docs: freeze the 2026-08-05 Phase 1 product-quality baseline

.gitignore

@@ -150,6 +150,11 @@ docs/accuracy.md
 # Typed DNS catalog baseline (2026-07-17): aggregate counts and generic
 # provider patterns only. No target names, target records, tenant IDs, or rows.
 !/validation/2026-07-17-typed-catalog-baseline.md
+# Phase 1 product-quality baseline (2026-08-05): generated by
+# scripts/quality_scorecard.py from the registered MCP surface and the shipped
+# catalog. Network-free and corpus-free, so it contains protocol byte counts and
+# catalog counts only. No target names, records, tenant IDs, or rows.
+!/validation/2026-08-05-quality-baseline-scorecard.md
 # Per-pattern base-rate reducer and its aggregate memo (2026-07-25):
 # provider-owned patterns and aggregate counts only; the reducer never
 # reads target-identifying fields and its output stays in ignored paths.

scripts/quality_scorecard.py

@@ -419,7 +419,9 @@ def _render_markdown(scorecard: dict[str, Any]) -> str:
     lines = [
         "# Product-quality baseline scorecard",
         "",
-        f"Commit `{commit[:12]}`, catalog `{(revisions['catalog_digest_sha256'] or 'unknown')[:12]}`, "
+        # The catalog digest carries its algorithm prefix so neither a reader nor
+        # the backticked-commit-receipt gate mistakes a content hash for a commit.
+        f"Commit `{commit[:12]}`, catalog `sha256:{(revisions['catalog_digest_sha256'] or 'unknown')[:12]}`, "
         f"Python {scorecard['environment']['python']}.",
         "",
         "Network-free, corpus-free, aggregate-safe. Diagnostic artifact, not a gate.",

tests/test_quality_scorecard.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import json
+import re
 from typing import Any
 
 import pytest
@@ -159,6 +160,18 @@ def test_markdown_memo_reports_the_measured_numbers(scorecard: dict[str, Any]) -
         assert entry["metric"] in memo
 
 
+def test_markdown_memo_never_prints_a_non_commit_hash_as_a_commit_receipt(scorecard: dict[str, Any]) -> None:
+    # test_documentation_integrity resolves every backticked hex string in
+    # validation/*.md as a git commit. The catalog digest is a content hash, so
+    # it must not be emitted bare or a committed memo breaks that gate.
+    memo = _render_markdown(scorecard)
+    receipts = set(re.findall(r"`([0-9a-f]{7,40})`", memo))
+    digest = scorecard["revisions"]["catalog_digest_sha256"] or ""
+
+    assert digest[:12] not in receipts
+    assert f"`sha256:{digest[:12]}`" in memo
+
+
 def test_markdown_memo_tolerates_an_absent_git_revision(scorecard: dict[str, Any]) -> None:
     # The harness must still render inside a source archive with no git metadata.
     detached = {**scorecard, "revisions": {**scorecard["revisions"], "commit": None}}

validation/2026-08-05-quality-baseline-scorecard.md

@@ -0,0 +1,67 @@
+# Product-quality baseline scorecard
+
+Commit `8a0536b34caa`, catalog `sha256:004afd17fe96`, Python 3.14.4.
+
+Network-free, corpus-free, aggregate-safe. Diagnostic artifact, not a gate.
+
+## MCP context cost
+
+What an agent pays before it does any work.
+
+| Measure | Value |
+|---|---:|
+| Registered tools | 22 |
+| Discovery payload | 81,880 bytes |
+| Instruction preamble | 9,996 bytes |
+| **Session context before first call** | **91,876 bytes** |
+| Order-of-magnitude tokens | ~22,969 |
+
+Where the discovery payload goes:
+
+| Component | Bytes | Share |
+|---|---:|---:|
+| `outputSchema` | 51,358 | 63.9% |
+| `description` | 21,867 | 27.2% |
+| `inputSchema` | 4,696 | 5.9% |
+| `annotations` | 1,945 | 2.4% |
+| `name` | 441 | 0.5% |
+
+Output schemas are 63.1% of discovery. Removing them entirely would leave 30,170 bytes, which is an upper bound on headroom rather than a proposal: structured output is part of the contract clients validate against.
+
+Cross-tool definition duplication accounts for only 6,424 bytes across 5 shared definition bodies, so a shared-definition scheme is not the lever either.
+
+Largest and smallest registered tools:
+
+- `reevaluate_domain` at 14,729 bytes
+- `clear_ephemeral_fingerprints` at 583 bytes
+- top 5 tools carry 47.4% of discovery
+
+## Catalog surface
+
+| Measure | Value |
+|---|---:|
+| Entries | 860 |
+| Detection rules | 1,070 |
+| Rules carrying a verification date | 26 (2.4%) |
+| Undated rules | 1,044 |
+
+## Measured elsewhere
+
+| Metric | Owner |
+|---|---|
+| component latency and peak allocation | `scripts/characterize_performance.py` |
+| claim lineage and provenance completeness | `scripts/check_default_claim_audit.py` |
+| typed catalog coverage from real results | `validation/catalog_baseline.py` |
+| claim-family precision, benefit, and safety | `docs/quality-baseline-preregistration.md` |
+
+## Unmeasured
+
+Named rather than omitted, because an absent metric reads as a passing one.
+
+| Metric | Blocked by |
+|---|---|
+| classified versus unclassified observable surface | private corpus |
+| CT marginal signal gain relative to latency cost | network and private corpus |
+| end-to-end cold and warm p50/p95 for single, batch, graph, and MCP workflows | network |
+| degraded-source rate and partial-result rate | network and private corpus |
+| MCP result payload bytes under real lookups | network |