# Release artifact matrix — RP-00 inventory (T27) > **Reconciled 2026-11-17** against the Phase 0 baseline audit > ([PHASE-0-AUDIT.md](PHASE-0-AUDIT.md), PR #16, branch > `docs/phase-0-baseline-audit`). The audit is the evidence authority > for what exists; [decision 0075](../decisions/0075-release-scope-mvp.md) > is the release-scope authority. > > Authority: [RELEASE-ENGINEERING-PLAN.md](../RELEASE-ENGINEERING-PLAN.md). > Execution ledger: [EXECUTION-BOARD.md](EXECUTION-BOARD.md) — one atomic row > per unit of work, with the evidence and the blockers for each. > > One row per artifact. Each row states **what exists today**, **how it > is produced**, and **its evidence state** — a status word from the > campaign vocabulary plus the receipt that supports it. > > Status vocabulary (campaign §3.2): `NOT STARTED` · `READY` · > `IN PROGRESS` · `BLOCKED` · `IMPLEMENTED` · `BUILD VERIFIED` · > `INSTALLED` · `LAUNCHED` · `SMOKE TESTED` · `PUBLISHED` · `CERTIFIED` · > `DEFERRED` · `REJECTED`. > > **Recipe ≠ built ≠ installed ≠ launched ≠ published ≠ certified.** > Those are separate states and are never collapsed. A row with no > receipt is `IMPLEMENTED` or `BLOCKED`, never `PUBLISHED` or > `CERTIFIED`. ## Reading the matrix - **Status** is the artifact's state on the campaign's 11-rung evidence ladder (audit §3): 1 recipe exists → 11 downloaded published artifact independently verified. - **Receipt** names what was observed, on what host, on what date. A build receipt without a date is not a receipt. - **Blocker** is the verified blocker ID from the audit's register (§4). A blocker ID means the condition was *observed*, not inferred. - The evidence ladder has a hard ceiling at **rung 8** for everything in this repository: no artifact is signed and no provenance/attestation exists (B-14), so **no artifact can reach rung 9**, and therefore **nothing is `CERTIFIED`**. ## Target matrix | # | Target | Format | What exists today | How it is produced | Status | Rung | Receipt | Blocker | |---|---|---|---|---|---|---|---|---| | 1 | Linux x86_64 | tar.gz (desktop) | `firecraft-0.2.5-linux-x86_64.tar.gz` | `xtask dist --native` → `dist/native/` | `PUBLISHED` | 11 | GitHub release v0.2.5 asset, 22,084,726 B. Downloaded and checksum-verified 2026-11-09; `sha256sum -c` passes against the published `MANIFEST.sha256`; unpacks to `firecraft-0.2.5/bin/{firecraft,firecraft-cli}` + `share/licenses/LICENSE`. **The workflow's own build path now produces this artifact**: the `v0.2.5-dryrun.6` rehearsal published `firecraft-0.2.5-linux-x86_64.tar.gz` (24,389,908 B, sha256 `cb55a525…`) on **draft** 449279892 — a draft is not a publication, so those bytes are evidence of a working build, not of a published asset; they are also **byte-identical across two independent rehearsals** (dryrun.5 and dryrun.6) | B-18 (a rebuild of the current tree does not reproduce the *published* bytes — the rehearsal's tarball is 2,305,182 B larger), B-21 (published out of band, not by the release workflow) | | 2 | Linux x86_64 | tar.gz (CLI) | `firecraft-cli-0.2.5-linux-x86_64.tar.gz` | `xtask dist --native` → `dist/native/` | `PUBLISHED` | 11 | GitHub release v0.2.5 asset, 8,072,377 B. Downloaded and checksum-verified 2026-11-09; the extracted `firecraft-cli` **launches** and prints its usage. **The workflow's own build path now produces this artifact, and it attests its own build**: the `v0.2.5-dryrun.6` rehearsal published `firecraft-cli-0.2.5-linux-x86_64.tar.gz` (8,158,124 B, sha256 `953633e4…`) on **draft** 449279892; the archive was downloaded from that draft, re-hashed locally (the digest matches) and extracted, and the extracted binary prints `firecraft-cli 0.2.5`. It is byte-identical across two independent rehearsals | B-18, B-21 (the production release's bytes were published out of band) | | 3 | Linux x86_64 | DEB | `firecraft_0.2.5_amd64.deb` | `xtask dist --native` from the pinned `packaging/deb/control`; the gate reads the package back with `dpkg-deb` and fail-closes on the declared libc6 floor against the binaries' max GLIBC symbol version (via `objdump`) and on the presence of the desktop entry and icon in the package contents | `PUBLISHED` | 11 | GitHub release v0.2.5 asset, 14,819,904 B. Downloaded and checksum-verified 2026-11-09; `dpkg-deb --info` reads `Version: 0.2.5`, `Architecture: amd64`; the extracted `firecraft-cli` launches. **Clean-container receipts on the published v0.2.3 package** (Debian 12): install → `PASS: installed firecraft 0.2.3`; CLI document round trip → `firecraft-cli info` prints the document JSON, `firecraft-cli export` writes a 97 B PNG with the correct signature `89504e470d0a1a0a`; GUI launch under Xvfb → exit 124, i.e. the binary started and stayed up for the 25 s launch window | **B-7 — OPEN DEFECT.** The declared glibc floor is below the binaries' real requirement, confirmed on the **published** artifact by `objdump -p`, not only on a local rebuild. The floor numbers and the `objdump` floor logic are owned by the stacked `fix/deb-glibc-floor` PR; this row deliberately does not restate them. A package built from this tree under-declares its dependency, so it installs cleanly and then fails to start on older distributions. **This is a defect, not a settled fact.**
**B-24 — OPEN DEFECT (owner: PR #41).** The published package carries **no desktop integration at all**: `dpkg-deb --contents` on the published v0.2.5 lists only `./usr/bin/firecraft`, `./usr/bin/firecraft-cli` and `./usr/share/doc/firecraft/copyright`. The install verification run against the published v0.2.3 package in a clean Debian 12 container fails at exactly that point — `FAIL: /usr/share/applications/firecraft.desktop is not installed` — so the package installs cleanly and is still not a desktop application. PR #41 adds the desktop entry and the icon to the payload and makes `dist --native` fail when either is absent from the built package The published package's missing desktop entry and icon are also tracked as board T27-22. **The fix is in the tree and the workflow's own build proves it**: the `v0.2.5-dryrun.6` rehearsal published `firecraft_0.2.5_amd64.deb` (16,228,104 B, sha256 `79283696…`) on **draft** 449279892, and the release workflow's package verification ran the repository's eight-stage script against that package in a clean Debian 12 container — **all eight stages passed**, including the desktop entry, the icon and the **upgrade** stage that the distro matrix does not exercise. The gap is therefore between the *published* package and the tree, not inside the tree | B-7 and B-24 remain defects **of the published v0.2.5 package**, which this tree no longer reproduces; they close only when a release is cut from this tree | | 4 | Linux x86_64 | AppImage | recipe only — `packaging/build-appimage.sh` | operator-run on a host with `linuxdeploy`; the script is `bash -n` parse-checked by `dist --native` | `NOT STARTED` | 1 | recipe exists and passes the gate's `bash -n` check; **never executed**. On this branch the recipe gained a preflight that fails when the desktop entry or the icon is missing from the tree, so the "no icon → an image with no icon" silent defect cannot reach `linuxdeploy` (audit §11.6) | B-8 (`linuxdeploy` absent, no pinned copy vendored), B-10 (the script stages `target/release/…` and ignores `CARGO_TARGET_DIR`) B-9 (the icon file and the `Icon=` key are in-tree since `a4d74a0`, but `linuxdeploy` has still never run). | | 5 | Linux x86_64 | Flatpak | recipe only — `packaging/build-flatpak.sh` + `packaging/flatpak/dev.firecraft.FireCraft.yml` (both added on this branch; **no recipe on `main`**) | operator-run on a host with the Flatpak SDK; the script is `bash -n` parse-checked and the manifest is YAML-parse-checked by `dist --native` | `NOT STARTED` | 1 | recipe and manifest exist on this branch and the gate parse-checks them; **never built**. On this branch the recipe no longer passes the unsupported `flatpak run --repo` argument, installs the bundle into a user installation, launches it with `flatpak run --command=firecraft-cli`, removes it and fails if it is still listed; the manifest installs the icon and points at the canonical desktop entry (audit §11.6) | B-11 (`flatpak` and `flatpak-builder` absent) The manifest's module source is `type: dir, path: ..` — the local working tree, not an immutable upstream revision — so a bundle built from it is not reproducible from the tag SHA (board T27-29). | | 6 | Windows x86_64 | portable ZIP | recipe only — `packaging/build-windows.ps1` | operator-run on a Windows host with MSVC | `NOT STARTED` | 1 | recipe exists; the gate **existence-checks** it only — it cannot parse PowerShell on this fleet; **never executed** | B-12 (`pwsh` absent; no MSVC linker on this host). A GNU cross-build and a wine smoke are recorded on the PR #9 branch — preview evidence only, never native certification (campaign §8) The required Windows artifact does not yet exist: the MSVC build, the installer and the clean-Windows-VM install are all outstanding (board T27-36/37/38). | | 7 | Windows x86_64 | Inno Setup installer | recipe only — `packaging/windows/firecraft.iss` | compiled by the PowerShell recipe on a Windows host with Inno Setup 6 | `NOT STARTED` | 1 | recipe exists and is existence-checked by the gate; the `__APPVERSION__` placeholder is substituted by the PowerShell recipe, which has never run; **never built** | B-12 (`ISCC` absent). The PowerShell recipe *skips* the installer when `ISCC` is missing, so a Windows run without Inno Setup ships no installer and says nothing | | 8 | macOS universal | DMG | recipe only — `packaging/build-macos-dmg.sh` | operator-run on macOS: two `apple-darwin` slices → `lipo` → `hdiutil` | `NOT STARTED` | 1 | recipe exists and is `bash -n` parse-checked by the gate; **never executed** | needs macOS + Xcode CLT; `lipo`/`hdiutil` are absent on this Linux host, so the recipe is off-fleet by design | | 9 | Web (WASM) | ZIP bundle | `firecraft-web-0.2.5.zip` | `xtask dist --web` → `dist/web/`, then zipped | `PUBLISHED` | 11 | GitHub release v0.2.5 asset, 7,174,826 B. Downloaded 2026-11-09: 6 entries (bundle `firecraft-web-f9dbe2f14305b2d9{,_bg.wasm}`, `index.html`, `host.html`, `appinfo.json`, inner `MANIFEST.sha256`); `sha256sum -c` passes on all 5 manifest entries | **B-21 / F-16 — the published bundle was never built by the release workflow.** Its entries are stamped 2026-11-03 12:08, six hours before the workflow's own build finished at 18:15:59Z, and its wasm is a different size than the workflow's. No recorded bridge smoke ever ran against the published bytes. **The workflow's own build path now produces this artifact, and the smoke ran against it**: the `v0.2.5-dryrun.6` rehearsal published `firecraft-web-0.2.5.zip` (5,853,417 B, sha256 `898be851…`) on **draft** 449279892, and the `build artifacts` job's log reads `[xtask] dist web: 7 file(s), wasm 15780814 B …, smoke green` and `[xtask] dist --web ok`. The ZIP's **payload is byte-identical** across two independent rehearsals (all seven files hash the same); only the ZIP container's own timestamps differ. Separately, the **live deployment** serves a fourth bundle (`2274ac8c2925b61d`) that matches no release — B-6; the live comparison was **re-verified 2026-11-10** and is recorded in [Live web deployment](#live-web-deployment-b-6-reconciliation) below | | 10 | WebOS | `appinfo.json` + `host.html` | staged into `dist/web/` by `dist --web` from `packaging/webos/` | `xtask dist --web` | `BUILD VERIFIED` | 5 | staged `appinfo.json` reads 0.2.5; the web manifest covers `appinfo.json` and `host.html`; the live site serves the same `appinfo.json` bytes — re-verified 2026-11-10, byte-identical (`7d039c36…`) | device/runtime smoke is `UNVERIFIED` (audit §6); the Pages deploy is manual and currently stale (B-6) | | 11 | CLI binaries | `firecraft-cli` (inside #1 and #2) | `firecraft-cli` | `xtask dist --native` | `PUBLISHED` | 11 | the **published** `firecraft-cli` was downloaded and extracted 2026-11-09 and **launches**, printing its full usage string; the locally rebuilt one does the same | — | | 12 | Source | tar.gz | `firecraft-0.2.5-source.tar.gz` — the canonical name; **not on `main`** | `xtask dist --source` (git archive of HEAD; the verb verifies a clean extraction and runs `cargo metadata --locked` inside it). The verb exists on this branch and on the PR #9 / #12 trees | `SMOKE TESTED` | 8 | the `v0.2.5-dryrun.6` rehearsal published `firecraft-0.2.5-source.tar.gz` (3,176,024 B, sha256 `0cbee184…`) on **draft** 449279892; the archive was downloaded from that draft, re-hashed locally (the digest matches) and extracted on a **second host** (Fedora), where `cargo metadata --locked` succeeds and `cargo build --locked` finished the dev profile in 53.78 s on cargo 1.99.1. The workflow's own step reported `clean-extract + cargo metadata --locked verified`. The archive's only `.claude/` entry is `.allow_config_edits` — the scratch script rehearsal 5 shipped is gone and the archive is 806 B smaller. The naming defect the rehearsal-3 receipt found (the capitalised `FireCraft-0.2.5-source.tar.gz`) is closed | F-9 (the verb is absent from `main`), B-13 (never published to a *release* — a draft is not a publication) | | 13 | SBOM | CycloneDX 1.5 JSON | `firecraft-0.2.5-sbom.cdx.json` — **not on `main`** | `xtask dist --sbom` over the locked dependency graph; the serial is a deterministic UUIDv8 derived from the locked graph. The verb exists on this branch and on the PR #9 / #12 trees | `SMOKE TESTED` | 8 | the `v0.2.5-dryrun.6` rehearsal published `firecraft-0.2.5-sbom.cdx.json` (369,745 B, sha256 `e50516b2…`) on **draft** 449279892; the workflow's step reported `562 components, 562 locked package(s), schema-valid` and the aggregate gate validated the download. The document is validated against the vendored CycloneDX 1.5 schema **before** it is written and its component inventory is compared with `Cargo.lock`'s own `[[package]]` sections ([decision 0119](../decisions/0119-the-sbom-is-validated-against-the-schema-it-claims.md)). Two independent rehearsals' SBOMs are identical in every field except `metadata.timestamp` — same components, same dependency graph, same serial number | F-9 (the verb is absent from `main`), B-13 (never published to a *release*) | | 14 | Metadata | checksums / manifest | `dist/native/MANIFEST.sha256` (3 native artifacts) and `dist/web/MANIFEST.sha256` (5 bundle files) | `xtask dist --native` / `dist --web`; `release.yml` additionally writes a release-wide `RELEASE_SHA256.txt` over 7 assets | `PUBLISHED` | 11 | `MANIFEST.sha256` attached to release v0.2.5 (398 B, 3 native files) and embedded in the web ZIP (532 B). Both checked against the published downloads on 2026-11-09 — 3/3 and 5/5 pass. **The release-wide list now exists and verifies**: the `v0.2.5-dryrun.6` rehearsal published `RELEASE_SHA256.txt` (664 B, sha256 `af054b99…`) on **draft** 449279892; downloaded from the draft, its 7 entries match the API digests of the 7 assets it covers — **7 of 7**. The list names each file explicitly because the globs over-match, and every published asset now carries the lowercase canonical name. The native `MANIFEST.sha256` is byte-identical across two independent rehearsals | B-17 (the published release still carries no release-wide list; the list exists only on the rehearsal draft), B-21 (the manifest and the assets it covers were published out of band, so the list is self-consistent but not independently attested) | | 15 | Metadata | provenance / attestation | **nothing on any branch** | no generator, no `gh attestation` step, no in-toto/SLSA material anywhere in the repository | `NOT STARTED` | 0 | nothing exists | B-14 | | 16 | Metadata | release manifest (JSON) | `dist/release-manifest.json` — the workflow now generates it and publishes it as a release asset | `xtask dist --manifest` (schema `firecraft.release-manifest/1`); `xtask dist --verify-manifest` re-checks it against the files; `release.yml` generates it after the other assets and publishes it as a ninth release asset | `SMOKE TESTED` | 8 | the `v0.2.5-dryrun.6` rehearsal published `release-manifest.json` (3,295 B, sha256 `54b8a231…`) on **draft** 449279892; the aggregate gate validated it against the downloaded assets, and the file was re-hashed locally (the digest matches). **It names its tag and its run**: `tag` `v0.2.5-dryrun.6`, `source_sha` `c396b18afba0a…`, `run_url` `…/actions/runs/44188107435`, `run_id` `44188107435`, `built_at` `2026-11-17T22:31:44Z`, 8 artifact entries. Two gaps had to be closed first — the caller now states the tag and `xtask` fails closed when the stated tag does not name the workspace version, and the run URL is reconstructed from `GITHUB_SERVER_URL`, `GITHUB_REPOSITORY` and `GITHUB_RUN_ID` because GitHub Actions exports no `GITHUB_RUN_URL`. Shape: [decision 0089](../decisions/0089-release-manifest-shape.md) | B-14 (nothing is signed and no provenance exists), and the manifest has never been published to a *release* — the draft is not a publication | | 17 | Metadata | signature (Authenticode / Developer ID / GPG) | **nothing**; no signing identity is configured anywhere | no signing step in any workflow; signing remains an operator ceremony | `NOT STARTED` | 0 | nothing exists | B-14 — no artifact can reach rung 9 until this is resolved, so **no artifact can be `CERTIFIED`** | | 18 | Android arm64 | APK/AAB | feasibility report only — [android-feasibility.md](../research/android-feasibility.md) | not started: no native entrypoint, no NDK/SDK CI image | `DEFERRED` | 0 | the report concludes "feasible with bounded work"; the gate decision is recorded in [decision 0121](../decisions/0121-the-android-gate-defers-the-native-port.md) — defer the port, keep Android experimental | no NDK/SDK image, no physical ARM64 device | | 19 | Linux | RPM | no spec, no recipe | — | `DEFERRED` | 0 | [decision 0070](../decisions/0070-rpm-deferred.md), accepted 2026-11-06, re-validated 2026-11-16 (T27-31: no RPM demand on the issue tracker; Fedora-family users are served by the tar.gz) | decision 0070 is not on `main`; RPM is **not** reopened by [decision 0075](../decisions/0075-release-scope-mvp.md) | | 20 | Linux aarch64 | any | assessment only — [T27-32](../tasks/T27-32-linux-aarch64-assessment.md) | RP-01 assessment (verdict: feasible-later) | `DEFERRED` | 0 | the assessment records the verdict **feasible-later**: the default native build is pure Rust and architecture-agnostic, so the target is a cross compile plus an architecture-parameterized packaging path — but the fleet has no ARM64 host, and emulation would not carry the GL stack, so the artifact would stop at a build receipt | extended scope ([decision 0075](../decisions/0075-release-scope-mvp.md)); the assessment's four conditions | | 21 | Windows ARM64 | any | assessment only — [T27-41](../tasks/T27-41-windows-arm64-assessment.md) | RP-02 assessment (verdict: feasible-later) | `DEFERRED` | 0 | the assessment records the verdict **feasible-later**: the target's standard library installs, the Windows dependency graph is architecture-agnostic, and GitHub's hosted `windows-11-arm` runner can build an ARM64 Windows artifact for this public repository without new hardware — but the link needs the MSVC C runtime from a Windows host, and eframe's `glow` backend draws through Microsoft's Direct3D-backed OpenGL compatibility pack, which a GPU-less runner cannot prove, so the artifact would stop at a build receipt | extended scope ([decision 0075](../decisions/0075-release-scope-mvp.md)); the assessment's four conditions | ### Status counts | Status | Count | Artifacts | |---|---|---| | `NOT STARTED` | 7 | AppImage, Flatpak, Windows ZIP, Windows installer, macOS DMG, provenance/attestation, signature | | `READY` | 0 | — | | `IN PROGRESS` | 0 | — | | `BLOCKED` | 0 | — | | `IMPLEMENTED` | 0 | — | | `BUILD VERIFIED` | 1 | WebOS package | | `INSTALLED` | 0 | — | | `LAUNCHED` | 0 | — | | `SMOKE TESTED` | 3 | source archive, SBOM, release manifest JSON | | `PUBLISHED` | 6 | desktop tar.gz, CLI tar.gz, DEB, web bundle, CLI binaries, checksums/manifest | | `CERTIFIED` | 0 | — | | `DEFERRED` | 4 | Android arm64, Linux RPM, Linux aarch64, Windows ARM64 | | `REJECTED` | 0 | — | **No artifact has reached `CERTIFIED`.** Six artifacts (desktop tar.gz, CLI tar.gz, DEB, web bundle, CLI binaries, checksums/manifest) reach **rung 11** — they were downloaded from the published release and verified against their published checksum lists on 2026-11-09, and the CLI binary was launched. Nothing has reached rung 9: no artifact is signed and no provenance exists. **The rehearsal path reaches rung 8 and no further.** The `v0.2.5-dryrun.6` rehearsal (run 44188107435, draft 449279892) produced all nine artifacts, ran the web bridge smoke green, passed the DEB's eight-stage package verification in a clean Debian 12 container, and had three of its assets independently re-verified from the draft (7/7 checksum entries, 3/3 re-hashes, one clean build on a second host). A **draft is not a publication**, so those bytes stop at rung 8 until a release is cut from the tag. The receipt is [RECEIPT-0.2.5-DRYRUN-6.md](RECEIPT-0.2.5-DRYRUN-6.md); the scope that this evidence can honestly certify is in [CERTIFICATION.md](CERTIFICATION.md). ## Linux platform compatibility **What is tested on which Linux distribution is published separately**, in [LINUX-COMPAT.md](LINUX-COMPAT.md) — it is the authoritative table for "does this work on my machine", and it lists the untested combinations explicitly rather than leaving them to be inferred. As of 2026-11-17 the matrix has been run across three distro families and two display servers (Ubuntu 24.04, Debian 12, Fedora 45 × X11 and Wayland); all six cells pass, and the DEB's eight-stage package verification passes on both Debian-family cells. The AppImage and Flatpak rows above remain **never built**, and the RPM row remains **deferred** — the compat table says so in the same words. ## Build surfaces (what actually exists) - `xtask dist --native` — pinned Linux x86_64: builds `--locked`, stages, produces desktop tar.gz, CLI tar.gz and DEB (read back with `dpkg-deb --info`/`--contents`), writes `MANIFEST.sha256`, and parse-checks the packaging recipes. - `xtask dist --web` — trunk wasm build, wasm size ceiling, manifest, and the headless bridge smoke. The smoke has no skip switch: a missing browser or a failed round trip fails the dist run (B-5 is remediated). - `xtask dist --source` and `xtask dist --sbom` — **on this branch and the PR #9 / #12 trees; not on `main`** (F-9). Both are wired into `release.yml`. - `.github/workflows/release.yml` — tag-driven: native + web + source + SBOM artifacts, a release-wide `RELEASE_SHA256.txt`, a fail-closed eight-asset aggregate validation, and a **draft** GitHub release whose assets are passed positionally. The Cloudflare Pages deploy step is **commented out** — deployment is manual. - `.github/workflows/ci.yml` — backup CI (`fmt` + `gate`). The job's `Web bundle smoke` step is **tag-gated** (`if: startsWith(github.ref, 'refs/tags/v')`) and is therefore **skipped on branch pushes**. - `.gitforge.yml` — the eight-job pipeline; its `release` job builds the web bundle **with** the bridge smoke, in the headless chromium the CI image bakes (B-5 remediated). - Recipes (operator-owned, off-fleet): `build-windows.ps1`, `build-macos-dmg.sh`, `build-appimage.sh`, `build-flatpak.sh`. ## Live web deployment (B-6 reconciliation) **The deployed bundle is not the released bundle.** They are different bytes, and the difference is recorded here rather than left to be rediscovered at release time. ### What the live site serves (re-verified 2026-11-10) Fetched from `https://firecraft-web.pages.dev/` on 2026-11-10 and checked against the site's **own** `MANIFEST.sha256`: | File | Live sha256 | Size | Manifest verdict | |---|---|---|---| | `appinfo.json` | `7d039c36…` | 250 B | **OK** | | `firecraft-web-2274ac8c2925b61d.js` | `3b0403fc…` | 86,028 B | **OK** | | `firecraft-web-2274ac8c2925b61d_bg.wasm` | `f4f0a3f0…` | 21,231,377 B | **OK** | | `host.html` | expected `379720cc…` | 2,209 B served | **FAILED** | | `index.html` | expected `2617be68…` | 2,409 B served | **FAILED** | The two HTML failures are **not** corruption: Cloudflare Pages injects its analytics beacon (`static.cloudflareinsights.com/beacon.min.js`) into the served HTML after the manifest was written, so the served bytes can never equal the built bytes. A live `sha256sum -c` therefore fails on those two entries **by construction** — the manifest is only verifiable against the pre-deploy bytes. ### Deployed vs released (same table, two builds) | | Live deployment | Released v0.2.5 ZIP | |---|---|---| | bundle stem | `2274ac8c2925b61d` | `f9dbe2f14305b2d9` | | `.js` sha256 | `3b0403fc…` | `be288850…` | | `_bg.wasm` sha256 | `f4f0a3f0…` | `a8a4fa4f…` | | `_bg.wasm` size | 21,231,377 B | 21,238,451 B (7,074 B larger) | | `host.html` (pre-injection) | `379720cc…` | `379720cc…` — **identical** | | `index.html` (pre-injection) | `2617be68…` | `fb2a5140…` — different | | `appinfo.json` | `7d039c36…` | `7d039c36…` — **identical** | **Relationship: different artifact bytes; source relationship unknown unless established by a build receipt.** The shared `host.html` and `appinfo.json` show the two bundles carry the same WebOS shell files; the differing wasm (7,074 B) and differing `index.html` establish that the artifacts differ, but do not establish different Rust sources. No receipt ties `2274ac8c2925b61d` to any commit SHA, so the deployed build cannot be attributed to a source revision at all — it is an orphan. It is also *not* the CI-built bundle recorded in the audit (21,231,517 B, `cf1bec8315c7bfe5`): the live wasm differs from that by 140 B, so a **fifth** distinct wasm build of nominally the same 0.2.5 release now exists. ### Decision Recorded as [decision 0088](../decisions/0088-web-deployment-source-of-truth.md). **The release is the source of truth; the deployment is redeployed from the release bytes.** The reverse — cutting the release from the deployed bytes — is rejected, for two reasons that are both evidence-backed: 1. The deployed bundle is **unattributable**: no build receipt, no commit SHA, no CI run ties `2274ac8c2925b61d` to any revision. 2. The deployed bytes are **not attestable**: two of the five files the deployment's own manifest covers can never verify against it, so a release cut from those bytes would publish an artifact whose checksum list fails by construction. Consequences, recorded so the next release starts from a known state: - The next web release must be **redeployed** from the release ZIP's bytes, and the redeploy must record the deployed bundle stem and the source SHA in a receipt. Deploying is an operator action (the Pages step in `release.yml` is commented out) and is **not** performed by this repository's automation. - The live `MANIFEST.sha256` must **not** be used as the deployment's verification list. Verification happens against the pre-deploy `dist/web/MANIFEST.sha256`; the live HTML entries are expected to differ. - Until a redeploy happens, the live site is **stale and unattributed**: it is not evidence of any release, and no rung above 5 may be claimed for it. **Status: `BLOCKED`** — the reconciliation is complete and the decision is recorded; the redeploy itself is gated behind explicit maintainer approval (campaign §20 forbids publishing or deploying production artifacts without it). Tracked by issue #81 / task #158, defect **B-6**. ## Known gaps against RP-00 acceptance 1. **Windows/macOS/AppImage/Flatpak outputs are recipe-only.** No artifact of those kinds has ever been produced on a target-native host. The GNU cross-build and wine smoke recorded on the PR #9 branch are preview evidence, never native certification (campaign §8). 2. **The DEB's declared glibc floor is below the binaries' real requirement** — confirmed on the published artifact (B-7). This is an open defect with an owner (the stacked `fix/deb-glibc-floor` PR), not a closed item. The floor numbers and the `objdump` check are that PR's to correct; this matrix deliberately does not restate them. 3. **The published DEB carries no desktop integration** — no desktop entry, no icon (B-24). The install verification run against the published v0.2.3 package fails at the desktop-entry stage in a clean Debian 12 container. PR #41 owns the fix; until that lands and a release is cut from it, the published package installs cleanly and still is not a desktop application. 4. **The live web deployment is not reconciled with a release bundle** (B-6), and its own checksum list cannot verify it — Cloudflare Pages rewrites the two HTML files after the manifest is written, so `sha256sum -c` against the live site fails on those two entries by construction. **Reconciled 2026-11-10** — see [Live web deployment](#live-web-deployment-b-6-reconciliation): the deployed bundle is a different, unattributable build, and the decision is that the site is redeployed from the release bytes. The redeploy itself is gated behind maintainer approval. 5. **The release workflow has never produced a release** (B-4, B-21). All six published releases were operator-published by hand. The workflow's own path now runs end to end and produces **drafts** — `v0.2.5-dryrun.5` and `v0.2.5-dryrun.6`, both green — but a draft is not a release. 6. **The source archive and the SBOM have never been published** (B-13), and their verbs do not exist on `main` (F-9). Both are now published on the rehearsal **drafts** and independently verified there. 7. **No release-wide checksum file has ever been published** (B-17), and every published asset reports a null GitHub digest. The workflow writes and publishes one on its drafts, where it verifies 7 of 7. 8. **Nothing is signed and no provenance exists** (B-14), so nothing can be `CERTIFIED`. 9. **RPM is deferred** by [decision 0070](../decisions/0070-rpm-deferred.md) and is **not** reinstated by [decision 0075](../decisions/0075-release-scope-mvp.md). 9. **The published artifacts are not reproducible** from the current tree (B-18), and the bundler version is not recorded in any build receipt (F-17). Four builds of nominally the same release differ by 42,806 B. ## Toolchain / secrets inventory - Fleet (Linux x86_64, Debian 13): rustup stable, trunk, dpkg-deb, chromium (bridge smoke). Per-artifact detail and the parse-check status of every recipe: [TOOLCHAINS.md](../TOOLCHAINS.md). - Off-fleet operator: Windows MSVC + Inno Setup; macOS Xcode CLT + `hdiutil` + `lipo`; `linuxdeploy` for AppImage; Flatpak SDK for Flatpak. None of these hosts exists on this fleet. - Secrets: `CLOUDFLARE_API_TOKEN` — optional, deploy step disabled. Whether it is configured in repository secrets is `UNVERIFIED` (audit §6). No signing credentials are configured anywhere — Windows Authenticode, macOS Developer ID, and GPG/nonce provenance are all absent (RP-06 scope, B-14). ## Scope note This matrix describes **evidence state**, not release scope. Which of these artifacts constitute an acceptable release cut is decided by [decision 0075](../decisions/0075-release-scope-mvp.md), which is **accepted** (Michael Kinney, 2026-11-17): the approved release scope is the MVP candidate scope in that record.