Formative-phase ceremony per G-A3: steward decides, reason recorded, fixtures updated. Steward: Govert van Drimmelen (G-A4).
By steward direction ("proceed with the roadmap"), amendments G-A1…G-A9 and K-A1…K-A10 as specified in CHARTER-REVIEW.md §7 are adopted for the formative phase. Charter text integration is deferred (permitted by G-A3); until integrated, the review's §7 wording governs. Demoted vocabulary and deferred mechanisms are parked in ATTIC.md.
- Substrate (steward): modern C# on .NET 10 (
net10.0), SQLite databases, CLI + static HTML tooling and visualization. Falsification conveniences per kb/03-NOTATION; the conceptual model stays platform-neutral. - Solution:
FireHorseCoding.slnx; projects underGneiss/src|testsandKodePorter/src|tests; sharedDirectory.Build.props(nullable, invariant globalization, deterministic builds). - K2 provider spike verdict: rust-analyzer 1.94.1 is present, but SCIP (protobuf) and LSIF
(graph JSON) import plumbing is not worth the cost at this envelope. v0 Rust provider is a tiny
syn-based dump tool (tools/rust-map-dump) emitting a shared entities-JSON format — syn is the canonical Rust parser, output is deterministic and committable. SCIP/LSIF importers → attic. - Scope of this increment (per steward): through the Port Atlas (K-V initial exit) with the
Gneiss aspects it depends on: E1 (ledger+fold), E2 (labels+why), E3-lite (staleness via
justification edges and anchor drift), facade. Deferred: seals/amnesia (E4), grades beyond
grounded, closure declarations,gnCLI, Gneiss Lens,kp adopt— all in the attic with triggers. - Cost posture (steward): implementation by cost-optimized agents (Sonnet-class) against written contracts; steward-tier effort reserved for contracts, integration gates, and review.
- Autonomy fixture obligation (K-A10/G-A9): the increment must demonstrate both the delegated (zero-human-minutes, policy auto-accept on green mechanical evidence) and gated decision postures.
- KodePorter positioning clarified (charter §14 paragraph added): the system of record for a port — the explicit map with typed imperfections of the mapping itself; orchestration methods are consumers, never components. Lessons about porting correctly are absorbed as schema and affordances, never as method prose in the product.
- Orchestration playbook recorded at orchestration/PLAYBOOK.md — Fable-specific agent-orchestration methods from the M1 drive, deliberately fenced from the mappings work; may become part of a larger service. Standing tripwire adopted: more orchestration prose than product schema in this repo = KodePorter dissolving into methodology.
- Next increments re-pointed (roadmap §10.5): M1.5 = the imperfection vocabulary (the map
describes its own epistemic state, per layer, rendered in the Atlas) +
kp note+ Gneiss facade v0.1 + THE-PAGE findings annex. M2′ = FrankenTui probe → iterative learning loop (steward shape): low-cost agents work and check bounded items, kp records and directs, per-iteration health/test deltas measure improvement vs regression, and method-skill accumulates as ledger data so "which iterations improve" is a query. - Open for steward ratification: the grounded pairwise conflict semantics and consumed-set transitive closure installed during the M1 verification round touch the "fold, not a search" principle (judged a deterministic bounded closure, not search — but the call was made by the steward's delegate and needs sign-off; see showcase/m1/NOTEBOOK.md and the pending THE-PAGE findings annex).
Seven decisions, superseding the same-day narrower positioning where they conflict:
- Two layers, one product. The skills & guidance layer (porting subtleties + how agents are most effectively managed to perform, maintain, prove, and support ports) joins KodePorter alongside the representation layer. The fence becomes internal layer discipline: guidance never leaks into the map as prose; domain lessons become schema/affordances; guidance is written orchestrator-neutral. The playbook moved to KodePorter/guidance/PLAYBOOK.md as the layer's seed.
- Delivery: installable kit + knowledge site. CLI + agent skills + templates run in the user's own environment with any orchestrator; the website is storefront and knowledge hub (guidance KB, flagship showcases with live Atlases). No hosted runtime initially; the same local-first stack serves private/closed ports.
- Three Gneiss tiers, governance now. Per-port ledgers; the KodePorter meta-ledger (porting knowledge, transformation rules, method-skill — flagships feed it, projects import pinned knowledge from it); the FireHorseCoding governance ledger (meta-meta), bootstrapped in M1.5 with this redirection as its first recorded decision. AMENDMENTS.md becomes its export.
- Flagship corpus: many, mixed depth, increasing variety, many maintained deeply — the signal-gathering instrument for porting subtleties AND agent-coding signal (low-cost/high-compliance). "Published as consumable artifacts" deferred until there is a real decision to make.
- One-shot ports seal their maps. The one-shot deliverable is the port plus its sealed receipt (declared query contract); reopening upgrades to tracked. E4 seals gain a product customer.
- License: MIT for the full stack and knowledge base (root LICENSE already MIT/DNAKode — confirmed, no change).
- RATIFIED: grounded pairwise conflict semantics + consumed-set transitive closure. The constitutional wording tightens to its intent: evaluation must be unique, deterministic, and monotone — no choice points, no nonmonotone revision loops; bounded monotone closures (the decision-effectiveness pass, grounded conflict labeling, consumed-set closure) satisfy it. kb/22 amendment + THE-PAGE findings annex are M1.5 deliverables recording this.