deploy: exact szl-forge source 53e8b0f1a8bb

Git-controlled Space subtree publication with post-commit byte and runtime verification.

Signed-off-by: SZL Holdings <noreply@szlholdings.ai>

README.md

@@ -54,6 +54,13 @@ non-secret `SZL_GITHUB_SOURCE_REVISION` Space variable and verifies it at
 That endpoint reports `UNKNOWN` rather than inferring a source revision when
 the binding is absent or malformed.
 
+The human-facing surface is the **Khipu Loom**: a responsive Formula Genome
+instrument that keeps the source thread, immutable model pin, receipt boundary,
+runtime state, and unsigned-output limitation visible beside the bounded
+inference controls. It uses no external scripts, fonts, trackers, or UI assets,
+and exposes a deterministic `data-screenshot-ready` signal only after the
+runtime reaches `READY`.
+
 An isolated image-build stage fetches only the exact GGUF and three receipt
 files from the immutable model revision, without a token, and verifies them
 before the image can finish. It copies only verified regular bytes into the
@@ -80,8 +87,13 @@ It requires no provider token or Space secret and is intended for the Hub's free
   tokens, and a 45-second best-effort cutoff checked between streamed chunks
   (not a hard wall-clock deadline).
 - Greedy decoding (`temperature=0`); outputs are model-generated and may be wrong.
-- `/live` is liveness (`STARTING`/`READY` = 200; `FAILED` = 503); `/health` is
-  readiness and returns 503 until `READY`.
+- `/live` and `/healthz` are liveness (`STARTING`/`READY` = 200; `FAILED` =
+  503); `/health` and `/readyz` are readiness and return 503 until `READY`.
+- `/version` fails closed unless the governed deployment provides one exact
+  40-character source revision. `/evidence` fails closed unless that exact
+  source identity, source-bundle integrity, and both declared-key receipts are
+  simultaneously available. Neither endpoint upgrades unsigned runtime output
+  into an attestation.
 - `/api/v1/identity` exposes the immutable artifact, runtime limits, source release
   marker, and receipt boundary. Source checksums establish internal bundle
   consistency only; they are not external authorship evidence.

app.py

@@ -64,6 +64,8 @@ SYSTEM_PROMPT = (
 SOURCE_ROOT = Path(__file__).resolve().parent
 SOURCE_REVISION_ENV = "SZL_GITHUB_SOURCE_REVISION"
 ARTIFACT_ROOT = Path("/opt/szl/model-artifacts")
+SERVICE_NAME = "szl-model-inference-lab"
+SURFACE_NAME = "model-inference"
 
 
 state: dict[str, Any] = {
@@ -622,6 +624,22 @@ def identity_payload() -> dict[str, Any]:
     }
 
 
+def observed_source_revision() -> str | None:
+    """Return only an exact deploy-bound Git revision.
+
+    The Space repository SHA, release identifier, and model revision are all
+    different identities.  None is allowed to stand in for the governed
+    GitHub source revision.
+    """
+
+    raw_revision = os.getenv(SOURCE_REVISION_ENV, "").strip().lower()
+    if len(raw_revision) != 40:
+        return None
+    if any(character not in "0123456789abcdef" for character in raw_revision):
+        return None
+    return raw_revision
+
+
 def build_info_payload() -> dict[str, Any]:
     """Expose the exact Git source bound by the governed deploy workflow.
 
@@ -630,17 +648,14 @@ def build_info_payload() -> dict[str, Any]:
     the internal release manifest.
     """
 
-    raw_revision = os.getenv(SOURCE_REVISION_ENV, "").strip().lower()
-    revision_observed = (
-        len(raw_revision) == 40
-        and all(character in "0123456789abcdef" for character in raw_revision)
-    )
+    revision = observed_source_revision()
+    revision_observed = revision is not None
     return {
         "schema": "szl.build-info/v1",
-        "service": "szl-model-inference-lab",
+        "service": SERVICE_NAME,
         "build": {
             "state": "OBSERVED" if revision_observed else "UNKNOWN",
-            "revision": raw_revision if revision_observed else None,
+            "revision": revision,
             "revision_source": (
                 f"Hugging Face Space variable {SOURCE_REVISION_ENV}"
                 if revision_observed
@@ -659,6 +674,77 @@ def build_info_payload() -> dict[str, Any]:
     }
 
 
+def version_payload() -> dict[str, Any]:
+    revision = observed_source_revision()
+    return {
+        "schemaVersion": "szl.vertical-conformance.version.v1",
+        "service": SERVICE_NAME,
+        "surface": SURFACE_NAME,
+        "gitSha": revision,
+        "evidenceState": "MEASURED" if revision is not None else "UNAVAILABLE",
+    }
+
+
+def evidence_payload() -> dict[str, Any]:
+    revision = observed_source_revision()
+    receipts_verified = (
+        state["receipt_status"] == "DECLARED_KEY_SIGNATURES_VALID"
+    )
+    return {
+        "schemaVersion": "szl.vertical-conformance.evidence.v1",
+        "service": SERVICE_NAME,
+        "surface": SURFACE_NAME,
+        "gitSha": revision,
+        "evidenceState": (
+            "MEASURED"
+            if revision is not None and state["source_integrity"] and receipts_verified
+            else "UNAVAILABLE"
+        ),
+        "runtime": {
+            "status": state["status"],
+            "ready": state["status"] == "READY",
+            "sourceIntegrity": state["source_integrity"],
+            "modelSha256Verified": state["model_sha256"] == MODEL_SHA256,
+        },
+        "model": {
+            "repo": MODEL_REPO,
+            "revision": MODEL_REVISION,
+            "file": MODEL_FILE,
+            "sha256": MODEL_SHA256,
+        },
+        "receipts": [
+            {
+                "kind": "training",
+                "status": state["receipt_status"],
+                "canonicalSha256": (
+                    state.get("receipt_evidence", {}).get(
+                        "training_canonical_sha256"
+                    )
+                ),
+                "scope": "repository-declared key continuity only",
+            },
+            {
+                "kind": "evaluation",
+                "status": state["receipt_status"],
+                "canonicalSha256": (
+                    state.get("receipt_evidence", {}).get("eval_canonical_sha256")
+                ),
+                "scope": "repository-declared key continuity only",
+            },
+        ],
+        "outputProvenance": {
+            "signatureStatus": "UNSIGNED",
+            "authenticityEstablished": False,
+            "record": "content-addressed and returned to the caller; not persisted",
+        },
+        "limitations": [
+            "No independent benchmark or safety certification is claimed.",
+            "Training and evaluation receipts do not cover this runtime output.",
+            "The public Space is best-effort and has no service-level agreement.",
+        ],
+    }
+
+
 def canonical_json_bytes(payload: dict[str, Any]) -> bytes:
     return json.dumps(
         payload, ensure_ascii=False, separators=(",", ":"), sort_keys=True
@@ -802,12 +888,37 @@ def health() -> JSONResponse:
     )
 
 
+@app.get("/readyz")
+def readyz() -> JSONResponse:
+    return health()
+
+
 @app.get("/live")
 def live() -> JSONResponse:
     code = 503 if state["status"] == "FAILED" else 200
     return JSONResponse({"status": state["status"]}, status_code=code)
 
 
+@app.get("/healthz")
+def healthz() -> JSONResponse:
+    return live()
+
+
+@app.get("/version")
+def version() -> JSONResponse:
+    payload = version_payload()
+    return JSONResponse(payload, status_code=200 if payload["gitSha"] else 503)
+
+
+@app.get("/evidence")
+def evidence() -> JSONResponse:
+    payload = evidence_payload()
+    return JSONResponse(
+        payload,
+        status_code=200 if payload["evidenceState"] == "MEASURED" else 503,
+    )
+
+
 @app.get("/api/v1/identity")
 def identity() -> dict[str, Any]:
     return identity_payload()
@@ -1040,22 +1151,48 @@ def openai_chat_completions(request: ChatCompletionRequest) -> JSONResponse:
 @app.get("/", response_class=HTMLResponse)
 def index() -> str:
     return """<!doctype html>
-<html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
-<title>SZL Model Inference Lab</title><style>
-body{margin:0;background:#080b13;color:#e9efff;font:16px/1.5 system-ui,sans-serif}main{max-width:820px;margin:auto;padding:40px 22px}
-.card{background:#101728;border:1px solid #2a385b;border-radius:18px;padding:24px;box-shadow:0 18px 60px #0007}h1{margin:.1em 0;color:#8ce8ff}
-textarea{box-sizing:border-box;width:100%;min-height:130px;background:#080d19;color:#fff;border:1px solid #40537f;border-radius:10px;padding:12px}
-button{margin-top:12px;background:#7ce6ff;color:#06101a;border:0;border-radius:999px;padding:10px 18px;font-weight:700;cursor:pointer}pre{white-space:pre-wrap;background:#080d19;padding:14px;border-radius:10px;min-height:52px}code{color:#a7f3d0}.fine{color:#aebbd4;font-size:.9rem}a{color:#8ce8ff}</style></head>
-<body><main><div class="card"><p class="fine">FREE CPU BASIC · ONE REQUEST AT A TIME · IMMUTABLE Q4_K_M</p><h1>SZL Model Inference Lab</h1>
-<p>Real, bounded CPU inference for <code>SZL-Khipu-1.5B-GGUF</code>. Max 1,200 input characters and 32 generated tokens.</p>
-<textarea id="prompt" maxlength="1200">Reply with one short sentence describing what a cryptographic receipt can prove.</textarea><br>
-<button id="run" disabled aria-disabled="true">Run bounded inference</button><pre id="out" role="status" aria-live="polite">Checking runtime…</pre>
-<p class="fine">Upstream receipt signatures verify against the repository-declared key only; no independent identity or key-ownership binding is claimed. No independent benchmark, post-quantization quality claim, or safety certification. Prompts are not intentionally stored. <a href="/api/v1/identity">Machine identity</a> · <a href="/health">Health</a></p></div></main>
+<html lang="en" data-screenshot-ready="false"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1">
+<meta name="theme-color" content="#0c0811"><meta name="color-scheme" content="dark">
+<meta name="description" content="A bounded, source-bound GGUF inference instrument with visible model, receipt, and unsigned-output boundaries.">
+<meta property="og:type" content="website"><meta property="og:title" content="SZL Model Inference Lab - Khipu Loom">
+<meta property="og:description" content="Every token leaves a thread: inspect exact model bytes, deploy identity, receipt evidence, and bounded execution together.">
+<meta property="og:url" content="https://szlholdings-szl-model-inference-lab.hf.space/">
+<link rel="canonical" href="https://szlholdings-szl-model-inference-lab.hf.space/">
+<title>SZL Model Inference Lab - Khipu Loom</title>
+<style>
+:root{--ink:#0c0811;--ink-2:#120d19;--panel:#171020;--line:#3b2a45;--text:#f7f0f4;--muted:#bcaebc;--fiber:#ef86d7;--ember:#ffbd69;--mint:#63e6d4;--danger:#ff6f82;--radius:22px;--shadow:0 30px 90px #0009}
+*{box-sizing:border-box}html{scroll-behavior:smooth}body{margin:0;background:radial-gradient(circle at 78% 12%,#301733 0,transparent 34%),radial-gradient(circle at 7% 78%,#172b30 0,transparent 30%),var(--ink);color:var(--text);font:16px/1.55 Inter,ui-sans-serif,system-ui,-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif;min-height:100vh}
+body:before{content:"";position:fixed;inset:0;pointer-events:none;opacity:.2;background-image:linear-gradient(#fff1 1px,transparent 1px),linear-gradient(90deg,#fff1 1px,transparent 1px);background-size:48px 48px;mask-image:linear-gradient(to bottom,#000,transparent 82%)}
+a{color:inherit}.skip{position:fixed;left:16px;top:-60px;background:var(--text);color:var(--ink);padding:10px 14px;z-index:20}.skip:focus{top:16px}
+.shell{width:min(1180px,calc(100% - 40px));margin:auto;position:relative}.nav{display:flex;align-items:center;justify-content:space-between;gap:24px;padding:22px 0;border-bottom:1px solid #ffffff14}.brand{display:flex;align-items:center;gap:12px;text-decoration:none}.mark{width:34px;height:34px;border:1px solid var(--fiber);border-radius:50%;position:relative;box-shadow:inset 0 0 18px #ef86d744}.mark:before,.mark:after{content:"";position:absolute;background:var(--mint)}.mark:before{width:1px;height:46px;left:16px;top:-7px}.mark:after{width:46px;height:1px;left:-7px;top:16px}.brand strong{letter-spacing:.08em}.brand small{display:block;color:var(--muted);font:700 10px/1.2 ui-monospace,monospace;letter-spacing:.18em;text-transform:uppercase}.links{display:flex;gap:22px;align-items:center}.links a{color:var(--muted);font-size:13px;text-decoration:none}.links a:hover{color:var(--text)}
+.hero{display:grid;grid-template-columns:minmax(0,1.02fr) minmax(360px,.98fr);gap:58px;align-items:center;padding:82px 0 50px}.eyebrow,.mono{font-family:ui-monospace,SFMono-Regular,Consolas,monospace}.eyebrow{color:var(--mint);font-size:12px;font-weight:700;letter-spacing:.18em;text-transform:uppercase}.hero h1{font-size:clamp(52px,7vw,92px);line-height:.94;letter-spacing:-.06em;margin:20px 0 26px;max-width:760px}.hero h1 span{color:var(--ember);font-family:Georgia,serif;font-style:italic;font-weight:500}.lede{font-size:clamp(18px,2vw,22px);color:var(--muted);max-width:660px}.hero-actions{display:flex;flex-wrap:wrap;gap:12px;margin-top:30px}.button,.ghost{display:inline-flex;align-items:center;justify-content:center;min-height:48px;padding:0 20px;border-radius:999px;font-weight:800;text-decoration:none}.button{border:0;background:var(--text);color:var(--ink);cursor:pointer}.button:hover{background:var(--ember)}.button:disabled{cursor:not-allowed;opacity:.48}.ghost{border:1px solid var(--line);color:var(--text);background:#ffffff08}.ghost:hover{border-color:var(--fiber)}
+.loom{min-height:470px;border:1px solid #ffffff1b;border-radius:32px;background:linear-gradient(145deg,#ffffff0d,#ffffff03);box-shadow:var(--shadow);position:relative;overflow:hidden}.loom:before{content:"";position:absolute;inset:28px;border:1px solid #ffffff12;border-radius:24px}.cord{position:absolute;left:13%;right:13%;height:1px;transform-origin:center;background:linear-gradient(90deg,transparent,var(--fiber),var(--ember),var(--mint),transparent);box-shadow:0 0 14px #ef86d766}.cord.c1{top:23%;transform:rotate(14deg)}.cord.c2{top:43%;transform:rotate(-9deg)}.cord.c3{top:65%;transform:rotate(5deg)}.cord.c4{top:78%;transform:rotate(-15deg)}.knot{position:absolute;width:18px;height:18px;border-radius:50%;background:var(--panel);border:3px solid var(--ember);box-shadow:0 0 0 7px #ffbd6917,0 0 28px #ffbd6955}.k1{left:24%;top:29%}.k2{left:52%;top:39%;border-color:var(--fiber)}.k3{left:72%;top:58%;border-color:var(--mint)}.k4{left:34%;top:70%}.loom-label{position:absolute;padding:9px 12px;border:1px solid var(--line);border-radius:12px;background:#0c0811db;font:700 10px/1.2 ui-monospace,monospace;letter-spacing:.12em;text-transform:uppercase}.l1{left:9%;top:12%;color:var(--fiber)}.l2{right:8%;top:33%;color:var(--ember)}.l3{left:12%;bottom:13%;color:var(--mint)}.loom-core{position:absolute;left:50%;top:50%;width:130px;height:130px;transform:translate(-50%,-50%);border:1px solid #ffffff20;border-radius:50%;display:grid;place-items:center;background:#100b17cc;box-shadow:0 0 0 22px #ffffff05,0 0 80px #ef86d72b;text-align:center}.loom-core b{font:700 12px/1.2 ui-monospace,monospace;letter-spacing:.13em}.loom-core small{display:block;color:var(--muted);margin-top:5px}
+.status-strip{display:grid;grid-template-columns:repeat(4,1fr);border:1px solid var(--line);border-radius:20px;background:#0c0811b8;overflow:hidden;margin:4px 0 72px}.metric{padding:20px 22px;border-right:1px solid var(--line)}.metric:last-child{border:0}.metric b{display:block;font:800 14px/1.25 ui-monospace,monospace;word-break:break-word}.metric small{color:var(--muted);font-size:11px;letter-spacing:.1em;text-transform:uppercase}.state{display:inline-flex;align-items:center;gap:8px}.state:before{content:"";width:8px;height:8px;border-radius:50%;background:var(--ember);box-shadow:0 0 14px currentColor}.state.ready{color:var(--mint)}.state.ready:before{background:var(--mint)}.state.failed{color:var(--danger)}.state.failed:before{background:var(--danger)}
+.section{padding:72px 0;border-top:1px solid #ffffff12}.section-head{display:flex;align-items:end;justify-content:space-between;gap:30px;margin-bottom:30px}.section h2{font-size:clamp(34px,5vw,58px);line-height:1;margin:10px 0;letter-spacing:-.04em}.section-copy{color:var(--muted);max-width:600px}.composer{display:grid;grid-template-columns:1.08fr .92fr;border:1px solid var(--line);border-radius:28px;overflow:hidden;background:var(--panel);box-shadow:var(--shadow)}.input-pane,.output-pane{padding:28px}.output-pane{background:#0a0710;border-left:1px solid var(--line)}label{display:block;font-weight:800;margin-bottom:10px}textarea{box-sizing:border-box;width:100%;min-height:180px;resize:vertical;background:#0b0710;color:var(--text);border:1px solid #4a3554;border-radius:16px;padding:16px;font:15px/1.55 ui-monospace,monospace}textarea:focus{outline:2px solid var(--fiber);outline-offset:2px}.composer-meta{display:flex;justify-content:space-between;gap:14px;color:var(--muted);font-size:12px;margin:10px 2px}.run-row{display:flex;align-items:center;justify-content:space-between;gap:18px;margin-top:18px}.output-label{color:var(--mint);font:700 11px ui-monospace,monospace;letter-spacing:.14em;text-transform:uppercase}.output-pane pre{white-space:pre-wrap;margin:18px 0 0;min-height:220px;color:#e8dfe8;font:15px/1.7 ui-monospace,monospace}.thread-id{color:var(--muted);font-size:11px;word-break:break-all}
+.evidence-grid{display:grid;grid-template-columns:repeat(3,1fr);gap:16px}.evidence-card{min-height:220px;border:1px solid var(--line);border-radius:20px;background:#ffffff08;padding:22px}.evidence-card .number{color:var(--ember);font:700 12px ui-monospace,monospace}.evidence-card h3{font-size:20px;margin:40px 0 10px}.evidence-card p{color:var(--muted);font-size:14px}.evidence-card a{color:var(--mint)}
+.boundary{display:grid;grid-template-columns:1fr 1fr;gap:16px}.boundary article{padding:26px;border-radius:22px;border:1px solid var(--line)}.boundary article:first-child{background:#17302e55}.boundary article:last-child{background:#34182255}.boundary h3{margin-top:0}.boundary ul{padding-left:20px;color:var(--muted)}footer{display:flex;justify-content:space-between;gap:24px;padding:38px 0 56px;color:var(--muted);font-size:12px;border-top:1px solid #ffffff12}
+:focus-visible{outline:2px solid var(--mint);outline-offset:3px}@media(max-width:850px){.links{display:none}.hero{grid-template-columns:1fr;padding-top:56px}.loom{min-height:390px}.status-strip{grid-template-columns:1fr 1fr}.metric:nth-child(2){border-right:0}.metric:nth-child(-n+2){border-bottom:1px solid var(--line)}.composer{grid-template-columns:1fr}.output-pane{border-left:0;border-top:1px solid var(--line)}.evidence-grid{grid-template-columns:1fr}.boundary{grid-template-columns:1fr}.section-head{display:block}}@media(max-width:520px){.shell{width:min(100% - 24px,1180px)}.hero{gap:34px}.hero h1{font-size:52px}.loom{min-height:330px}.status-strip{grid-template-columns:1fr}.metric{border-right:0;border-bottom:1px solid var(--line)!important}.metric:last-child{border-bottom:0!important}.input-pane,.output-pane{padding:20px}.run-row{align-items:stretch;flex-direction:column}.button{width:100%}footer{display:block}}
+@media(prefers-reduced-motion:reduce){html{scroll-behavior:auto}*,*:before,*:after{animation:none!important;transition:none!important}}
+</style></head>
+<body><a class="skip" href="#main">Skip to inference lab</a><div class="shell">
+<nav class="nav" aria-label="Primary"><a class="brand" href="/"><span class="mark" aria-hidden="true"></span><span><strong>SZL / KHIPU LOOM</strong><small>Bounded inference instrument</small></span></a><div class="links"><a href="#run-lab">Run</a><a href="#evidence">Evidence</a><a href="/api/v1/identity">Identity</a><a href="/.well-known/szl-inference-contract.json">API contract</a></div></nav>
+<main id="main"><section class="hero"><div><p class="eyebrow">Formula genome 01 / public cpu instrument</p><h1>Every token leaves a <span>thread.</span></h1><p class="lede">A bounded inference surface where the exact model bytes, source revision, receipt chain, limits, and missing guarantees remain visible together.</p><div class="hero-actions"><a class="button" href="#run-lab">Enter the loom</a><a class="ghost" href="/evidence">Inspect machine evidence</a></div></div>
+<div class="loom" aria-label="Abstract Khipu provenance loom showing source, model, and receipt threads"><i class="cord c1"></i><i class="cord c2"></i><i class="cord c3"></i><i class="cord c4"></i><i class="knot k1"></i><i class="knot k2"></i><i class="knot k3"></i><i class="knot k4"></i><span class="loom-label l1">source / exact sha</span><span class="loom-label l2">model / q4_k_m</span><span class="loom-label l3">output / unsigned</span><div class="loom-core"><div><b>KHIPU<br>1.5B</b><small>CPU bound</small></div></div></div></section>
+<section class="status-strip" aria-label="Live runtime summary"><div class="metric"><small>Runtime</small><b id="runtime-state" class="state">CHECKING</b></div><div class="metric"><small>Git source</small><b id="source-sha">UNAVAILABLE</b></div><div class="metric"><small>Model pin</small><b>67d60ec...f4a4b5d25</b></div><div class="metric"><small>Output proof</small><b>UNSIGNED / HASHED</b></div></section>
+<section class="section" id="run-lab"><div class="section-head"><div><p class="eyebrow">Bounded generation</p><h2>Pull one thread.</h2></div><p class="section-copy">One request at a time. At most 1,200 characters, 800 formatted prompt tokens, and 32 generated tokens. Greedy decoding. No tools, streaming, or hidden fallback.</p></div>
+<div class="composer"><div class="input-pane"><label for="prompt">Prompt</label><textarea id="prompt" maxlength="1200">Reply with one short sentence describing what a cryptographic receipt can prove.</textarea><div class="composer-meta"><span id="char-count">0 / 1,200 characters</span><span>24 output tokens</span></div><div class="run-row"><button class="button" id="run" disabled aria-disabled="true">Run bounded inference</button><span class="mono thread-id" id="request-state">Waiting for readiness</span></div></div><div class="output-pane"><span class="output-label">Model output</span><pre id="out" role="status" aria-live="polite">Checking the runtime and evidence threads...</pre></div></div></section>
+<section class="section" id="evidence"><div class="section-head"><div><p class="eyebrow">Inspectable by default</p><h2>The evidence bay.</h2></div><p class="section-copy">The interface does not turn provenance into decoration. Every status below resolves to a machine-readable surface.</p></div><div class="evidence-grid"><article class="evidence-card"><span class="number">01 / SOURCE</span><h3>Exact deploy identity</h3><p id="source-detail">The Git revision must come from the governed Space variable. Missing or malformed identity fails closed.</p><a href="/version">Open /version</a></article><article class="evidence-card"><span class="number">02 / RECEIPTS</span><h3>Declared-key chain</h3><p id="receipt-detail">Training and evaluation receipts are verified against the repository-declared key and canonical hash chain.</p><a href="/evidence">Open /evidence</a></article><article class="evidence-card"><span class="number">03 / RUNTIME</span><h3>Exact model bytes</h3><p>Q4_K_M GGUF bytes are fetched at image build from an immutable revision, size-checked, hash-checked, and loaded offline.</p><a href="/api/v1/identity">Open identity</a></article></div></section>
+<section class="section"><div class="boundary"><article><p class="eyebrow">What is measured</p><h3>Internal integrity and bounded execution</h3><ul><li>Exact GGUF revision, size, and SHA-256</li><li>Exact deploy-bound Git revision when configured</li><li>Declared-key training/evaluation receipt chain</li><li>Runtime readiness and deterministic request limits</li></ul></article><article><p class="eyebrow">What is not claimed</p><h3>Authenticity and quality remain bounded</h3><ul><li>No independent identity or key-ownership binding</li><li>No post-quantization quality or safety certification</li><li>No signed output or reproducible execution claim</li><li>No provider SLA, sensitive-data handling guarantee, or autonomy</li></ul></article></div></section></main>
+<footer><span>SZL Holdings / Khipu Loom / Apache-2.0</span><span class="mono">MEASURED, REPORTED, UNKNOWN, or UNAVAILABLE - never implied</span></footer></div>
 <script>
-const b=document.querySelector('#run'),o=document.querySelector('#out'),p=document.querySelector('#prompt');
+const b=document.querySelector('#run'),o=document.querySelector('#out'),p=document.querySelector('#prompt'),root=document.documentElement;
+const runtimeState=document.querySelector('#runtime-state'),sourceSha=document.querySelector('#source-sha'),requestState=document.querySelector('#request-state'),charCount=document.querySelector('#char-count'),sourceDetail=document.querySelector('#source-detail'),receiptDetail=document.querySelector('#receipt-detail');
 let running=false,hasResult=false;
-function renderStatus(status,failure){const ready=status==='READY';b.disabled=!ready||running;b.setAttribute('aria-disabled',String(b.disabled));if(running)return;if(ready){if(!hasResult)o.textContent='Runtime READY. You can run bounded inference.';return}hasResult=false;o.textContent=status==='FAILED'?'Runtime FAILED'+(failure?': '+failure:'')+'.':status==='STARTING'?'Runtime STARTING…':'Runtime status unavailable; retrying…'}
-async function checkHealth(){try{const r=await fetch('/health',{cache:'no-store'}),j=await r.json();renderStatus(j.status,j.failure_code)}catch(_){renderStatus('UNAVAILABLE')}}
-b.onclick=async()=>{if(b.disabled)return;running=true;hasResult=false;b.disabled=true;b.setAttribute('aria-disabled','true');o.textContent='Running on free CPU…';try{const r=await fetch('/api/v1/infer',{method:'POST',headers:{'content-type':'application/json'},body:JSON.stringify({prompt:p.value,max_new_tokens:24})}),j=await r.json();hasResult=true;o.textContent=r.ok?j.output:JSON.stringify(j)}catch(e){hasResult=true;o.textContent='Request failed: '+e}finally{running=false;await checkHealth()}};
-checkHealth();setInterval(checkHealth,5000);
+function short(value){return typeof value==='string'&&value.length>12?value.slice(0,7)+'...'+value.slice(-7):value||'UNAVAILABLE'}
+function countChars(){charCount.textContent=p.value.length.toLocaleString()+' / 1,200 characters'}
+function renderStatus(status,failure){const ready=status==='READY';runtimeState.textContent=status||'UNAVAILABLE';runtimeState.className='state '+(ready?'ready':status==='FAILED'?'failed':'');b.disabled=!ready||running;b.setAttribute('aria-disabled',String(b.disabled));root.dataset.screenshotReady=String(ready);if(running)return;if(!hasResult)requestState.textContent=ready?'Runtime ready':status==='FAILED'?'Runtime failed':'Warming exact model bytes';if(ready){if(!hasResult)o.textContent='Runtime READY. Pull a thread when you are ready.';return}if(!hasResult)o.textContent=status==='FAILED'?'Runtime FAILED'+(failure?': '+failure:'')+'.':status==='STARTING'?'Runtime STARTING. Verifying exact model bytes and receipts...':'Runtime status unavailable; retrying...'}
+async function getJson(path){const r=await fetch(path,{cache:'no-store'});let j={};try{j=await r.json()}catch(_){j={}}return{ok:r.ok,status:r.status,json:j}}
+async function refresh(){const [health,version,evidence]=await Promise.allSettled([getJson('/health'),getJson('/version'),getJson('/evidence')]);if(health.status==='fulfilled')renderStatus(health.value.json.status,health.value.json.failure_code);else renderStatus('UNAVAILABLE');if(version.status==='fulfilled'){const sha=version.value.json.gitSha;sourceSha.textContent=short(sha);sourceDetail.textContent=sha?'Governed Git source '+sha+' is bound to this deployment.':'Exact governed Git source is unavailable; /version fails closed.'}if(evidence.status==='fulfilled'){const receipts=evidence.value.json.receipts||[];const verified=receipts.filter(x=>x.status==='DECLARED_KEY_SIGNATURES_VALID').length;receiptDetail.textContent=verified===2?'Two declared-key receipts are visible and chain-verified. Runtime outputs remain unsigned.':'Receipt evidence is not yet available; no green state is inferred.'}}
+p.addEventListener('input',countChars);b.addEventListener('click',async()=>{if(b.disabled)return;running=true;hasResult=false;b.disabled=true;b.setAttribute('aria-disabled','true');requestState.textContent='Inference in progress';o.textContent='Pulling the bounded model thread on free CPU...';try{const r=await fetch('/api/v1/infer',{method:'POST',headers:{'content-type':'application/json'},body:JSON.stringify({prompt:p.value,max_new_tokens:24})});const j=await r.json();hasResult=true;o.textContent=r.ok?j.output:JSON.stringify(j);requestState.textContent=r.ok?'Completed / output unsigned':'Request refused / '+r.status}catch(_){hasResult=true;o.textContent='Request unavailable. No result was fabricated.';requestState.textContent='Network unavailable'}finally{running=false;await refresh()}});
+countChars();refresh();setInterval(refresh,5000);
 </script></body></html>"""

release.json

@@ -1,15 +1,15 @@
 {
   "schema": "szl.space-source-release/v1",
-  "release_id": "145e9425-ff95-444d-825e-b644530bf520",
+  "release_id": "c893e93c-9bd8-47e9-941e-f8613a608992",
   "source_files": {
     ".dockerignore": "dba1fb4c6538a44cd8d7e77f133af90a8e7927d3dee23f8694fe6d867238333c",
     "Dockerfile": "6d982909b8678c6ab3bfc21766f3356c1dedfff1c2efea5ba0fbc64ed7cff4b8",
     "LICENSE": "0a41cda63c0751e5cdb240864525fc8bf29b2d3f2cb5c6406b9bae97f1e1407b",
-    "README.md": "7f3b1bac2c3b2648b6b5f3320f5ac7855e7bb4954ca323447eabcc3d521ac666",
-    "app.py": "5f382fe7a6303a8e8cb5e0a44b60f13f8edc46d0e3a8f30dd94e8f759aff390b",
+    "README.md": "c2edfc0957ccd9f2719ca3c8e71baefd23a9e88d61cfc5d3445009ae470618cb",
+    "app.py": "992badf3f02189d7f44bb4cbb5881dce837ecc72f183a7dd40952cf0106a8e37",
     "download_artifacts.py": "aa9f98b3d19753953864beae167b5e74872270c880e78430bb345a5f06b4ef36",
     "requirements.txt": "9f690793b7b9c6e1adce88244de01d77898d117779c3428b80190eabaf4658f9",
-    "tests/test_app.py": "736b419a6f4b213661a9da33e6b68ae202cb636939cab63309fe415131f7af11",
-    "verify_execution_record.py": "7ff66006fa6f405c8d31c1f2b9f6aca7c288e9eb3aef23e38122b1565d3cd6db"
+    "tests/test_app.py": "6538ee45b40e5d0dc12b1849b29c4dcb6b3184542199b0e6669c6239a73888e0",
+    "verify_execution_record.py": "1a64e3c8097409c19d971f711d63fbeb2066e3bfe804da31f684d0b795a529a6"
   }
 }

tests/test_app.py

@@ -285,6 +285,77 @@ class AppContractTests(unittest.TestCase):
             else:
                 os.environ[app.SOURCE_REVISION_ENV] = original
 
+    def test_operational_aliases_preserve_liveness_and_readiness(self):
+        original = app.state["status"]
+        try:
+            app.state["status"] = "STARTING"
+            self.assertEqual(200, app.healthz().status_code)
+            self.assertEqual(503, app.readyz().status_code)
+            app.state["status"] = "READY"
+            self.assertEqual(200, app.healthz().status_code)
+            self.assertEqual(200, app.readyz().status_code)
+            app.state["status"] = "FAILED"
+            self.assertEqual(503, app.healthz().status_code)
+            self.assertEqual(503, app.readyz().status_code)
+        finally:
+            app.state["status"] = original
+
+    def test_version_fails_closed_without_exact_deploy_revision(self):
+        original = os.environ.pop(app.SOURCE_REVISION_ENV, None)
+        try:
+            response = app.version()
+            payload = json.loads(response.body)
+            self.assertEqual(503, response.status_code)
+            self.assertIsNone(payload["gitSha"])
+            self.assertEqual("UNAVAILABLE", payload["evidenceState"])
+
+            os.environ[app.SOURCE_REVISION_ENV] = "b" * 40
+            response = app.version()
+            payload = json.loads(response.body)
+            self.assertEqual(200, response.status_code)
+            self.assertEqual("b" * 40, payload["gitSha"])
+            self.assertEqual("model-inference", payload["surface"])
+        finally:
+            if original is None:
+                os.environ.pop(app.SOURCE_REVISION_ENV, None)
+            else:
+                os.environ[app.SOURCE_REVISION_ENV] = original
+
+    def test_evidence_requires_source_and_verified_receipts(self):
+        original_revision = os.environ.get(app.SOURCE_REVISION_ENV)
+        original_state = dict(app.state)
+        try:
+            os.environ[app.SOURCE_REVISION_ENV] = "c" * 40
+            app.state.update(
+                {
+                    "status": "READY",
+                    "source_integrity": True,
+                    "model_sha256": app.MODEL_SHA256,
+                    "receipt_status": "DECLARED_KEY_SIGNATURES_VALID",
+                    "receipt_evidence": {
+                        "training_canonical_sha256": "d" * 64,
+                        "eval_canonical_sha256": "e" * 64,
+                    },
+                }
+            )
+            response = app.evidence()
+            payload = json.loads(response.body)
+            self.assertEqual(200, response.status_code)
+            self.assertEqual("MEASURED", payload["evidenceState"])
+            self.assertEqual(2, len(payload["receipts"]))
+            self.assertEqual("UNSIGNED", payload["outputProvenance"]["signatureStatus"])
+
+            app.state["receipt_status"] = "NOT_CHECKED"
+            response = app.evidence()
+            self.assertEqual(503, response.status_code)
+        finally:
+            app.state.clear()
+            app.state.update(original_state)
+            if original_revision is None:
+                os.environ.pop(app.SOURCE_REVISION_ENV, None)
+            else:
+                os.environ[app.SOURCE_REVISION_ENV] = original_revision
+
     def test_openai_model_catalog_is_immutable_and_unsigned(self):
         response = app.openai_models()
         payload = json.loads(response.body)
@@ -736,11 +807,21 @@ class AppContractTests(unittest.TestCase):
     def test_root_waits_for_readiness_before_enabling_inference(self):
         html = app.index()
         self.assertIn('id="run" disabled aria-disabled="true"', html)
-        self.assertIn("Checking runtime…", html)
-        self.assertIn("fetch('/health'", html)
+        self.assertIn("Checking the runtime and evidence threads", html)
+        self.assertIn("getJson('/health')", html)
+        self.assertIn("getJson('/version')", html)
+        self.assertIn("getJson('/evidence')", html)
         self.assertIn("status==='READY'", html)
         self.assertIn("status==='STARTING'", html)
         self.assertIn("status==='FAILED'", html)
+        self.assertIn("if(running)return;if(!hasResult)requestState", html)
+        self.assertIn("if(!hasResult)o.textContent", html)
+        self.assertIn('data-screenshot-ready="false"', html)
+        self.assertIn("prefers-reduced-motion:reduce", html)
+        self.assertIn("Every token leaves a", html)
+        self.assertIn('name="description"', html)
+        self.assertIn('property="og:title"', html)
+        self.assertIn('rel="canonical"', html)
 
     def test_body_limiter_rejects_chunked_overflow(self):
         downstream_called = False

verify_execution_record.py

@@ -34,7 +34,7 @@ EXPECTED_MODEL = {
     "sha256": "13c1a1993063e1dff92f7413ccf48eaca6d48efc8801ae9af35961ae3396623a",
 }
 EXPECTED_SPACE_ID = "SZLHOLDINGS/szl-model-inference-lab"
-EXPECTED_RELEASE_ID = "145e9425-ff95-444d-825e-b644530bf520"
+EXPECTED_RELEASE_ID = "c893e93c-9bd8-47e9-941e-f8613a608992"
 EXPECTED_RECORD_HASH_SCOPE = (
     "canonical UTF-8 JSON with sorted keys and compact separators, "
     "excluding record_sha256"