release: v1.2.0 — preview rules now gate (consolidation 2026-06-26)

CHANGELOG.md

@@ -7,10 +7,34 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.2.0] - 2026-06-30
+
+### Fixed
+- **Soundness — preview-maturity rules now gate (and baseline) exactly like stable
+  rules** (wardline-4ada23bb09). The `--fail-on` gate predicate silently excluded
+  `maturity: preview` findings, so six ERROR-severity rules — `PY-WL-118` (SQL
+  injection), `PY-WL-119` (no-op/degenerate trust boundary), `PY-WL-120` (stored
+  taint reaches trusted state), `PY-WL-121` (XXE), `PY-WL-122` (SSTI), and
+  `PY-WL-124` (native-library load) — fired as active ERROR defects but
+  `wardline scan --fail-on ERROR` passed **green** on them (`would_trip_at: null`).
+  `maturity` is now purely informational ("predicates may still sharpen"), matching
+  the long-standing documented contract that preview rules "participate in the gate,
+  baseline, waivers, and judge exactly like stable rules". Preview findings are also
+  now baselineable, so a preview false-positive has the usual escape hatch.
+  **Behavior change:** a repository that scans clean today but contains one of these
+  flows will now correctly **fail** `--fail-on`. Fix the boundary/sink, or reach for
+  an escape hatch — but mind the secure default: the gate evaluates the *unsuppressed*
+  population, so a committed baseline/waiver clears it only under `--trust-suppressions`
+  (a trusted local checkout). In **CI**, scope with `--new-since <merge-base>`; a
+  baselined or waived finding alone will not turn the build green. This applies at
+  every threshold, not just ERROR: the WARN-severity preview rules
+  (`PY-WL-116`/`117`/`123`/`126`) now gate under `--fail-on WARN`, and `PY-WL-125`
+  (INFO) under `--fail-on INFO` — a CI pinned below ERROR is affected too.
+
 ## [1.1.0] - 2026-06-29
 
 ### Changed
-- **BREAKING (unreleased contract):** attest bundle schema bumped `wardline-attest-1` → `wardline-attest-2`; each boundary now carries `content_hash` (entity-body span blake3 binding key, null when unresolved). `wardline-attest-1` bundles no longer verify.
+- **BREAKING:** attest bundle schema bumped `wardline-attest-1` → `wardline-attest-2`; each boundary now carries `content_hash` (entity-body span blake3 binding key, null when unresolved). `wardline-attest-1` bundles no longer verify.
 - Default scan artifacts now anchor to the weft-project root (the `weft.toml` directory)
   rather than the scan cwd, so a subdirectory scan writes to `<project-root>/.wardline/`.
   Retention is therefore project-root-wide across heterogeneous subdir/root scans sharing

UPGRADING.md

@@ -2,6 +2,30 @@
 
 Migration notes for changes that can alter a previously-green run. Newest first.
 
+## To the next release (recommended v1.2) — preview rules now gate (soundness)
+
+`wardline-4ada23bb09`. The `--fail-on` gate previously **ignored** any rule whose
+`maturity` is `preview`, so a scan could pass green while an active ERROR defect
+was present. `maturity` is now purely informational; **preview rules gate (and
+are baselineable) exactly like stable rules**, matching the documented contract.
+
+**Who is affected.** A repository that scans green today but contains one of the
+previously-non-gating preview findings will now correctly **fail**. At
+`--fail-on ERROR`: `PY-WL-118` (SQL injection), `PY-WL-119` (no-op/degenerate
+trust boundary), `PY-WL-120` (stored taint → trusted), `PY-WL-121` (XXE),
+`PY-WL-122` (SSTI), `PY-WL-124` (native-library load). At lower thresholds also
+`PY-WL-116`/`117`/`123`/`126` (WARN) and `PY-WL-125` (INFO).
+
+**What to do.** This is a real finding, not noise — fix it at the boundary/sink.
+If you must defer, mind the secure default: the gate evaluates the *unsuppressed*
+population, so a committed baseline or waiver clears it **only** under
+`--trust-suppressions` (a trusted local checkout), not in a default CI run. In
+**CI**, scope the gate with `--new-since <merge-base>` so it fires only on changed
+code; a baselined/waived finding alone will not turn the build green. (`wardline
+baseline` / the `waiver_add` MCP tool still record the suppression for the
+trusted-checkout and `--new-since` paths.) There is no config flag to restore the
+old "preview never gates" behavior.
+
 ## To v1.0 — Weft config/store consolidation (BREAKING)
 
 Wardline's operator config and machine state moved onto the Weft federation

docs/product/current-state.md

@@ -9,9 +9,9 @@
 `roadmap.md` → Now) — **UNCHANGED**. The open frontier remains the **seam-health probe** —
 PRD-0002 criteria 1 + 2 (Layer-1 `doctor --seams` self-check with a mandatory
 machine-readable `reason`; Layer-2 consumer round-trip that never trusts a self-reported
-status field). *This session did NOT advance the seam bet* — it was an **owner-directed
-examination** of the parked Q4 strategic question (now resolved, below). The seam probe is
-still the highest-blast-radius core unbuilt.
+status field). *The last two sessions did NOT advance the seam bet* — the Q4 examination
+(resolved, below) and this session's reactive **install-friction fix** (PDR-0010) were both
+detours. The seam probe is still the highest-blast-radius core unbuilt.
 
 - *Metric it moves:* **G2-seam** (`metrics.md`): `BASELINE (2026-06-15): 3 of 6 surfaces
   lie or can't self-report → TARGET: 0 of 6 by 2026-07-31`. crit-3 closed the
@@ -52,10 +52,13 @@ No vision change (anti-goal unchanged); the instrumentation was within grant.
 ## Open questions / blocked-on-owner
 
 1. **PR #69 merge + release (owner gate) — SCOPE GREW AGAIN.** PR #69
-   (`release/consolidation-2026-06-26` → main, HEAD now `53a1424d`) carries the prior detour
-   (Part A + pack-bridge + layering `cfe546ed`) **plus** the concurrent session's Part B
+   (`release/consolidation-2026-06-26` → main, HEAD now `8c950e02`) carries the prior detour
+   (Part A + pack-bridge + layering `cfe546ed`), the concurrent session's Part B
    (`652d3bf3`), Part C (`b5170e22`), de-elspeth refactor (`9886280c`), glossary re-sync
-   (`53a1424d`). Merging to main / any PyPI release are outward-facing — your call.
+   (`53a1424d`), the **v1.1.0 release prep** (`adb42a0e` version bump + CHANGELOG, `35401454`),
+   **plus this session's install-friction fix** (`87f13b0d`/`8c950e02`, PDR-0010). The
+   install fix is **unpublished** — it reaches users only via the owner-gated release.
+   Merging to main / any PyPI publish are outward-facing — your call.
 2. **Install the pack in elspeth + relay corrected guidance (cross-repo / owner).** "blake3 will
    NOT fix the gate; install the pack." Pack-bridge + Part C both shipped generically wardline-side.
 3. **3.12 fingerprint release note (owner).** Carry-over from PDR-0006 — one-line note on release.
@@ -68,12 +71,19 @@ No vision change (anti-goal unchanged); the instrumentation was within grant.
 
 ## What this checkpoint did
 
-- **PDR-0009** — Q4 owner decision A+C (hold vision, instrument); metric-bound non-self-sealing
-  reversal trigger (≥ 5; baseline = 1).
-- **metrics.md** — dated G3 reading (inert-gate prevalence; no trigger crossed).
-- **roadmap.md** — added option B to Later as PARKED+gated; Now unchanged.
-- **Reconciled drift** — Part B/C were stale-as-"open" in the prior brief; now recorded DONE+
-  committed; `bd9d1e65cb` recorded CLOSED. Grant re-confirmed 2026-06-29 (date-only).
+- **PDR-0010** — install-friction fix: scan-pipeline extras self-include scanner (the
+  `loomweave` whack-a-mole under `uv tool install`) + a shared install-hint helper naming
+  both installers + a regression guard. Commits `87f13b0d`/`8c950e02` (pushed); also
+  installed the fixed build into the local uv tool (single `[loomweave]` install now keeps
+  scanner — verified).
+- **metrics.md** — dated G4 reading (per-release extras re-check; base stays 0-dep; no
+  trigger crossed).
+- **Tracker** — filed + closed `wardline-c8d7e020e8` (the dogfood install defect).
+- **Cross-repo (not wardline's to own):** this session was primarily the plainweave↔warpline
+  requirements-enrichment federation gate + an exit-code investigation; those product records
+  live in the sibling workspaces (plainweave PDR-017, warpline PDR-0008), not here.
+- The **Now bet was not advanced** (reactive install-fix detour). `roadmap.md` untouched;
+  grant unchanged (re-confirmed 2026-06-29 by the prior session).
 
 ## Where the next session starts
 
@@ -86,5 +96,6 @@ No vision change (anti-goal unchanged); the instrumentation was within grant.
 
 Decisions: `0001` bootstrap, `0002` Now rotation, `0003` doctor seam, `0004` ACCEPT PRD-0001,
 `0005` ACCEPT crit-3 + source-drift CI, `0006` fingerprint determinism, `0007` inert-gate
-visibility, `0008` elspeth pack-bridge, `0009` Q4 hold-vision+instrument. Tactical truth is the
-tracker; intent lives here and in `roadmap.md`.
+visibility, `0008` elspeth pack-bridge, `0009` Q4 hold-vision+instrument, `0010` extras
+self-include scanner (install-friction fix). Tactical truth is the tracker; intent lives here
+and in `roadmap.md`.

docs/product/decisions/0010-extras-self-include-scanner.md

@@ -0,0 +1,64 @@
+# PDR-0010: Scan-pipeline extras self-include scanner (fix uv-tool install-friction)
+
+Date: 2026-06-29
+Status: accepted
+Author: agent:claude (engineering session; recorded at /product-checkpoint)
+Owner sign-off: autonomous within grant — a repo-local, reversible defect fix on the
+install surface (the grant authorizes accept/dispatch of reversible repo-local work).
+Committed + pushed on `release/consolidation-2026-06-26` (PR #69). Reaching users via a
+**PyPI publish is owner-gated** and is NOT performed here.
+Related: `vision.md` thesis invariant #1 (plug-and-play install); G4 (extras/weight
+discipline) + G3 (zero-config activation) in `metrics.md`; tracker `wardline-c8d7e020e8`
+(filed + closed this session); PDR-0007/0008 (the elspeth dogfood that surfaced
+install-surface friction).
+
+## Context
+
+A dogfood report from `~/elspeth` (installed wardline 1.0.7, uv-tool install) hit an
+install whack-a-mole: `uv tool install wardline[loomweave]` *uninstalled* the scanner deps
+(pyyaml/jsonschema/click) and installed only blake3, so `wardline init`/`scan` then errored
+"requires the scanner extra"; reinstalling scanner dropped blake3 again. Root cause: `uv
+tool install` REPLACES the tool env with exactly the named extras (it does not merge), and
+`loomweave = ["blake3>=1.0"]` was not self-sufficient — unlike the `rust` extra, which
+already self-includes `wardline[scanner]` for exactly this reason. This breaks thesis
+invariant #1 ("install it and it stands itself up"): installing a capability extra broke
+the tool.
+
+## Options considered
+
+1. **Guidance only** — tell users to combine extras (`wardline[scanner,loomweave]`).
+   Rejected: pushes the uv-tool gotcha onto every user; the single-extra install (the
+   natural action, and what the doctor hint advised) stays broken.
+2. **Make scan-pipeline extras self-sufficient** (CHOSEN) — `loomweave` self-includes
+   `wardline[scanner]`, mirroring `rust`; a single-extra install carries its prerequisites.
+3. **Collapse loomweave into base/scanner** — rejected: violates G4 (capability stays
+   behind opt-in extras) and the zero-dep base.
+
+## The call
+
+**Make scan-pipeline extras self-sufficient.** `loomweave = ["wardline[scanner]",
+"blake3>=1.0"]` (loomweave's taint-store writes fire only during `wardline scan`, so it
+genuinely needs the pipeline; there is no loomweave-without-scanner path — the CLI refuses
+to start without scanner). Plus a shared `extra_install_hint(extra)` naming both installers
+(`uv tool install` vs `pip install`, since pip targets the wrong env for a uv-tool install)
+across every extra hint, and a regression guard (`test_extras_composition.py`) pinning the
+self-sufficiency invariant. Verified: built-wheel METADATA resolves `loomweave →
+blake3+click+jsonschema+pyyaml`; a single `[loomweave]` install now keeps scanner (no
+whack-a-mole, confirmed live in the uv tool); 256 install+cli tests + layering conformance
+green. Commits `87f13b0d` + `8c950e02`.
+
+## Rationale
+
+Serves invariant #1 directly — the install surface is part of "plug-and-play." The fix is
+the established in-repo pattern (`rust`), keeps the base zero-dep and capability opt-in (G4
+holds), and closes a previously **untested** invariant. Engineering-tracked in
+`wardline-c8d7e020e8`; recorded as a product decision because the install surface is a
+thesis-invariant-#1 concern, not incidental hygiene.
+
+## Reversal trigger
+
+Reopen the bundling decision if a legitimate consumer needs `wardline[loomweave]` WITHOUT
+the scanner pipeline (none exists today), or if the extras-composition guard is ever
+relaxed — watched under G4's per-release extras re-check (`metrics.md`). The
+install-friction class itself reopens if a single-extra `uv tool install` whack-a-moles
+again on any extra after this fix.

docs/product/metrics.md

@@ -151,6 +151,14 @@ gates) enters the tool.
 - `BASELINE → TARGET`: `BASELINE: base = 0 runtime deps at 1.0.6 → TARGET: base
   stays 0 runtime deps; new deps only behind a named extra, re-checked each
   release`
+- **Reading 2026-06-29 (PDR-0010):** per-release extras re-check — base package stays
+  **0 runtime deps** (G4 holds); capability stays behind named extras. Fixed an
+  install-friction defect: the `loomweave` extra was not self-sufficient under `uv tool
+  install` (which replaces, not merges, extras), so a bare `wardline[loomweave]` dropped
+  the scanner deps and broke the CLI (whack-a-mole). Now `loomweave` self-includes
+  `wardline[scanner]` (mirrors `rust`) so single-extra installs carry their prerequisites.
+  **No weight creep** (loomweave still opt-in; base unchanged) and **no G4 trigger
+  crossed**. Pinned by `test_extras_composition.py`; tracker `wardline-c8d7e020e8`.
 
 ## Notes