Why v1.3 is a minor bump.

The v1.2 line was a philosophy shift. v1.3 is a locus shift. Different shape of change, different version digit.

v1.2 told the harness to coach instead of punish. v1.2.1 patched the bits that didn't work. v1.2.2 hardened what did, with 29 days of clean production data behind it.

v1.3 is not the next step in that line. It changes where discipline lives.

In v1.2.x, discipline lived in rules plus hook-enforcement. You said the thing. The hook nagged when you broke it. The audit caught what slipped through.

In v1.3, discipline lives in structure that makes the failure mode hard to occur. Same discipline. Shapes that don't need ceremony to enforce.

Rules need case studies before they're rules. Schemas migrate when at-scale evidence says they should, not when someone remembers. The harness stops policing every interaction and gets out of the way. Public commitments get watched by a folder, not by a person remembering. That's a paradigm shift. Hence minor, not patch.

One more concept lands in v1.3. It took 90+ days to surface and it changes how Hard Rules sit relative to the protocol family.

The protocol is the brewery. Yeast, water, grain, the brew cycle. These are constants. They make you "a kaygee protocol" at all. Files are memory. Generated beats maintained. Build the fence, not the net. RECORD after every logical unit.

uno, duo, tre. These are flavours. Recipes brewed in the same brewery. Different hops, different fermentation temperatures, different malt roasts. Same fundamentals.

The mistake we'd been making: treating the flavours as if they should be symmetric. Asking "why doesn't uno have a DECISIONS.md if duo does?" misses the point. uno is an OS spec, POSIX-shaped, monolithic. duo is a construction site, ADR-shaped, split. They're different beers. The right shape for an ale isn't the right shape for a lager.

What v1.3 codifies: Hard Rules sit at two levels.

• → Protocol-level rules apply to every flavour. You live and breathe them because you ARE a kaygee protocol. Files-are-memory. The case-study gate itself. RECORD after every logical unit.

• → Flavour-level rules apply because you chose this recipe. You live and breathe them because you ARE the uno flavour (or the duo flavour). N1 docs-travel-with-code ships in protocol-duo because the convergent evidence came from two Construct-flavour repos. It stays silent in protocol-uno and the maintainer itself. They're not brewing code.

The flavours can morph into each other. A duo project matures into uno-mode substrate when construction ends and ongoing-operations begin. A uno deployment spins up a duo sub-project for a bounded build. Rigid walls between flavours aren't there. Only the brewery is constant.

And even that's contingent. The definition of "beer" will change as frontier models change, as harnesses change. v1.3 is one snapshot. v1.4 will be another.

• →Fewer judgment calls at audit time. Rules either cite case studies or they're deletion candidates. The trim becomes a lookup, not a debate. The anti-pattern table stops growing without bound.
• →Fewer hook errors. The default doesn't police every interaction. If you want hooks, opt in. If you don't, the protocol still works, possibly better.
• →Clearer rule placement. Protocol-level rules apply because you ARE a kaygee protocol. Flavour-level rules apply because you chose this recipe. Both layers are visible. No more "why doesn't uno have what duo has?" Different beer.
• →A folder watching the falsification. The public claims we make at /learn now have a structural witness. If Cursor ships hooks, if Codex's /goal adopts an external judge, if native memory absorbs project-scoped state, we notice through the folder, not by chance.
• →Migration cost is real but bounded. Front-matter changes shape. Hookless settings get rewritten. Existing rules without case studies get flagged at the next audit. A 60-day compat shim covers the transition. After that, the old shape stops working. That's the point.