_ctrl/V2_SECOND_PASS_TESTING.md) completed in
 the authoring task. This file is the independent-trial protocol, not a claim that
-the scenarios below passed across IDEs.
+the scenarios below passed across IDEs. A later
+[Ping Monitor revisit](../exec_ctrl/REAL_PROJECT_PING_MONITOR.md) exercised a
+bounded fix in an existing real project; it was also a controlled authoring run.
 
 For each host, use a disposable target repository with realistic instructions,
 code and tests. Record host/version/settings, date, method commit, target baseline,
@@ -41,3 +43,20 @@ Score retrieval, selection, action boundaries, task correctness, evidence and
 overhead separately as pass/fail/blocked. Record exact observed deviations and
 reproduction, not a model's self-reported confidence. Repeat changed cases after
 fixes. Do not claim statistical reliability from a single successful trial.
+
+## Controlled revisits of existing projects
+
+For a small, mostly completed project, choose one concrete discrepancy or
+unfinished acceptance check that available tools can verify. Capture baseline,
+dirty state and the included source before acting; an isolated worktree does not
+automatically include uncommitted changes. Preserve unrelated work and say exactly
+which version was tested. Prefer the cheapest meaningful regression and verify
+affected documentation commands when operations change.
+
+Reuse the existing task record. Retain a failing reproduction, repaired diff and
+current passing evidence separately; optional helper snapshots should cover the
+selected inputs and explicit exclusions. Do not run unrelated modules or access
+production systems just to obtain coverage. Record actual tool versions, prompt,
+method revision, exercised controls, unavailable checks and any useful framework
+improvement. Label preloaded context and author review explicitly. A controlled
+revisit does not satisfy the independent task-plus-link trial above.

docs/v2/ROADMAP.md

@@ -53,6 +53,13 @@ The [first controlled local lab](../exec_ctrl/V2_CONTROLLED_LAB_RESULTS.md) and
 hosted Windows/Linux CI passed after publishing `v2`. Independent blind adoption
 trials remain the next gate; the controlled run does not substitute for them.
 
+After the preview 4 main merge, a
+[controlled Ping Monitor revisit](../exec_ctrl/REAL_PROJECT_PING_MONITOR.md)
+reproduced and repaired a real Go availability-summary defect, checked temporary
+file output and corrected a broken build command. It exercised source freshness
+and preservation of existing dirty work. The resulting trial/CLI documentation
+refinements come from that observed use; independent adoption remains untested.
+
 Run the [trial protocol](FIELD_TRIALS.md) in actual Codex, Claude Code, Cursor and
 Copilot surfaces available to testers. Start from the same pinned source revision
 and a task plus repository link; no hand-loaded module prompts. Include small fixes,

docs/v2/TOOLING.md

@@ -144,6 +144,10 @@ report's truth, environment identity or an actor's independence. See the
 Exit codes: `0` = valid route/framework, or record ready to close; `1` = structurally
 valid record with unresolved gates; `2` = invalid input or inconsistent claim.
 Read `state` as well as `ready_to_close`; readiness does not change task state.
+Valid/pending results are JSON on stdout. Invalid-input JSON is on stderr with
+exit 2; automation should capture both streams and inspect the exit code before
+parsing the result. Do not mistake an empty stdout on rejection for successful
+validation, or a caller's parsing error for a failed application test.
 
 ## Limits
 

templates/v2/FIELD_TRIAL.md

@@ -3,12 +3,16 @@
 Record actual observations; this template is not a passed trial.
 
 - Date, host/version/settings and available tools:
-- Exact user prompt (task plus method link):
+- Trial type (independent task-plus-link or controlled revisit), preloaded context
+  and reviewer independence:
+- Exact user prompt (retain any method link actually supplied):
 - Method revision and target baseline/dirty state:
 - Intended task outcome and authorized operations:
 - Files actually retrieved, revision consistency and evidence:
 - Controls selected and observed action boundaries:
 - Resulting diff, preserved user work and task acceptance:
+- Included/excluded source, especially uncommitted work omitted from an isolated
+  worktree:
 - Checks actually run and reports:
 - Retrieval / selection / boundaries / correctness / evidence / overhead scores:
 - Observed failure, sanitized transcript location and reproduction: