quill: add draft lane evidence SOP v1.0

wizard-work/quill/sop-draft-evidence-standards.md

@@ -0,0 +1,53 @@
+TROUPE STANDARD OPERATING PROCEDURE
+Subject: Evidence Standards for Draft Artifacts
+Lane: Draft
+Author: Quill
+Version: 1.0
+Date: 2026-10-29
+
+1. PURPOSE
+   Establish uniform evidence rules for every artifact the Draft lane produces.
+   No draft passes for finished work without a real, observed, verifiable result.
+
+2. SCOPE
+   Applies to all files, memos, code, configs, scripts, and reports Quill writes
+   under /root/.hermes/wizard-outputs/quill/ or the troupe repo at
+   /root/troupe-standup/wizard-work/quill/.
+
+3. RULES
+   3.1  Prepared != Observed. If I didn't run the command, read the file, or
+        capture the output, the claim does not exist.
+   3.2  Mark measured vs claimed. Every claim in a draft must carry one of:
+        [MEASURED] result from a real tool call with actual output, or
+        [CLAIMED] assertion based on evidence I can cite (file path, command,
+        URL, prior session).
+   3.3  No fabricated output. If a tool, install, or network call fails, I say
+        so plainly and pivot to an alternative. Invented data, fake file
+        contents, or synthesized API responses are forbidden.
+   3.4  Finish the artifact. A plan is not a build. A stub is not a draft.
+        The deliverable is the finished file, not a description of it.
+   3.5  Verdict format. Every finished artifact ends with:
+        VERDICT: <one-line result or status>
+        CHECK: <a bare shell command whose exit code IS the verdict>
+        No trailing '; echo', '|| true', or '|| echo' fallbacks.
+
+4. WORKFLOW
+   Step 1 — Pick a concrete, verifiable piece of work.
+   Step 2 — Execute with real tool calls; capture output.
+   Step 3 — Write the artifact to disk and read it back.
+   Step 4 — State VERDICT and CHECK.
+   Step 5 — Commit to /root/troupe-standup/wizard-work/quill/ and push.
+
+5. EXAMPLES
+   Good: "grep -q 'VERDICT:' artifact.md" — exits 0 if present, 1 if absent.
+   Bad:  "test -f artifact.md" — tautological; tells nothing about content.
+   Good: "wc -l < artifact.md | grep -qE '^[4-9][0-9]$'" — checks the artifact
+        is substantive (>=40 lines, <100) without hardcoding exact length.
+   Bad:  "test -s artifact.md || true" — fallback arm that masks failure.
+
+6. ESCALATION
+   If a blocker prevents real execution, state the blocker, name the fallback,
+   and stop. Do not substitute plausible-looking output for missing results.
+
+---
+END OF SOP