chore(release): cut nexus v2.1.0 — RAMZA default planner + composition/manifesto reseat (#437)

Closes the deferred composition.md + MANIFESTO reseat and cuts the version that
carries the RAMZA default-planner succession (Stage 3, merged in #436).

- composition.md reseated to ATLAS -> RAMZA -> Vivi -> IDG via ECL v2.1.1:
  9 new RAMZA planner-seat contracts in Rynaro/eidolons-ecl (mirror the SPECTRA
  edge set, default downstream coder now Vivi; validate against
  handoff-contract.v1.json) + composition.md.j2 template reseat. Nexus
  composition-drift pin bumped v2.0.1 -> v2.1.1 and composition.md regenerated;
  the pin bump also lands the Vivi coder edges deferred since ECL v2.0.2. Drift
  gate simulated locally: template byte-identical, regen byte-identical.
- MANIFESTO.md team table + pipeline prose reseated (RAMZA/Vivi default,
  SPECTRA/APIVR-Δ opt-in fallback; Kupo + CRYSTALIUM added).
- VERSION 2.0.0 -> 2.1.0; CHANGELOG [Unreleased] closed to [2.1.0].

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: Rynaro

.github/workflows/composition-drift.yml

@@ -32,7 +32,7 @@ on:
 # The pinned ECL release tag. Bump this when you accept a new
 # `contracts/` shape from `eidolons-ecl`.
 env:
-  ECL_PINNED_TAG: v2.0.1
+  ECL_PINNED_TAG: v2.1.1
 
 jobs:
   composition-drift:

CHANGELOG.md

@@ -8,11 +8,15 @@ Format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). Version
 
 ## [Unreleased]
 
+## [2.1.0] — 2026-07-06 — RAMZA takes the default planner seat
+
 ### Changed
 
 - **RAMZA takes the default planner seat (Stage 3 of the SPECTRA succession).** Promoted from `in_construction` to `shipped` at v1.0.0 (attested release) after the Stage-2 measurement closed green: AC-001 (scorer calibration), AC-002 (7/7 canary missions, 0 MUST-fail), and AC-003 — a pre-registered, budget-matched, holdout-gated planner A/B vs `spectra@4.11.0`. **24/24 cells, both arms 6/6 tasks pass², 0 MUST failures; RAMZA non-inferior (6 ≥ spectra 6 − 1) with holdout consistency intact.** Descriptive (not gated): RAMZA produced contract-conformant plans at ~51% of SPECTRA's verbosity with machine-verifiable gate audit trails; one run's maker≠checker critic caught and closed a real criteria-desync (an audit-enumeration gap) at cycle-2, independent cycle-3 PASS. Adjudication + full matrix: `.spectra/research/ramza-stage2/AC-003-ADJUDICATION.md`.
   - **Seat flip (routing is DATA):** `roster/routing.yaml` now carries a `ramza` planner block with `default_for_class: planner` + `fallback: spectra`; the four planner chains (plan-before-build, ship-fast, decide-then-implement, scout-then-spec) and the no-spec confidence signal reseat to RAMZA. `EIDOLONS.md` cortex, the `pipeline`/`plan-and-build`/`full`/`errands` presets, and the README reseat to RAMZA-default.
-  - **SPECTRA retained as the conservative opt-in fallback** (`eidolons add spectra`) — still shipped, still reachable by named dispatch (the +0.5 name bonus). Mirrors the Vivi/APIVR-Δ coder precedent. Deferred follow-up (matching that precedent): `methodology/composition.md` ECL contract reseat.
+  - **SPECTRA retained as the conservative opt-in fallback** (`eidolons add spectra`) — still shipped, still reachable by named dispatch (the +0.5 name bonus). Mirrors the Vivi/APIVR-Δ coder precedent.
+
+- **Composition + manifesto reseat to the current default pipeline (`ATLAS → RAMZA → Vivi → IDG`).** Added nine RAMZA planner-seat ECL contracts to `Rynaro/eidolons-ecl` (`atlas→ramza`, `ramza→vivi`, `ramza→apivr`, `ramza→forge`, `ramza→kupo`, `forge→ramza`, `human→ramza`, `kupo→ramza`, `vigil→ramza`; all validate against `schemas/handoff-contract.v1.json`) and reseated the `composition.md.j2` template — cut as **ECL v2.1.1**. Bumped the nexus `composition-drift.yml` pin `v2.0.1 → v2.1.1` and regenerated `methodology/composition.md`; the same pin bump also lands the Vivi coder edges deferred since ECL v2.0.2. `MANIFESTO.md` team table + pipeline prose reseated (RAMZA/Vivi default, SPECTRA/APIVR-Δ fallback; Kupo + CRYSTALIUM added).
 
 ### Added
 

MANIFESTO.md

@@ -18,7 +18,7 @@ The second wave tried **sub-agent pipelines inside a single system** — one har
 
 ### 1. Each member is an individual
 
-The Eidolons are not interchangeable cogs. Each has a name, a capability class, a named methodology with a phase cycle, and a refusal boundary. ATLAS refuses to write; SPECTRA refuses to implement; IDG refuses to research. This isn't a limitation — it's the point. Sharp boundaries let each member go deeper in its domain than a generalist ever could.
+The Eidolons are not interchangeable cogs. Each has a name, a capability class, a named methodology with a phase cycle, and a refusal boundary. ATLAS refuses to write; RAMZA refuses to implement; IDG refuses to research. This isn't a limitation — it's the point. Sharp boundaries let each member go deeper in its domain than a generalist ever could.
 
 ### 2. The team is portable
 
@@ -30,7 +30,7 @@ No "because it feels right." Every design decision in every Eidolon maps to a re
 
 ### 4. Composition is first-class
 
-The team is more than the sum of its members. The canonical pipeline — ATLAS → SPECTRA → APIVR-Δ → IDG, with FORGE (reasoning) and VIGIL (forensic debugging) as lateral specialists — is a real compositional asset. Handoff contracts are structured artifacts, not free-form messages. Partial-team deployment is supported by design: bring just ATLAS to an audit-heavy project, bring the whole pipeline to a greenfield, bring ATLAS + IDG when you want to understand and document without changing anything.
+The team is more than the sum of its members. The canonical pipeline — ATLAS → RAMZA → Vivi → IDG, with FORGE (reasoning) and VIGIL (forensic debugging) as lateral specialists — is a real compositional asset. Handoff contracts are structured artifacts, not free-form messages. Partial-team deployment is supported by design: bring just ATLAS to an audit-heavy project, bring the whole pipeline to a greenfield, bring ATLAS + IDG when you want to understand and document without changing anything.
 
 ---
 
@@ -40,7 +40,7 @@ The team is more than the sum of its members. The canonical pipeline — ATLAS 
 
 **Unbounded reflection loops.** Research (CorrectBench 2025) shows they degrade prose quality. Every Eidolon has a fixed verification gate and a bounded revision budget.
 
-**Internal sub-agent pipelines inside one member.** If ATLAS needs planning, it hands off to SPECTRA; it does not grow an internal SPECTRA. Roster bloat dilutes the team.
+**Internal sub-agent pipelines inside one member.** If ATLAS needs planning, it hands off to RAMZA; it does not grow an internal planner. Roster bloat dilutes the team.
 
 **Hard-coded vendor names.** We speak in capability classes — reasoning-class, speed-class — not model names. An Eidolon must work on Claude, GPT, Gemini, a local Llama.
 
@@ -63,21 +63,25 @@ The Eidolons are opinionated. They are also humble about where the evidence is t
 | Eidolon | Role | Status |
 |---------|------|--------|
 | **ATLAS** | Scout — read-only codebase intelligence | shipped |
-| **SPECTRA** | Planner — decision-ready specifications | shipped |
-| **APIVR-Δ** | Coder — brownfield feature implementation | shipped |
+| **RAMZA** | Planner (default) — mechanized, gate-enforced specifications | shipped |
+| **SPECTRA** | Planner (opt-in fallback) — prose-methodology specifications | shipped |
+| **Vivi** | Coder (default) — loop-native brownfield implementation | shipped |
+| **APIVR-Δ** | Coder (opt-in fallback) — conservative non-loop implementation | shipped |
 | **IDG** | Scriber — documentation synthesis | shipped |
 | **FORGE** | Reasoner — deep reasoning, trade-offs, counterfactuals | shipped |
 | **VIGIL** | Debugger — forensic root-cause attribution | shipped |
+| **Kupo** | Executor — localized, verifier-backed micro-tasks | shipped |
+| **CRYSTALIUM** | Memory — shared four-layer memory substrate | shipped |
 
 See [`roster/index.yaml`](roster/index.yaml) for the machine-readable registry and [`methodology/composition.md`](methodology/composition.md) for how they work together.
 
 ---
 
 ## What you can do with them
 
-- **Plan-before-build.** ATLAS maps the unfamiliar codebase, SPECTRA turns the scout report into a decision-ready spec, APIVR-Δ implements against that spec, IDG chronicles what happened.
+- **Plan-before-build.** ATLAS maps the unfamiliar codebase, RAMZA turns the scout report into a decision-ready spec, Vivi implements against that spec, IDG chronicles what happened.
 - **Audit without touching.** ATLAS + IDG. Explore a codebase, produce a read-only findings report and a change-narrative — no code modified.
-- **Ship a feature fast.** SPECTRA + APIVR-Δ. Skip the scout if you know the terrain; plan, then build.
+- **Ship a feature fast.** RAMZA + Vivi. Skip the scout if you know the terrain; plan, then build.
 - **Bring rigor to ambiguous decisions.** FORGE. Deliberate on trade-offs, name counterfactuals, produce a verdict with confidence tier.
 
 ---

VERSION

@@ -1 +1 @@
-2.0.0
+2.1.0