[adamjohnwright]

Three reactions in Mitotic G1 make products that no Release 97 reaction uses. Each reaction's summation says what its product does.

src/summation_regulations.py writes each of these as a structural rule. A rule applies only within a pathway that holds its source reaction:

rule	regulator	effect	targets
E1	RB1:E2F	inhibits	every E2F1/2/3-stimulated transcription (15)
E2	AKT's p-T-p21/p27	inhibits	Cip/Kip binding to cyclin:CDK
E3	the p-Y-Cip/Kip CDK4/6 complex	activates	CDK4/6-catalysed phosphorylations

The regulators are added after variant enumeration, to every copy, as one pool of their variants (<stId>::pool). This means they never multiply copies or push a reaction past the cap. A first build that wired them as curated participants broke the p130/p107 recycle pools; it was never benchmarked.

Measured (deltasignal specs/102, pre-registered, against canonical defaults):

• Experimental: +6 (11 / 5), reaching 641/849.
• Curator tuning: +173 (206 / 33, Mitotic G1 only).
• Curator held-out: 0.
• The gate passes. Adopted by Adam 2026-10-24.

These edges add to Release 97 rather than reproduce it. LNG_SUMMATION_REGULATIONS=0 builds the release alone.

Changes:

• Flag LNG_SUMMATION_REGULATIONS, default on.
• PIPELINE_SETTINGS.md row.
• tests/test_summation_regulations.py, including a live-database test of the registered targets.

Checks: suite passes except the known stale test_main_edges_proportional_to_best_matches[...983169]. mypy is clean.

🤖 Generated with Claude Code