ci(standards): move the praetor engine pin to main 846da59 and tighten the stale baseline

[lusoris]

Summary

standards-gate.yml installs praetor at e4b35cb3c7fe (praetor main as of 2026-09-18). Main is now 846da5908d15. This moves the pin — and, because the workflow comment says "the baseline is recorded against this engine", it first proves the move changes nothing.

The engine bump changes no finding

Three builds recorded .standards-baseline.json on the identical origin/master tree (551d35a63):

Engine	How built	Result
e4b35cb3c7fe	the previously pinned binary	1414 entries
846da5908d15	local build of praetor main	1414 entries
846da5908d15	go install …/cmd/standardsctl@846da590…, exactly what this workflow runs	1414 entries

The three JSON files are byte-identical after dropping the timestamp — same fingerprints, same messages, same order. praetorctl audit passes under all three.

The committed baseline was 19 looser than the tree

Every one of those recordings reported Total: 1414 infractions, previously 1433. The 19 are two files cleaned in merged PRs without the same-PR re-record ADR-1142 asks for:

File	Baseline said	Tree has	Cleaned by
core/src/svm.cpp	14	0	#1498
core/test/test_ciede_neon.c	5	0	an earlier PR

A too-loose baseline fails nothing — the audit errors only on an increase, or on unmatched fingerprints in a touched file — so those entries were quietly headroom for 19 new findings. Ledger L-80; reported upstream as cordanaLLM/praetor#349 (the audit should notice stale entries itself). Re-recorded here, in the PR that owns the engine, so #1499, #1500, #1501 and #1502 stop each correcting it in passing and conflicting with one another on the file.

Verification

Check	Result
Baseline recorded by old engine vs new engine vs CI-form go install	byte-identical
praetorctl audit (new engine) on this branch	passes
Findings changed by the bump	none

Type

• ci — CI / tooling change

Checklist

• Commits follow Conventional Commits.
• Pre-push hooks pass.
• I did not modify any assertAlmostEqual(...) score in the Netflix golden Python tests.
• docs/state.md: no row — no bug closed or opened in the fork's tracked set; the baseline drift is a ledger item (L-80) and an upstream praetor issue.

Deep-dive deliverables (ADR-0108)

• Research digest — no digest needed: trivial. The measurement is the three-way baseline diff above.
• Decision matrix — no alternatives: only-one-way fix. A pin either tracks the engine the baseline was proven against or it does not.
• AGENTS.md invariant note — no rebase-sensitive invariants: a workflow env var and a generated JSON.
• Reproducer / smoke-test command — below.
• CHANGELOG fragment — no changelog fragment needed: CI-internal, no user-visible delta.
• Rebase note — no rebase impact: fork-local CI files an upstream sync never touches.

Reproducer

GOBIN=/tmp/pb go install github.com/cordanaLLM/praetor/cmd/standardsctl@846da5908d15b3cf5581ca6b0205cc644b249599
git clone --no-hardlinks --branch master . /tmp/clean && cd /tmp/clean
/tmp/pb/standardsctl baseline --record     # prints: Total: 1414 infractions, previously 1433

Repeat with @e4b35cb3c7fe7abe91554e3476e0ffd9caadbfa4 and diff the two JSONs ignoring generated_at.