Decision records
One line per decision, full context in the linked source. A decision is SETTLED only when Yose confirms. Specs cite decisions as D-numbers. Never re-litigate a SETTLED decision in a spec; open a new record instead.
| # | Decision | Status | Position | Source / rationale |
|---|---|---|---|---|
| D01 | Blockchain for provenance? | SETTLED (Yose 2026-07-05) | No. Hash-linked transparency logs + threshold signatures + Pods. "Blockchain's auditability without blockchain's baggage." | 00_FRAMING §2 P1; consistent with Blog 03/05 (distillation §3.10) |
| D02 | Cultural data commercialization model | SETTLED (Yose 2026-07-05) | Collective corpus licensing under dewan custody (Karya pattern). Never an individual data marketplace. 70/20/10 retained. | 00_FRAMING §2 P2 |
| D03 | Consumer home assistant vertical? | SETTLED (Yose 2026-07-05) | No. BUMDes/koperasi vertical instead (same node, LPJ skill pack). | 00_FRAMING §2 P3 |
| D04 | Forum's role in agent infrastructure | SETTLED (Yose 2026-07-05) | Discovery, trust, audit metadata only. Memory lives in Pods behind CARE gates, never in PieFed's DB. | 00_FRAMING §2 P4 |
| D05 | Model gateway for Fase 0 | SETTLED (Yose 2026-07-05) | LiteLLM proxy (virtual keys, budgets, provider abstraction). OpenBao + attestation later. | 00_FRAMING §2 P5 |
| D06 | Forum scope: province vs island | SETTLED (Yose 2026-07-05) | One instance, Tanah Papua identity, region-namespaced communities, province-ready split via federation. | 00_FRAMING §2 P6 |
| D07 | Build own runtime? | SETTLED (Yose 2026-07-05) | No. Own the deterministic layers (aksara-cli, CARE engine, provenance); planner runtimes stay swappable. Version-pinned Hermes Agent for now. | 00_FRAMING §2 P7; CLIs report paradigm (b) |
| D08 | Demo-phase signing | SETTLED (Yose 2026-07-05) | Self-signed dev keys with visible PROTOKOL DEMONSTRASI labeling. No paid PSrE/e-meterai before pilot. | 00_FRAMING §2 P8 |
| D09 | Licensing split | SETTLED (Yose 2026-07-05) | ETNOS and network services AGPL-3.0 (matches PieFed upstream). aksara-cli, libraries, aksara-verify: Apache-2.0 plus trademark policy (Mozilla model, Blog 04): anyone implements, only certified builds use the Aksara name. Specs CC BY 4.0. Hardware CERN-OHL-S. Rationale: the moat is certification, trust network, and the standard itself; permissive code maximizes standard adoption, and P03 shows community origination is what cannot be forked. Consequence: v4.1 license table (AGPL cli, MPL libs) revised next edition. | Session 2026-07-05 |
| D10 | Canonical naming | SETTLED (Yose 2026-07-05) | Everything Aksara-branded. ZeroClaw and ZeroStamp retired (OpenClaw-era research carryovers, never chosen names). Verifier: aksara-verify. Runtime: "Aksara runtime, built on Hermes Agent" with loud credit. Secure element frozen: ATECC608B (608A not recommended for new designs). SKU names Pro/Aksara/Mini/Go retired in favor of v4.1 deployment packages (Paket MPP/Distrik/Kampung/...). | Session 2026-07-05 |
| D11 | ETNOS backend seam | SETTLED (Yose 2026-07-05) | Canonize the CLIs report resolution: PieFed-style ActivityPub now, ActivityPods/NextGraph later, pluggable behind a stable verb set. Master Memory's backbone claim is the target state, not the current state. | Distillation §3.2 |
| D12 | atproto agent-network README | SETTLED (Yose 2026-07-05) | External prior art only (Joel Hooks). Mine the role/trust-level tables; do not adopt the substrate (conflicts with sovereignty, offline-first, AP bets). | Distillation §3.1 |
| D13 | House style for all project docs | SETTLED (standing rule) | No em dashes, no contrast-marketing, honest labels on unbuilt things, Claude invisible in external docs, Indonesian-first user surface. | Master Memory §XII |
| D14 | Perpres 83/2025 framing | SETTLED (correction) | It establishes the digital-transformation committee. Never cite as an "efficiency mandate". Tailwind citations: GovTech/INA Digital, SPBE, Inpres 1/2025. | Domain dossier; distillation §3.9 |
| D15 | Liveness claims about ETNOS | SETTLED (honesty doctrine) | Scope all claims: frontend live on borrowed backend (piefed.social); tier 3/4 badges are stubs; pmy locale is a copy of id; dashboards demo-labeled. | Forum-readiness audit; distillation §3.5 |
| D16 | Paper submission timing | SETTLED (Yose 2026-07-05) | Defer rather than submit thin: P01/P03 wait for real pilot data. Publication order per Master Memory §X (Blog 04 first, then 02+03), arXiv mirrors within 30 days. | Distillation §3.6 |
| D17 | Colleague-memory stack | PROPOSED | Build "Memori Aksara": Graphiti-lite on SQLite (FTS5 + sqlite-vec, numpy fallback). Bi-temporal fact table (valid/invalid + recorded/superseded, invalidate-don't-delete), entity/person tables with consent + provenance, custom Hermes memory provider (Holographic SQLite provider as template). ~0.4-0.7 GB incl. multilingual embedder; extraction on resident Gemma 4 E4B with constrained JSON, nightly batch. Fallback: Hindsight (MIT, official Hermes plugin) if build stalls, cost ~1-1.5 GB embedded Postgres. LightRAG stays but ONLY as document/regulation graph-RAG, never colleague memory. Re-evaluate embedded graph DBs (LadybugDB) 2027-01. | research/2026-07-05_memory-stack.md |
| D18 | OpenViking claims correction | SETTLED (correction, refined 2026-07-05) | Two truths, two layers: data model IS markdown management (viking:// virtual filesystem, .md + .abstract/.overview tiered summaries; Yose's framing correct) BUT runtime is an AGPLv3 server with mandatory embedding+VLM models, no embedded mode ("built into Hermes" claim in our docs is wrong). Adopt the L0/L1/L2 tiered-summary CONVENTION inside Memori Aksara with zero AGPL/server baggage; keep OpenViking as lineage acknowledgment / Hub-tier candidate. Correct v4.1 §20 and Master Memory next revision. | research/2026-07-05_memory-stack.md (correction appended) |
| D19 | Display choice | SETTLED (Yose 2026-07-11, option A+B) | Demo/Prototype 1: Seeed XIAO ePaper EE03 kit, same ED103TC2 10.3 inch 1872x1404 panel class driven by ESP32-S3 display node ($85.00 + $29.90 = $114.90 verified, ~Rp 2.6jt landed, 42-46% below the IT8951 line at zero pixel/watt cost; TRMNL-style seam: OPi renders, ESP32 holds the page). Plus ONE Waveshare 7.5 inch HAT ($56.99, ~Rp 1.29jt) in the same cart to physically test the size floor ("Aksara Lite" candidate). Waveform discipline unchanged: DU partial 0.22 s NO FLASH for statline/strokes, GC16 masked as the daily page-turn ritual (partial-refresh quality through Seeed's stack = the TBV to burn down hands-on). Superseded for demo: Waveshare flexible + IT8951 (Rp 4-4.5jt line). Production path still Good Display volume + TCON; RLCD remains the endgame watch. Cheap IPS/projector rejected with receipts (projector: 260-300 real ANSI at Rp 776rb-1.3jt, ~2:1 contrast in a 500-lux office, 40-60 W). | research/2026-07-05_display-options.md, research/2026-07-11_cheap-display-paths.md |
| D20 | Memori Aksara: lane + packaging | PARTIAL (2026-07-10) | SETTLED: lane B (markdown vault = truth, ONE rebuildable SQLite sidecar; specs/memori-aksara/01_arsitektur.md); decoupled standard with aksara-cli as the only consent-gated writer; repo aksara-mem (family pattern aksara-cli / aksara-mem / aksara-ui; Yose 2026-07-10). Format spec DRAFT v0.1 = 02_format.md (conformance L0 Vault / L1 Indeks / L2 Hidup; Buku Tamu no-profiling rules). OPEN: naming pass for .aksara folder, laci names, "tidur", "buku tamu"; D17 refinement ratification (files=truth, db=index) implied by lane B but not formally stamped; tamu read-policy mechanics (security spec). | specs/memori-aksara/01-02, research/2026-07-10_memory-brain-landscape.md + _memory-substrate-team-agents.md |
| D21 | Hermes Agent integration shape | PROPOSED, H0 EVIDENCE SHIPPED (2026-07-21) | Use unmodified, version-pinned Hermes Agent as the v0.x reference reasoning runtime. aksara-agent is the Aksara distribution and policy layer: profile bundle, standalone plugins, PieFed adapter, MCP wiring, runtime-neutral envelope, and conformance suite. aksara-cli retains deterministic authority; aksara-mem retains canonical memory. No permanent Hermes fork; weekly compatibility CI tests pinned release plus upstream main. H0 private repo and 17-test scaffold validate the package boundary; ratification still waits on H1 and a private PieFed sandbox proof. | research/2026-07-21_hermes-agent-integration.md; specs/agentic/01_hermes-runtime-profile.md |