Specify Docker-bracketed release conveyor
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add REQ-039: release gate requires VM-creation + runner-registration proof
Per ADR-0006 (revised after 3-model consensus review). Planned until
#110 (Windows register-runner live-proof) completes and the gate's
actual release.yml implementation lands.
Promote REQ-038 to Verified (live): Ubuntu register-runner real workflow_dispatch proof
Real workflow_dispatch proof against a genuine, freshly-built disposable
VM (register-runner-proof-ubuntu-20261107, upstream run 42403330593) --
the registered runner independently confirmed online via gh api with
exactly the expected labels, not merely the phase's own exit code.
Three real bugs found and fixed during this live-dispatch process:
(1) missing SSH_ASKPASS setup (every ssh/scp call failed with Permission
denied), (2) gh api --jq doesn't support jq's own --arg flag, (3) a
re-registration ordering bug (config.sh remove requires the systemd
service to be stopped/uninstalled first). Each merged as its own PR
(#107, #108, #109).
Disposable test runner deregistered and the temporary
RUNNER_REGISTRATION_PAT secret removed immediately after.
Add REQ-038: Ubuntu register-runner phase parity with Windows
Implementation merged via VITechnologies/projectbuddy#106 (issue #105).
Marked 'implementation complete; live verification pending' until a real
workflow_dispatch produces a runner independently confirmed online via
gh api (dispatch in progress: a fresh disposable VM build, upstream run
42400820191, is running specifically for this purpose).
- SRS REQ-038 added
- RTM row added
- Architecture Packet Component View row added (install-actions-runner.sh,
complete-runner-registration.sh)
Promote REQ-037 to Active: Decode-LbaCapture.py implemented and verified
- scripts/Decode-LbaCapture.py implemented per #102, 23 synthetic-fixture
pytest tests (99% line coverage), real verification run against the
genuine run-1794072448721 capture (reproduces the 31.724 MB/s peak-write
finding at frame 432, reports zero false anomalies on this healthy
capture per ADR-0005's Addendum correction)
- Architecture Packet: Container View row for 'Agent diagnostics' no
longer 'planned'; added matching Component View row
- RTM REQ-037 row updated to Verified (live)
Correct REQ-037: dhashHex/screen null is structurally expected, not an anomaly
Matches ADR-0005's own Addendum, written after inspecting the real
labview-benchmark-actor source (local clone at the matching v1.4.14
version): the live 'Capture LabVIEW Launch' command never computes a
perceptual hash or emits screen dimensions, so null dhashHex/screen is
normal for every real capture it produces, not a settle/completeness
signal. Corrected anomaly flags: capture-status.json state of
failed/capturing (not stopped), dualPacket.outcome != authoritative, and
inconsistent (not uniform) dhashHex population across one capture.
Add ADR-0005: agentic decoding of labview-benchmark-actor capture output
- REQ-037 (Planned): dedicated read-only decoding tool for
labview-benchmark-actor capture output, agent/operator diagnostic use
only, schema-pinned to labview-benchmark-actor/launch-capture@1
- RTM row for REQ-037, tracked via #102
- Architecture Packet: ADR-0005 index entry + planned Container View row
for 'Agent diagnostics'
Grounded in this session's real capture analysis (run-1794072448721):
found the MCP viewer tools show fixed demo data regardless of new
captures, no upstream 'latest capture' pointer exists, resources.disk is
a flat array (richer per-disk/per-frame data lives in the top-level
frames[] array instead), and all 631 frames had a null dhashHex.
Correct REQ-018/REQ-032: no separate LabVIEW activation exists on Ubuntu/Linux
Real workflow_dispatch proof (#62, run 42394542555) found this engine's
Ubuntu-LabVIEW appliance has no separate human-interactive LabVIEW
activation step at all -- NI's Linux LabVIEW distribution's authenticated
package-repository access is itself the licensing gate. VIPM activation
(automated via vipm activate --direct-vipm-io) is the only activation
the complete-activation phase performs or needs.
REQ-018's and REQ-032's prior verification relied solely on
complete-activation.sh's own usage text claiming a prior manual LabVIEW
activation precondition -- an unverified assumption carried over from
the Windows family, never actually true for Linux, now corrected with
real receipt evidence (perpetual, valid VIPM activation; a passing
LabVIEW known-answer probe) rather than code review alone.
Refs VITechnologies/projectbuddy#62
Record #94 verification review findings and #97 fix-round re-verification
REQ-034/REQ-035/REQ-036: append fresh upstream re-verification evidence
after #97's fix round (missing engineCommit, report-completeness gap,
suite-commit validation, Linux VIPM-detection paths, installMethod
manifest value, ADR-0004 waiver authorization, Xvfb documentation
overclaim). Fresh runs reproduce the same PASS results as before the fix,
now also recording engineCommit correctly.
Refs VITechnologies/projectbuddy#94, #95, #96, #97
Update REQ-035/036: Phase 2 Linux LUnit gate confirmed feasible, live-dispatch-proven on fork
REQ-035: feasibility spike outcome (a) confirmed -- no Xvfb/DISPLAY
needed, no NSIS-equivalent installer needed for Linux.
REQ-036: now fully self-contained with its own three-waiver list
(the shared Windows waiver plus two new Linux-specific ones: Test
Window Position.vi, Test INI Settings.vi), rather than cross-
referencing REQ-034. Confirmed via direct local Docker testing and a
real workflow_dispatch run on the fork (run 42372457595): exact match,
10 tests, 3 named waivers, vipmPresence: false.
Refs VITechnologies/projectbuddy#87
Correct REQ-033/034/035/036 per round-2 verification review of ADR-0003
Second 3-model review (Claude Opus, GPT-6-astra, Gemini-4-argon) of the
revised ADR-0003 found, despite genuine fixes, several residual defects:
a lingering 'already proven upstream' contradiction, an over-broad waiver
clause, REQ-035/036 still not self-contained, and REQ-033's 'pinned-commit'
ambiguity between this engine's release commit and the external suite's
commit. Fixes all of it: REQ-035 restated as an investigative
determine-whether requirement (outcome (a) or (b), neither release-gating
by itself), REQ-036 fully self-contained (no 'same as REQ-034'
cross-reference), waiver narrowed to the single named upstream test only,
and REQ-033 explicitly disambiguates the two distinct commit identities.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Correct SRS/Architecture Packet per ADR-0003's pre-implementation consensus review
Rewrites REQ-033/034/035, adds REQ-036 (Linux feasibility spike split from
execution), to fix: embedded implementation detail in requirement text,
REQ-035's non-self-contained dependency on REQ-034, the premature
'verified-working' claim, and the undecided runner-class gap -- all
independently flagged by 3 models (Claude Opus, GPT-6-astra,
Gemini-4-argon) reviewing ADR-0002 before any implementation began.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add REQ-033/034/035: ephemeral-container LUnit release gate (planned, ADR-0002)
Spec-first per this wiki's own Constraints section: requirements added
before the corresponding code change (tracked in #87) is merged. Also
adds a planned Container View entry for the new capability.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Operator-resolve wiki-side findings from all 9 milestone issues
- REQ-001: formally scoped to exclude 2 disclosed narrow-purpose
exceptions; marked Verified (manual), no longer NOT met.
- REQ-003/004/005/021/024: reworded to accurately scope real guarantees.
- REQ-010/REQ-013: fixed sibling-location drift (SRS vs RTM).
- REQ-018/REQ-019: merged (REQ-019's own Fit Criterion read 'Same
evidence as REQ-018', failing 29148 Necessary); REQ-019 formally
retired, every cross-reference updated.
- Added Notation + Glossary sections to the SRS.
- Architecture Packet: added 7 missing components + 2 new containers
(Release automation, Agent orientation); fixed Context View staleness
(feature/* branch claim, SHA-pinning overclaim); reframed the
validate-vm-provisioning-dispatch.py evidence citation to not rely on
a single ephemeral run ID, contradicting the SRS's own stated
principle; disclosed vm-fleet-status.yml's distinct fork-specific
failure mode; fixed the renamed projectbuddy/* schema namespace
reference; fixed the Test-SbomArtifact.ps1 'only unit-tested code'
stale claim.
- Home.md + cm-plan.md (main repo) + SRS Constraints: corrected the
wiki's own 'branch-then-push, reviewed' overclaim -- real git history
shows zero branches were ever used; disclosed honestly.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fix round-6 semantic-audit findings in wiki (Architecture Packet, SRS, RTM)
- Deployment View: reframe the validate-vm-provisioning-dispatch.py
restoration evidence to cite the script's own reproducible code/logic
as primary evidence, not a single historical run ID -- this SRS's own
evidence-citation principle explicitly says a run ID 'is ephemeral and
does not, by itself, prove anything about the current state of the
code', which round 5's own fix had directly contradicted.
- Deployment View: disclose vm-fleet-status.yml's distinct fork-specific
failure mode (no vm-provisioner runner registered, so scheduled runs
queue and are cancelled), separate from upstream's clean no-op.
- Document Control: fix stale 'Initial draft... v0.3.10'/'Draft' status;
fix 2 references to a v0.3.11 tag that was never actually released.
- REQ-010 (SRS): reflect the v0.3.28 exception already in its RTM row.
- REQ-013 (RTM): cite the actually-invoked build-virtualbox.sh instead
of its never-called .ps1 sibling, matching its own SRS row.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fix round-5 semantic-audit findings in wiki (Architecture Packet, SRS, RTM)
- Deployment View: fix self-contradiction with round-4's own code fix
(said validate-vm-provisioning-dispatch.py 'does not exist' in the
same round that restored it); cite a real, first-ever successful
validate-phase dispatch as positive evidence.
- Component View: fix the shared VIPM-exit-codes/schema validation row
(Linux container pipeline does its own separate inline check, not
pipeline-config.schema.json); fix the build-virtualbox no-default row.
- REQ-001: stop asserting 'Verified (manual)' against a Fit Criterion
now known false (#68). REQ-002: remove the stale 'ISO paths, etc.'
no-default overclaim (round 4 fixed this in README/Architecture
Packet but not here). REQ-010: reflect the one disclosed v0.3.28
GitFlow exception. REQ-013: cite the actually-invoked
build-virtualbox.sh instead of the orphaned .ps1 sibling.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fix round-4 semantic-audit findings in wiki (Architecture Packet, SRS)
- Deployment View: all Ubuntu phases were unreachable, not just
package-ready (#62's full scope); VIPM activation is not actually
LabVIEW-gated at runtime (REQ-032 correction propagated here).
- Component View: Pipeline.Stages.ps1 has 6 stage functions, not 8,
and the policy gate is a helper inside Publish-DependencyTrack, not
a preceding stage; no package-ready receipt verifier exists;
pipeline-config.schema.json is not validated against by the Linux
container pipeline; build-virtualbox.ps1/.sh no-default claim
corrected; added 6 previously-undocumented components.
- REQ-010: Fit Criterion now accounts for the no-op merge-back
exception.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fix round-3 semantic-audit findings (REQ-007/032, Architecture Packet)
A verification-focused re-run of this session's multi-model semantic
audit found real residual gaps from the previous round's wiki fixes:
- REQ-032: corrected a false claim that VIPM's automated activation is
"gated on the operator having already completed LabVIEW's own
interactive activation" -- capture-activation.sh runs VIPM activation
unconditionally whenever a VIPM env file is supplied, with no check of
LabVIEW's actual activation state. The precondition is a documented
operator convention (the phase's usage text), not a runtime-enforced
gate. Also corrected the real call chain: activate-vipm.sh is invoked
by capture-activation.sh, not directly by complete-activation.sh.
- REQ-007: still listed Run-Pipeline.sh as a "native SBOM pipeline
script" with Dependency-Track fail-closed behavior it doesn't
implement (it's a separate, unrelated LabVIEWCLI tool -- the same
class of gap already fixed in README.md/docs/ in the previous round,
just missed in this specific SRS row).
- Architecture Packet Container/Component Views: still described
Run-Pipeline.ps1/.sh together as the "Native SBOM pipeline" -- split
into two rows, with Run-Pipeline.sh correctly described as the
unrelated tool it actually is.
- Architecture Packet Deployment View: no longer asserts the Ubuntu
package-build/signed-.vip capability is proven, since
provision-ubuntu-vm.yml's package-build phase cannot currently succeed
(VITechnologies/projectbuddy#62, open).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Narrow REQ-013 wording to the base-OS ISO specifically
Found by this session's full 5-area multi-model semantic audit: REQ-013's
"installer/ISO media" wording implied coverage for all attached media,
but the auxiliary NSIS/service-bootstrap bootstrap ISOs
(-NsisBootstrapIso/NSIS_BOOTSTRAP_ISO, SERVICE_BOOTSTRAP_ISO) are only
existence-checked, not SHA-256 hash-verified, unlike the base-OS ISO this
requirement's own Fit Criterion/Verification actually describe. Narrowed
the Requirement/RTM wording to match what's actually verified; the
auxiliary-media gap is tracked separately (VITechnologies/projectbuddy#63).
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Fix semantic-audit findings: REQ-013/017/018/019/025/030 + add REQ-032
Corrects 6 real spec-accuracy gaps found by this session's multi-model
semantic audit (projectbuddy#52), each independently verified against
the real repo source before fixing:
- REQ-013: Fit Criterion/Verification referenced the wrong check
(provision-windows-labview-vm.yml's VIPM-installer hash, not the
Requirement's own claim about build-virtualbox.ps1/.sh's base-OS ISO
hash). Now cites the real, matching check in both scripts.
- REQ-017: broadened Fit Criterion/Verification to the now-complete
guest/WinRM-password-file + PAT-file cleanup (see the matching code
fix in VITechnologies/projectbuddy#52 finding 2).
- REQ-018/REQ-019: narrowed from "LabVIEW/VIPM" to "LabVIEW" -- VIPM
activation on the Ubuntu-LabVIEW family is genuinely automated via
activate-vipm.sh's non-interactive activation call, which contradicted
the broader wording. Fixed REQ-019's broken cross-reference (was
"Same evidence as REQ-017", should be REQ-018) and its Verification
column's inaccurate claim that complete-activation never performs
activation itself (true for LabVIEW, false for VIPM).
- REQ-025: Requirement wording now matches its own Fit Criterion's real
logic (completed failed-gate state, not an in-progress running stage).
- REQ-030: Requirement/Fit Criterion/Verification now describe the real
fix (filter all matching entries before appending, not a first-match
lookup) so duplicate pre-existing entries are actually guaranteed to
collapse to one.
- New REQ-032 documents VIPM's deliberate, narrower automated-activation
exception to REQ-018/REQ-019, so the SRS no longer has an unscoped
LabVIEW/VIPM claim that one real code path contradicts.
Corresponding code fixes shipped in VITechnologies/projectbuddy
(fix/semantic-audit-findings-req017-req030 branch): guest/WinRM
password-file shredding added to every write site in
provision-windows-labview-vm.yml, and add-project.yml's upsert logic
now filters all matching entries instead of a first-match lookup.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Add REQ-030/031 (project registration upsert + Issues-enablement safety); disclose broken validator call chain in Architecture Packet
Found while applying the spec-driven reproducibility audit to project
registration (VITechnologies/projectbuddy#48): add-project.yml and
Add-Project.ps1 both call a credential-reference validator and a
runtime validator that have never existed in the repository, meaning
registration cannot currently succeed for any input. Not something the
SRS can honestly claim as a verified 'shall' (it doesn't work), so
instead:
- Added REQ-030/REQ-031 for the two genuinely-verified parts of the
registration flow this audit also confirmed (name-keyed upsert logic;
the Issues-enablement side effect's never-fail-the-registration
guarantee), each single-shall-verified against the real code.
- Updated the Architecture Packet's Component View to show the real
call graph (Add-Project.ps1 does call a credential validator and a
runtime validator) while pointing at the main repo's new
docs/project-registration/procedure.md for the current-state detail,
matching this wiki's existing split (verified structure/requirements
here; known-gap disclosure in the main repo's procedure docs, same as
#38's handling in docs/vm-provisioning/procedure.md).
Verified locally via the assurance-workbench Docker image before
pushing: all 5 required gates still PASS/High, requirements_quality_check.py
reports zero findings.
Fix #35: resolve stale CI-integration text + remove ephemeral run-ID citations
Closes VITechnologies/projectbuddy#35: the SRS's Assumptions section
pointed readers to the Architecture Packet's Correspondence And
Rationale section for 'the wiring question still open at
ci_integration=not_yet_sure' -- but that section never actually
discussed it. Added the real resolution there (standards-assurance-
scan.yml, merged v0.3.11) and updated the SRS's own text to stop
describing it as open.
Also acts on new operator guidance: workflow run logs/summaries are
ephemeral and must never be cited as 'the evidence' in persistent
documentation. Replaced the one bare workflow-run-ID citation
(REQ-011's Verification/TestArtifact columns, run 42196529270) with a
description of the reproducible mechanism itself (re-runnable via
workflow_dispatch) instead. Updated the RTM's 'Verified (live)' legend
definition to state this principle explicitly, so future additions
follow it without re-litigating the question.
Add real SBOM-pipeline requirements/architecture coverage (VITechnologies/projectbuddy#32)
Closes the same shallow-coverage pattern #29 closed for VM provisioning,
this time for the native + containerized SBOM pipeline (~2900+ lines
across Run-Pipeline.ps1, Build-Sbom.ps1, internal/Pipeline.Stages.ps1,
plus 19 files under docker/linux-sbom/ and docker/windows-sbom/).
- SRS: 8 new requirements (REQ-022 through REQ-029), each single-shall
(verified programmatically), grounded directly in the real code:
the full-mode live-preflight fail-fast behavior, the security-gate
non-disableable guarantee, the lastOccurrence BOM-binding check, the
policy/reporting/build failure-categorization mechanism underlying
Test-SbomArtifact.ps1's exit codes, the path-traversal-hardened
source-checkout stage, the full immutable-release + in-toto attestation
+ tag-chain + manifest cross-match verification chain for LVLIBP
release dependencies, and the deliberate native-vs-container difference
in whether Grype's own findings are a gate.
- RTM: matching Req->Test->Code rows for all 8.
- Architecture Packet: Component View expanded with the real stage
library, shared schemas, and the narrow Build-LvlibpCandidate.ps1
script correctly scoped out of the main pipeline flow.
Verified before committing: single-shall discipline (all 29 total
requirements pass), all 7 ARCH text rules still match, RTM rtm-link
signal still present.
Add real VM-provisioning requirements/architecture coverage (VITechnologies/projectbuddy#29)
Closes the gap the operator flagged directly: VM provisioning had only 2
narrow REQ rows and 3 shallow architecture-table rows despite being ~37
scripts and 1200+ lines of real, dispatch-proven workflow behavior across
2 distinct families (Windows single-VM 5-phase lifecycle, Ubuntu
snapshot-based 10-phase dual-adapter lifecycle).
- SRS: 9 new requirements (REQ-013 through REQ-021), each single-shall
(verified programmatically), each grounded in real, already-proven
behavior read directly from the workflow YAML and its own inline
comments -- fail-closed SHA-256 verification, the triple-fallback
shutdown sequence, real WinRM round-trip readiness polling, credential
hygiene, the deliberate human-only activation boundary, vm-fleet-status's
active-verification design (fixing a real incident where self-reported
success masked a wrong-name runner registration), and the dual
Linux/Windows orchestrator-host adapter model.
- RTM: matching Req->Test->Code rows for all 9.
- Architecture Packet: Component View expanded with the real VM
provisioning components (manifest/inventory schemas, snapshot-receipt
verification chain, NSIS bootstrap installers); Deployment View expanded
with the actual node/lifecycle model for both VM families and the
active-verification pattern.
Verified before committing: single-shall discipline (regex-checked, all
21 total requirements pass), all 7 ARCH text rules still match.
Initial SRS, RTM, and Architecture Packet (spec-driven design)
Satisfies repo-standards-review's REQ and ARCH gates
(VITechnologies/projectbuddy#9, #10). Structured as an Obsidian vault
(.obsidian/ config) using native [[wiki-link]] syntax, which both
Obsidian and GitHub's own wiki renderer (Gollum) understand -- unlike the
main repo, where GitHub doesn't render [[...]] and repo-standards-review's
own lychee-based link checker doesn't support it either (confirmed via
lycheeverse/lychee#1650, open/unresolved).
- Software-Requirements-Specification.md: 12 REQ-### requirements, each
satisfying the single-shall discipline (verified programmatically, not
just by inspection), grounded in real, verified codebase behavior --
each cites its actual verification evidence (live smoke test, code
review, or "no automated check yet", never overstated).
- Requirements-Traceability-Matrix.md: Req -> Test -> Code linkage for
all 12 requirements, following repo-standards-review's
rtm-template.csv column shape.
- Architecture-Packet.md: all 4 required views (Context, Container,
Component, Deployment), named stakeholders and concerns, ADR Index
(cross-referencing ADR-0001, which stays in the main repo's
docs/architecture/adr/ since that's what repo-standards-review's own
ARCH:path:adr rule scans for).
ADR-0001 and the compliance stamp (docs/architecture/standards-compliance.md)
deliberately stay in the main repo, not this wiki -- see Home.md's
Related section.