Dialectic synthesis — 3.2.x design under white-team + red-team

Date: 2026-06-16. Method: a white-team squad corroborated the 3.2.x design/goals; a red-team squad was then mandated to refute them. This is the synthesis — what survives the adversarial test, what is refuted, and what remains genuinely contested. (Inputs: corroboration-* + caacs-delta-* [white]; redteam-* [red]; and ../naming-identity-ssot-strangler/ [the original investigation].)

Meta-finding (own it): the white-team squads were briefed to corroborate the operator's intuition/goals, which primed them toward agreement (red-team/priti, confirmed by commit timestamps). The corroboration is therefore weak as proof; its value is the assembled evidence, which the red team then re-read adversarially. The lesson for future scoping: run a neutral or adversarial scoping pass, not a confirmatory one.

Adjudication

#	Claim (white thesis)	Verdict	Basis
1	The static branch_naming/mid8() seam is the real, sound, load-bearing SSOT	SURVIVES	Conceded by all four red-teamers; it's the part doing the work.
2	The cheap mid8() routing of the ~14–26 bare [:8] sites is worth doing, low-risk	SURVIVES	Conceded by alphonso/debbie/priti; honest small win.
3	"Thread the ExecutionContext everywhere" is the central fix	REFUTED	Built-and-parked (2 fragment reads tree-wide, 0 on identity; git log -S "context.identity" = 0); internally split-brained (branch_name≠branch_ref.target_branch on one object, resolution.py:793-801); mutable; net more code. It's a builder redesign mislabelled as adoption.
4	Re-derivation is the root cause of the recurring defect class	REFUTED	The recurring bugs (#1949/#1589/#1990↔#1991) are wrong-logic-in-the-authority, fixed inside branch_naming.py; #1589 recurred via the seam. Threading prevents none.
5	The literal-ban ratchet is the mechanism and makes the class "can-never-regrow"	CONFOUNDED / OVERSTATED	Ratchet measures syntax, not coupling; it's a tripwire added after the cut (same squash), not its cause; passes today with 14 live [:8] sites; shape-shifting (mid[:8]/[0:8]/helper) defeats it. Keep it as a tripwire, not an oracle.
6	The strangler is measurably progressing (authorities cooled)	UNPROVEN	Per-block CC cooled for transitions/context (real), but src/ SLOC 123k→266k (+115%), core/execution_context.py deleted → +1,336-line mission_runtime/, and no legacy path retired in-range = extract-then-COEXIST, the failure precondition.
7	Status is the success model (−76% bypass)	CONTESTED → leans REFUTED	Deep imports 182→34 but facade imports 5→259, total +60%, files-coupled-to-status 44→78 (+77%). Red reads this as Goodhart/relabel; a charitable reading is correct facade adoption (more files use the facade, fewer reach inside). Unresolved without a coupling-quality (not count) analysis — do not cite −76% as settled.
8	Naming/identity is the right lead 3.2.1 slice	REFUTED on priority	It's the smallest, most-settled gap (one WP is verify-and-close) while P0 #1716 (launch-blocker), #1827 (unrecoverable merge-baseline), #1832 (read-path), #1891 and the god-modules (merge CC 102, mission 220) go untouched. Chosen for safety, not impact.
9	The 3.2.x goals are evidence-grounded continuations	WEAK / POST-HOC	The taxonomy was authored today; the corroboration was primed. The goals are legitimate operator intent, but "proven continuation" is unsupported. The underlying epics (#1619/#1666/#1868/#1878) are real and pre-existing.
10	Defer all of #1878 (coord/primary) to 3.3.x	REFUTED on sequence	The write/entry side authors identity + is the "highest blast radius" (white team's own §4) + holds the unrecoverable #1827. The read SSOT can't be more consistent than the write SSOT feeding it. Pull a bounded write-side slice forward.

What survives → the defensible 3.2.x core

• The static branch_naming/mid8() seam as the identity SSOT (already shipped).
• A cheap routing rider: route the ~14–26 bare mission_id[:8] sites through resolve_mid8/mid8(), plus a ratchet extension as an anti-regression tripwire (scoped honestly: partial, syntax-level, not a completeness guarantee).
• The DIR-031 bounded-context guardrails; the "cycle stays open" honesty about G1.

What is refuted/revised → drop or reframe

• Drop "thread the ExecutionContext everywhere" as a mission spine. It is a builder redesign (fix the internal branch_name/branch_ref inconsistency; make the composite genuinely immutable) — a separate, later, carefully-sequenced effort, applied only at the 2 sites that already hold a context, after the freeze and after the write-side authority is consistent.
• Reframe the ratchet from "the mechanism / can-never-regrow" to "a cheap regression tripwire" that measures syntax; pair it with shape-aware detection or accept its limits.
• Stop citing the confounded metrics ("−76% status", "cooled") as proof; the honest statement is "per-authority complexity improved; system-level reduction is unproven; adoption has not retired any legacy path."

Genuinely contested (needs a neutral follow-up, not a verdict here)

• Status coupling: relabel (Goodhart) vs correct facade-adoption — needs a coupling-quality analysis (does the facade hide internals, or just rename the import?).
• Whether per-authority cooling amid system growth counts as "progress" — depends on whether the growth is unrelated feature work (likely substantial) vs strangler-induced duplication.

Implications for 3.2.x scope (operator decision — see §below)

The red team did not refute that the engineering is real; it refuted the framing, the metrics-as-proof, and the sequencing. The synthesis points to: lead 3.2.1 with the high-impact/P0 work (#1716/#1827/#1832 + a bounded #1878 write-side slice), demote the naming work to a cheap routing+tripwire rider, and treat the ExecutionContext freeze as its own redesign — pending a neutral scoping pass (not a confirmatory one). The naming slice remains a legitimate low-risk opener if quick-win value + establishing the ratchet pattern is explicitly preferred over impact — but that should be a chosen trade-off, not a data-driven conclusion.