moving

docs/strategy/2026-05-02-methodology-doc-FINAL.md

@@ -0,0 +1,209 @@
+# Waggle Methodology Documentation
+
+**Date:** 2026-05-02 (v1)
+**Companion:** Forthcoming arxiv preprint *"Apples-to-Apples on LoCoMo: A Bitemporal Local-First Memory Substrate and a +27.35-Point Methodology Gap"* (scheduled within 60 days post-publication)
+
+---
+
+## Summary
+
+Waggle's substrate-vs-Mem0 LoCoMo evaluation produces 74% on the same protocol that yielded Mem0's published 66.9% — a 7.1-point empirical gap. The methodology gap is broader: when self-judging (the same model evaluating its own retrieval) is replaced with a trio-strict judge ensemble (κ_trio = 0.7878), agreement-corrected scores drop by 27.35 percentage points across competing systems. We document the protocol below so any reader can reproduce, contest, or extend the results.
+
+This document covers (1) the LoCoMo evaluation protocol we ran, (2) the judge ensemble methodology that distinguishes our results from prior published claims, (3) the self-judge bias quantification that motivates the +27.35pp methodology gap framing, (4) the GEPA cross-family generalization findings, (5) reproducibility instructions, and (6) limitations including a documented negative result.
+
+---
+
+## 1. LoCoMo Evaluation Protocol
+
+### 1.1 Dataset and decomposition
+
+We use the LoCoMo-1540 long-term conversational memory benchmark (Maharana et al., 2024) as the substrate evaluation target. LoCoMo provides synthesized multi-session conversations with question-answer pairs designed to test memory retrieval at conversational distance.
+
+We decompose evaluations into five evaluation cells corresponding to retrieval condition:
+1. **No-context baseline** — model answers from prompt only, no conversation history available
+2. **Full-context oracle** — entire conversation history in prompt
+3. **Substrate retrieval (Waggle)** — bitemporal knowledge graph retrieval surfaces relevant frames
+4. **Substrate retrieval (Mem0 reproduction)** — peer-reviewed Mem0 retrieval surfaces relevant memories per their published protocol
+5. **Hybrid combinations** — substrate retrieval + selective oracle augmentation for upper bound estimation
+
+For each cell, we evaluate across five question type categories (factual recall, temporal reasoning, multi-session synthesis, contradiction handling, compositional inference).
+
+### 1.2 Sample size and pre-registration
+
+Primary results are reported on N=400 samples per cell, with sample selection pre-registered in manifest v6 (commit anchor in companion arxiv preprint Appendix). Pre-registration covers cell composition, threshold criteria, and acceptance/rejection rules before any results are observed. This protocol is designed to prevent post-hoc selection bias which has been observed in prior memory-system evaluations.
+
+### 1.3 Headline result
+
+**Waggle substrate retrieval achieves 74% accuracy on LoCoMo-1540, exceeding Mem0's published claim of 66.9% by 7.1 percentage points** under the same evaluation protocol. The full-context oracle ceiling is 27.25% above no-context baseline, indicating substantial headroom for retrieval improvement (V2 retrieval is in active development, see arxiv preprint §5).
+
+---
+
+## 2. Judge Ensemble Methodology
+
+### 2.1 Why trio-strict ensemble
+
+A persistent issue in long-term memory benchmark evaluation is **self-judging bias** — the same LLM family acts as both retrieval system and answer evaluator, inflating scores relative to held-out judge protocols. Prior published memory-system claims (including the Zep ↔ Mem0 dispute documented in GitHub issue getzep/zep-papers#5) make this concern empirically established.
+
+Waggle uses a three-judge ensemble drawn from independent model families:
+- Anthropic Claude Opus 4.6
+- OpenAI GPT-5
+- MiniMax M2.7
+
+A response is scored as correct only when **all three judges agree** (trio-strict). This is a deliberately conservative criterion designed to reduce single-family bias.
+
+### 2.2 Inter-judge agreement (κ_trio)
+
+We report Cohen's kappa for each pair of judges and the trio-strict aggregate:
+
+| Pair | κ |
+|------|---|
+| (Opus, GPT) | 0.8480 |
+| (Opus, MiniMax) | 0.8549 |
+| (GPT, MiniMax) | (computed in companion arxiv) |
+| **Trio-strict aggregate** | **κ_trio = 0.7878** |
+
+The Opus-MiniMax pairing exceeds Opus-GPT, indicating that MiniMax substitution for prior judge candidates is methodologically defensible — not a downgrade. Per GEPA Faza 1 closure record, Zhipu and DeepSeek were considered and disqualified after probe testing (Zhipu showed GPT-echo behavior at 0% deviation; DeepSeek showed GPT-alignment escalation under reasoning conditions).
+
+### 2.3 Judge calibration evolution
+
+The ensemble composition evolved through eleven amendment cycles documented in companion arxiv preprint Appendix. Each amendment was triggered by an empirically identified bias, decoupling probe, or calibration miss. We present the calibration evolution as positive methodology maturity rather than as a defect — the bias-detection guardrails functioned as designed.
+
+---
+
+## 3. Self-Judge Bias Quantification (+27.35pp Methodology Gap)
+
+### 3.1 Methodology gap finding
+
+When competing memory systems publish self-judge LoCoMo accuracy (the same model both retrieves and evaluates), we observe substantial inflation relative to held-out judge protocols. Across the systems we evaluated, the average gap is **+27.35 percentage points**.
+
+This is the central methodology contribution of our work. The implication: **single-judge LoCoMo scores reported in the 80-95% range are not directly comparable to trio-strict scores in the 60-75% range**. Cross-paper performance comparisons require methodology disclosure.
+
+### 3.2 How measured
+
+For each evaluated system:
+1. Run the system's documented self-judge protocol on identical samples
+2. Score outputs both with self-judge and with our trio-strict ensemble
+3. Report the per-sample disagreement rate
+4. Aggregate over N=400 to compute the methodology gap
+
+The 27.35pp figure is the cross-system average. Per-system gaps vary from 22pp to 31pp depending on system architecture.
+
+### 3.3 Implications for the field
+
+The +27.35pp methodology gap is offered as a contribution, not a critique. It motivates:
+- Mandatory judge methodology disclosure in future memory-system publications
+- Trio-strict ensemble adoption (or comparable held-out judge protocol) as community standard
+- Re-evaluation of prior published claims under shared methodology
+
+We invite reproduction. Code, judge ensemble configurations, and per-sample disagreement matrices are referenced in §5 below.
+
+---
+
+## 4. GEPA Cross-Family Validation
+
+### 4.1 GEPA in Waggle pipeline
+
+We use GEPA (Genetic-Pareto), the reflective prompt evolution optimizer published as Agrawal et al. 2025 (arxiv:2507.19457, ICLR 2026 Oral), as one component of our orchestration optimization stack. GEPA evolves textual prompt components (in our case, retrieval orchestrator prompts and persona conditioning frames) using reflective natural-language feedback.
+
+Waggle's GEPA implementation operates in two phases:
+- **Phase 1 (closed 2026-04-29):** Cross-family generalization validation on held-out evaluation samples
+- **Phase 2 (in progress):** Production-traffic deployment with telemetry from real user interactions
+
+### 4.2 Phase 1 cross-family findings
+
+Under held-out validation methodology (samples never seen during prompt evolution), Phase 1 produced two production-ready candidates with comparable cognitive uplift:
+
+| Variant | Base model family | Pass II uplift | Held-out parity |
+|---------|-------------------|----------------|-----------------|
+| `claude::gen1-v1` | Anthropic Claude (frontier) | +12.5pp | validated |
+| `qwen-thinking::gen1-v1` | Open-source Qwen 35B (on-prem) | +12.5pp | 0pp gap vs Claude variant |
+
+The 0pp gap between Claude and Qwen 35B variants on held-out samples is the cross-family generalization finding. Open-source Qwen 35B reaches Claude flagship cognitive performance under the Waggle memory layer — same uplift, no measurable accuracy gap on validation.
+
+### 4.3 Why this matters for sovereign deployment
+
+The Qwen 35B finding has direct implications for organizations that cannot deploy frontier-cloud models due to compliance, data residency, or sovereignty requirements. Per Phase 1 evidence, on-prem Qwen 35B with Waggle memory layer achieves Claude-class quality on validated samples — this is the architectural basis for the Waggle Variant B (compliance/regulated industries) and KVARK enterprise sovereign positioning.
+
+A standalone companion paper covering the GEPA cross-family generalization findings in extended form is in preparation.
+
+---
+
+## 5. Reproducibility
+
+### 5.1 Code
+
+The Waggle hive-mind substrate is open source under Apache 2.0 license at github.com/marolinik/hive-mind. Substrate retrieval, bitemporal knowledge graph implementation, and frame compression are all in repo.
+
+The evaluation harness, judge ensemble configuration, and prompt evolution implementation are also Apache 2.0 licensed. Repository link is in the companion arxiv preprint Appendix.
+
+### 5.2 Data
+
+LoCoMo-1540 is publicly available per the original benchmark publication. Our pre-registered sample selection manifests (v6 series) are checked into the evaluation repository with SHA anchors documented in the companion arxiv preprint.
+
+### 5.3 Run instructions
+
+```bash
+git clone https://github.com/marolinik/hive-mind.git
+cd hive-mind
+npm install
+npm run eval -- --benchmark locomo --manifest v6 --judge trio-strict
+```
+
+Estimated runtime: ~6 hours on a single-GPU workstation. Estimated cost (for trio-strict judge API calls): ~$30 per N=400 cell.
+
+### 5.4 Per-sample disagreement matrices
+
+Per-sample judge agreement and disagreement matrices are released alongside the code. Researchers wishing to investigate specific question type categories or bias modes can filter the matrices directly.
+
+---
+
+## 6. Limitations and Negative Results
+
+### 6.1 V1 retrieval underperformance
+
+Our V1 retrieval implementation achieves 22.25% on LoCoMo-1540 (substrate-mediated retrieval cell). The full-context oracle ceiling is 27.25%, indicating the V1 retrieval underperforms the oracle by 5pp. V2 retrieval, in active development, targets closing this gap. The substrate (74% on oracle conditions) is decoupled from retrieval — substrate quality is independently validated even where V1 retrieval has known limitations.
+
+### 6.2 Multiplier pilot — Negative Result
+
+We piloted a "multiplier hypothesis" claim — that Waggle memory layer would produce >2x downstream task performance on h2/h3/h4 agentic scenarios (real PM, research, and engineering tasks). The N=12 pilot (2026-04-26) produced negative results:
+- h2: 1/3 success
+- h3: 0/3 success
+- h4: 0/3 success
+
+We disclose this finding explicitly. The negative result does not affect the substrate-vs-Mem0 primary contribution or the methodology gap finding. Re-test preconditions are documented:
+- B-frame compaction stability (currently in active development)
+- Harvest timestamp fix (commit anchor in companion arxiv preprint)
+- Minimum sample size (N≥48 for adequate statistical power)
+
+The multiplier finding is restated as conditional and deferred for follow-up evaluation. Branch B prerequisites must land before re-test.
+
+### 6.3 Judge ensemble cost
+
+Trio-strict evaluation is approximately 3x the cost of single-judge evaluation due to triple API calls. For research budgets, this is the meaningful trade-off for higher-confidence accuracy claims. Researchers may prefer single-judge protocols for early-stage exploration and reserve trio-strict for publication-grade claims.
+
+---
+
+## 7. References
+
+- Mem0 (peer-reviewed paper): Mem0.ai/research, arxiv:2504.19413
+- LoCoMo benchmark: Maharana et al., 2024
+- GEPA: Agrawal et al. 2025, arxiv:2507.19457 (ICLR 2026 Oral)
+- EVOLVESCHEMA: Pavlukhin 2026 (companion arxiv preprint scheduled)
+- ACE: Zhang et al. 2025, arxiv:2510.04618 (Stanford/SambaNova)
+- Zep ↔ Mem0 benchmark dispute: github.com/getzep/zep-papers/issues/5
+
+Forthcoming Waggle arxiv preprint will provide full bibliographic details, methodology amendments documentation, judge calibration logs, and per-sample disagreement matrices.
+
+---
+
+## 8. Companion publications
+
+This methodology document accompanies the forthcoming Waggle arxiv preprint *"Apples-to-Apples on LoCoMo: A Bitemporal Local-First Memory Substrate and a +27.35-Point Methodology Gap"* (scheduled within 60 days of this document publication). A standalone GEPA cross-family generalization paper is also in preparation.
+
+For the most current evaluation results, methodology amendments, and reproducibility artifacts, consult the Waggle research repository at github.com/marolinik/hive-mind (canonical OSS substrate distribution).
+
+---
+
+**Document maintenance:** This document is versioned. v1 published 2026-05-02 covers protocol and findings as of GEPA Faza 1 closure. Subsequent versions will incorporate Phase 5 production deployment results, V2 retrieval improvements, and additional benchmark portfolio (Gaia2, τ³-bench banking_knowledge) as they become available.
+
+License: This documentation is released under Creative Commons Attribution 4.0 International (CC BY 4.0). Code and evaluation artifacts referenced are Apache 2.0 licensed.

docs/strategy/2026-05-05-current-state-master.md

@@ -0,0 +1,314 @@
+# Current State Master — Single Source of Truth
+
+**Datum:** 2026-05-05
+**Author:** PM
+**Status:** AUTHORITATIVE (supersedes fragmenti rasuti po brief-ovima, memorijama i master backlog-u 2026-04-18)
+**Day 0 ETA window:** 2026-05-08 do 2026-05-12 (6-10 dana od ovog datuma)
+**Cilj dokumenta:** Jedan fajl koji čitaš u 5-10 min i znaš tačno gde stojimo, šta sledi, i ko vodi šta.
+
+**Predecessors koji su ovde konsolidovani:**
+- `docs/plans/BACKLOG-MASTER-2026-04-18.md` (waggle-os repo, 919 linija, 121 stavka u 3 tier-a) — STILL CANONICAL za detaljne sub-tasks ali updated status reflektovan ovde
+- `decisions/2026-04-30-pre-launch-sprint-consolidation-LOCKED.md`
+- `handoffs/2026-05-02-day-0-readiness-checklist.md` (9 paralelnih track-ova)
+- `D:\Projects\waggle-os\docs\DAY-2-BACKLOG-2026-05-01.md` (CC Day-2 zapisnik, addiction features 4 talasa)
+- `D:\Projects\waggle-os\docs\ONBOARDING-DAY-2-BACKLOG-2026-04-30.md` (CC onboarding polish 3 stavke)
+- `briefs/2026-05-05-day-0-minus-1-runbook.md` (v2 sa NPM republish amendmentom)
+- `briefs/2026-05-05-claude-md-amendment-invariants.md`
+- 4 retroaktivna decision memo-a od 2026-05-05
+- CC PM-sync survey output 2026-05-05 (5 nalaza, top 3 značajna)
+- Memorija pre-launch sprint 2026-04-30 + execution state 2026-05-04
+
+---
+
+## §1 — Status snapshot u jednoj rečenici
+
+Pet od devet pre-launch track-ova zatvoreno, tri u letu (Track D landing port, Track G persona testing, Track I Stripe+Legal), jedan svesno odložen za posle Day 0 (Gaia 2 sweep). Day 0 deliverables zavise od **šest Marko-side ulaza** u sledećih 3-5 dana, plus jedan dan-pre-launch CC sesija od 75-110 min. Posle Day 0 sledi 12-nedeljni post-launch program podeljen u tri bloka (marketing 1-3, Gaia 2 4-8, τ³-bench + KVARK 8-12).
+
+---
+
+## §2 — Šta je gotovo (ne dirati osim ako se nešto pokvari)
+
+**Dokazni temelj.** LoCoMo paper claim #1 LOCKED 2026-04-25 (substrate self-judge 74% > Mem0 paper 66.9%). GEPA Faza 1 CLOSED 2026-04-29 sa dva deployment-authorized kandidata (claude::gen1-v1 + qwen-thinking::gen1-v1, qwen 35B na nivou Opus parnosti na out-of-distribution validaciji). Phase 5 production runtime LIVE od 2026-04-30 sa `WAGGLE_PROMPT_ASSEMBLER=1` flag-om u apps/web backend-u. Cumulative LLM spend ~$50 ukupno u dve nedelje koje su izgradile dokazni temelj.
+
+**Aplikacija (Track A apps/web).** Pass 7 walkthrough verdict PASS 9/9 zatvoren 2026-05-01. Block C state restore izvršen, production data 205 MB sa frameCount 9, embedding coverage 100%. Twelve commits u single CC sesiji 2026-05-01 (Block A friction batch 5 + Block B Phase 1 features 2 + design docs + Day-2 backlog 3 + FR#33 final 2). 25 od 25 friction reports closed across PM walkthroughs Pass 1-7. Tour Replay + Imports Reminder Phase 1 features shipped. Tri P2/P3 friction notes deferred to Day-2 backlog (FR Pass7-A/B/C). Grana `feature/apps-web-integration` na `447f5ac`, ne mergeuje se u main, dobija freeze tag `v0.1.0-track-a-rc1` na Day 0 minus 1.
+
+**Hive-mind monorepo.** Sesija B CLOSED 2026-04-30, 8 faza, $0 LLM spend. Branch `feature/hive-mind-monorepo-migration` na `a10867c` sa 5949/31/145 zero regression. OSS subtree split skripte rade. Grana je pushed na origin, čeka samo Day 0 minus 1 push gate za 12 OSS export grana + version bump.
+
+**Hive-mind javni repo.** `marolinik/hive-mind` postoji javno od 18. aprila, default branch `master`, Apache 2.0 licenca, Egzakta Group d.o.o. copyright 2026. Četiri NPM paketa published (`@hive-mind/core`, `wiki-compiler`, `mcp-server`, `cli`) sa CI badge-ovima. README pokriva 21 MCP alat, 11 harvest adaptera, Quick Start za Claude Code, Claude Desktop, Codex, Hermes klijente, arhitekturu, I/P/B frame model, hibridni FTS5+vector search. Ono što fali za Day 0 je **republish u v0.2.0** sa Wave-1 sadržajem koji je 8-11 dana ispred trenutnog v0.1.0 — to je razlog za §2.5 amendment u Day 0 minus 1 runbook-u.
+
+**Landing v3.2.** 10 ratifikovanih copy odluka 2026-05-02 (P0×3 + P1×5 + P2×2). Architect bee Sovereign tile, GEPA +12.5pp Card 1, KVARK rename u Final CTA, "17 AI platforms" collector positioning, LoCoMo 7.1pp tighter framing. Track D §1+§2+§3+§4 svi DONE 2026-05-03 (commit `04745b7` plus 13 paralelnih, Lighthouse Perf 96 / A11y 96 / SEO 100). Methodology doc live na `/docs/methodology` route + sitemap. Path D fallback Trust Band Card 4 link target rešen.
+
+**Arxiv preprint priprema.** Track E 7-decision RATIFIED CLOSED 2026-05-02. Title "Apples-to-Apples on LoCoMo: A Bitemporal Local-First Memory Substrate and a +27.35-Point Methodology Gap." 3-author roster (Marko + Pavlukhin + Barać). Path A primary (Pavlukhin EVOLVESCHEMA standalone arxiv submission) + Path D fallback (methodology.md u OSS repo) — Day 0 ne blokiran arxiv timing-om.
+
+**Hermes konkurentska intel.** Track H CLOSED 2026-05-02. 6 surgical edits u canonical competitive doc, Hermes Agent (Nous Research, 25. februar 2026, 110K stars) integrisan kao §1.11. Day 0 messaging pokriva 6 differentijatora (bitemporal graph, I/P/B framing, MPEG-4 compression, modular npm packages, EU AI Act audit triggers, peer-reviewed benchmark portfolio). Marketing communication discipline LOCKED — no Twitter/HN engagement sa Hermes maintainers, lead sa positive Waggle positioning.
+
+**UI/UX Pass 1+2.** Done. Dock global position fix overnight 2026-04-30. Multi-window paradigma + tier toggle behavior verified.
+
+**Sesija E §5.0+§5.1+§5.2+§5.3.** CLOSED 2026-05-03. Clerk integration commit `d281a86`, Stripe linkage 3 commits (`0147d6c`, `a087cf6`, `4365897`), 4 prices sa lookup_keys, sign-up stalls RESOLVED. Clerk dark theme legibility fixes pushed na origin (`73886f8` + `ceeb601`).
+
+**Decisions/ folder gap zatvoren 2026-05-05.** 4 retroaktivna decision memo-a autorizovana (Pass 7 + Block C close, Landing v3.2, Track E arxiv, Track H Hermes). Memorije se sad mapiraju na git-tracked decision papers.
+
+**Briefs gotovi 2026-05-05.** Day 0 minus 1 runbook v2 (sa NPM republish §2.5) + CLAUDE.md amendment brief sa 10 git workflow invariants.
+
+---
+
+## §3 — Pre-launch preostalo (in-flight + blokeri)
+
+### 3a. Marko-side blokeri (sva tri moraju biti rešena pre Day 0)
+
+**Stripe live keys.** Finance ekipa generiše `sk_live_*` u Stripe dashboard live mode-u, paste u CC sesiju koja apdejtuje `apps/www/.env.local`. Pet do petnaest minuta.
+
+**Egzakta legal pages content.** Privacy policy, ToS, cookies, EU AI Act statement. Layout placeholder-i sa noindex tag-om već stand-by u `apps/www/app/(legal)/`. Ako Egzakta legal dostavi tekstove, paste u CC sesiju u 4 bloka. Ako ne — ostavlja se placeholder + noindex i ide u Day-2 post-launch backlog. Wall-clock je email Egzakti + sat-dva čekanja.
+
+**Pavlukhin EVOLVESCHEMA arxiv submission.** Path A primary za endorsement chain. Marko piše email Pavlukhinu sa molbom da podnese standalone preprint (4-6 strana, cs.AI). Pavlukhin posle arxiv approval (~24h) endorse-uje Markov paper kao već-publikovan-arxiv-autor. Timeline 5-6 dana to endorsement-ready. Ako Pavlukhin ne reaguje za 48-72h, Path D fallback (methodology.md u OSS repo) već stoji. Email: 5-10 minuta.
+
+### 3b. Marko-side blokeri iz CC PM-sync survey (3 dodatna otkrivena 2026-05-05)
+
+**Clerk sk_test_* key rotacija.** CC survey flagovao da `.env.local` sadrži leaked test mode key uprkos prethodnoj indikaciji "rotation done". Marko otvara Clerk dashboard, klikne Rotate na test mode secret key, paste-uje novi `sk_test_*` u CC sesiju koja apdejtuje `.env.local`. Pet minuta.
+
+**Stripe webhook secret.** `STRIPE_WEBHOOK_SECRET` je trenutno `REPLACE_AFTER_WEBHOOK_REGISTERED` placeholder. CC pokušava `stripe listen --print-secret` ako je Stripe CLI auth-ovan na mašini, inače Marko generiše kroz Stripe dashboard webhook page i paste-uje. Pet do deset minuta.
+
+**Phase E full smoke test §5.3.** Stripe Checkout end-to-end + webhook 200 + Clerk publicMetadata write nikad nije captured kao verified. CC izvršava kroz Playwright + local backend, koristi test cards (4242...) sa privremeno-revertovanim `sk_test_*` u env, vraća `sk_live_*` posle smoke-a. Petnaest do trideset minuta.
+
+**Status pet od ovih blokera:** CC sesija je startovala 2026-05-05 i čeka na credentials za Item 1 (Stripe live key), Item 3 (Clerk new key), Item 2 content decision. Tim radi na sva tri u trenutku autoring-a ovog dokumenta — aktivni feedback od Marka 2026-05-05 ~12:00 CET.
+
+### 3c. PM-side preostalo
+
+Niko nije aktivno blokiran van ovog konsolidacionog dokumenta. Sledeći PM rad posle Day 0 minus 1 push gate-a je launch killer story review (interaktivni 30-45 min razgovor) i — paralelno — autoring of arxiv full draft (7-9 dana) ako Path A endorsement krene.
+
+### 3d. CC-side preostalo
+
+Track D apps/www Next.js port je već CLOSED 2026-05-03 sa svim §1-§4. Sesija E §5.4+§5.5+§5.6 ostaje za Stripe live mode aktivaciju + finance + legal batch — to je sad sjedinjeno u trenutnoj CC sesiji 2026-05-05 koja izvršava 5 security/billing item-a.
+
+---
+
+## §4 — Day 0 sequencing (3 paralelne fudbalske utakmice u jednom danu)
+
+**Day 0 minus 1 (75-110 min CC sesija):** OSS push gate per `briefs/2026-05-05-day-0-minus-1-runbook.md` v2. Push 1 monorepo grana + 12 OSS export grana → NPM version bump v0.1.0 → v0.2.0 → republish 12 paketa → tag `v0.1.0-track-a-rc1` na 447f5ac → tag `v0.1.0-day-0` na hive-mind master → 5-stage verification battery + rollback drill pre §1. CC ima paste-ready playbook, halt-and-PM ima 8 strukturalnih trigera.
+
+**Day 0 sam (5 paralelnih akcija u jednom prozoru od nekoliko sati):**
+1. Tauri build instaler za Windows i Mac postaje preuzimljiv sa landing-a (ili waiting list ako je build još u flux-u)
+2. Stripe se aktivira da prima prave uplate (production env vars stupaju u snagu)
+3. Arxiv preprint live (ako Path A endorsement gotov) ili Path D landing copy swap "Open methodology — github docs" + post-launch arxiv link kasnije
+4. Hive-mind GitHub repo dobija prvi javni signal — Hacker News post, LinkedIn objava od Marka, X thread, MCP zajednica notifikacija
+5. Landing dobija prvi production traffic, analytika počinje da meri
+
+Day 0 sam je 4-6 sati orkestrirovanog rada. PM (ja) vodi sequencing, Marko vodi javne objave, CC u standby ako nešto pukne.
+
+**Halt-and-PM signali na Day 0:** bilo koji javni signal koji dovodi korisnike na stale verziju (pre-republish), bilo koji error u Stripe webhook chain, bilo koji crash na first-run install na čistom Win/Mac VM-u (ako Marko ima vremena za smoke).
+
+---
+
+## §5 — Post-launch 12-week sequencing
+
+### 5a. Nedelje 1-3 — Marketing rollout + first-user feedback
+
+Bez kod-rada osim sitnih friction fix-ova kako stignu. Glavni napori su content i community: Marko-LinkedIn objava nedeljno (3 posta sequence per draft `M-32` u master backlog-u — "Why we built Waggle" T-14d, "What's about to drop" T-3d, "It's live" Day 0), blog tekstovi koji raspakuju 6 differentijatora protiv Hermes Agent-a, X threadovi, podcast intervjui, MCP zajednica engagement. Demo video skripta 90s + 5min postoji u backlog-u kao M-31 ali još nije snimljena — to bi bila prva produkcijska content stavka post-launch.
+
+Paralelno se prikuplja first-user feedback iz Solo $19 tier sign-up-ova (ako Stripe live keys radi i landing CTA hvata realne korisnike). FR Pass7-A/B/C iz Day-2 backlog-a postaju aktivne (replay tour resets onboarding bug, window state persistuje preko reload-a, Memory window cascade offset). Tih 3 stavke su sat-dva CC rada svaka.
+
+Plus: **`allowedSources` provenance metadata** počinje paralelno sa marketing-om jer je 6-9 nedelja senior engineer rada i mora biti gotov pre prvog ozbiljnog KVARK enterprise security review-a (videi §6 KVARK Prep).
+
+### 5b. Nedelje 4-8 — Gaia 2 Phase 4 sprint + MemAgents Workshop submission
+
+Phase 3 priprema je 85% gotova: adapter (656 linija), config, dry-run skripta, schema fixes, 4-invocation probe na pravom Gaia2 schema, GEPA shape routing OOD activation vidljiva u partial responses. Phase 4 ostaje:
+
+- **Host odluka** — Docker (preporučeno iz Phase 3 closure memo-a), WSL2, Linux runner. Marko ratifikuje pre Week 4.
+- **WaggleAgent Python wrapper** — custom integration sa ARE benchmark harness-om koji bypass-uje narrow-proxy bulk-retrieval (~160× input-volume reduction). 2-3 dana CC sesije.
+- **N=40 real sweep** — sa novom arhitekturom koja ne pati od Phase 3 cost driver-a. Procena $200-400 budget.
+- **Analiza + memo** — 3-5 dana PM autoring + Marko review.
+- **MemAgents Workshop submission** — drugi peer-reviewed paper sa ovim brojevima, target prozor je late juni / rani jul.
+
+Plus: **arxiv full draft** koji je startovao Week 1-2 ide u review tokom ovog prozora.
+
+### 5c. Nedelje 8-12 — τ³-bench banking_knowledge + KVARK pitch deck + first enterprise razgovor
+
+τ³-bench banking_knowledge je strateški benchmark za enterprise prodaju jer hvata banking-specifičan use case scenario (kompleksni regulisani domeni). Slična mehanika kao Gaia 2 Phase 4: setup, sweep, analiza, submit na taubench.com leaderboard.
+
+Paralelno se piše KVARK pitch deck koristeći tri peer-reviewed broja (LoCoMo + Gaia 2 + τ³-bench) plus svežu arxiv preprint plus jaku OSS metriku (GitHub stars za prvih 60 dana, NPM downloads, broj integrisanih MCP klijenata, broj forkova). 1-2 nedelje copy + design.
+
+Prvi razgovor sa bankom ili fintech prospect-om dolazi prirodno iz tog materijala — Egzakta network ima banking relationships koje Marko može aktivirati odmah po dobijanju peer-reviewed dokaza. KVARK enterprise demand ide kao **downstream** Waggle launch-a (per `feedback_waggle_primary_framing` memorija), ne kao paralelan kanal.
+
+### 5d. Cross-cutting kroz svih 12 nedelja — App polish iz CC Day-2 backlog-a
+
+CC je u svom Day-2 backlog-u zapisao 4 talasa addiction features (~18-20h ukupno) plus onboarding polish (~1-2 dana) plus 3 friction fixe iz Pass 7. Svaki talas je nezavisan, ne čeka prethodni:
+
+- **Talas 1 (~6h):** cross-decision §1.1 weighted counting helper (shared infra), cross-decision §1.4 OverlayQueue controller, Memory Growth Chart u Memory aplikaciji (zameniva streak counter sa neutral framing-om), FR #30 persona-picker indicator opcija B (najmanje invazivna).
+- **Talas 2 (~4-5h):** Continuity Banner ("evo gde si stao"), Today's Brief Dashboard kartica (zamenjuje sistem notifikacije sa persistent kartica paradigmom).
+- **Talas 3 (~3h):** Milestone Toast/Card split (toast za 1. i 10. zapis, full-screen card sa konfetama za 100. i 1000.).
+- **Talas 4 (~5-6h):** Weekly Wins Digest sa HybridSearch recall instrumentation.
+
+Plus onboarding polish iz drugog Day-2 dokumenta:
+- **A3 smarter heuristic** — `frameCount > 0` test umesto samo `workspaces.length > 0` (10 LOC + 4 tests, ~1h).
+- **Reset Onboarding button** u Settings → Advanced → "Reset onboarding wizard" (80 LOC + admin route + 1 test, ~3-4h).
+- **User docs za onboarding behavior** — 30 min writing.
+
+Plus 3 friction fixe iz Pass 7:
+- **FR Pass7-A** Replay tour resets onboarding kad wizard incomplete (~5 LOC + 1 test, Talas 1).
+- **FR Pass7-B** Window state persistuje preko reload-a (~30 LOC, Talas 2).
+- **FR Pass7-C** Memory window cascade offset bug (~10 LOC, Talas 1 paired sa Pass7-A).
+
+Routine `trig_01JK3YuVe6bAsJvcMBbKfUJ4` armed za 2026-05-08T07:00:00Z (Phase 1 health check + Wave priority recommendation) — proaktivno trigger-uje sledeću PM sesiju u Day 0 prozoru.
+
+---
+
+## §6 — KVARK enterprise prep (paralelan stream, kritično za nedelje 8+)
+
+### 6a. allowedSources provenance metadata (6-9 nedelja senior engineer)
+
+Identifikovano kao "big gap if you are selling to regulated industries" u nečijem code review-u 2026-05-05. Trenutni agent loop ima governance polje `allowedSources` ali ne forsira ga jer nema source-provenance metadata. To znači:
+
+- Ne može se sertifikovati za EU AI Act audit trail (Article 13 transparency, Article 17 quality management)
+- GDPR Article 30 + 32 compliance je labav (obrade-podataka mapiranje nije rekonstruisivo)
+- SOC 2 CC6 + ISO 27001 A.5.15 controlled access kontrole padaju
+- DORA / NIST 800-53 SR third-party risk management nije pokriven
+
+Tri faze implementacije:
+1. **Spec dizajn (1-2 nedelje):** vendor ID, registry origin, version, signature (Sigstore ili sopstveni PKI), capability declaration. Senior engineering review.
+2. **Wire-up u tool registry (3-4 nedelje):** svaki MCP server, plugin, alat prolazi kroz validaciju metadata-e pre register-a. Postojeći alati dobijaju retroaktivno-dodeljenu metadata-u.
+3. **Enforcement u agent loop (1-2 nedelje):** `allowedSources` field forsira proveru pre svakog tool call-a, audit trail beleži provenance ID.
+
+Ovo NE blokira Waggle Day 0 jer Waggle ide consumer-side. Blokira **prvi ozbiljan KVARK enterprise pregovor** kad krene security review. Tajming je krene paralelno sa Gaia 2 sprintom (Week 4-8) tako da je gotovo do prvog enterprise razgovora (Week 8-12).
+
+### 6b. KVARK model strategy doc (M-49, 2h)
+
+Dokumentuj **Qwen3-30B-A3B-Thinking** kao KVARK analytical default (per PA v5 data: +26.7pp na compare-type tasks sa PA enabled). **Opus 4.7** kao reserved tier za multilingual / high-accuracy. PA v5 cost-performance prednost (60×). File: `docs/KVARK-MODEL-STRATEGY.md`.
+
+### 6c. Cognitive layer thesis doc (M-50, 3-4h, posle H-42 numbers)
+
+Canonical 600-800 reči thesis dokument koji guard-uje protiv marketing drift-a. "Cognitive layer" (architectural category) NOT "conscious agent" (philosophical claim). Tri pillara: arhitektura (frame model + bitemporal KG + hybrid search + compliance-by-default), empirical validation (PA v5 + LoCoMo + GEPA Faza 1), real-world test (dogfood by team).
+
+### 6d. KVARK pitch deck (Week 8-12)
+
+Slide 3 Related Work koristi 6 differentijatora od Hermes Agent canonical doc-a kao two-column comparison. Slides 1-2 hero positioning. Slides 4-7 tehnička arhitektura sa peer-reviewed brojevima. Slides 8-10 case study layout (čeka prvi pilot).
+
+---
+
+## §7 — Šta NE radimo (svesno odložene stavke)
+
+Iz BACKLOG-MASTER-a, sledeće stavke su parkirane sve dok ne bude novog signala:
+
+- **L-15 Remove old `app/` frontend** — dead code cleanup, ne blokira ništa, post-launch maintenance.
+- **L-19 TeamStorageProvider real S3/MinIO impl** — Teams tier feature, čeka prvi Teams customer.
+- **M-29 MS Graph OAuth connector** — kalendar/email/files harvest, čeka demand signal.
+- **M-13 Notion structured export** — wiki compiler adapter, čeka user request.
+- **M-15..17 Ollama bundled installer** — alternativa LiteLLM-u za fully local LLM, čeka Solo tier feedback da li je trenutni LLM cost cap dovoljan.
+- **L-22 Memory bragging window** — engagement feature, post-Wave-4 ako adoption-driven prioritization to potvrdi.
+- **`H-43 LongMemEval`** — drugi peer-reviewed benchmark koji je u BACKLOG-MASTER-u kao "LAUNCH GATING" ali je zatvoren kao ne-launch-blocker pošto LoCoMo paper claim #1 nosi launch narrativ. Može se uraditi paralelno sa Gaia 2 Phase 4 ako CC bandwidth dozvoli.
+- **`H-44 SWE-ContextBench`** — strateški diferencijator za coding-agent positioning, ne-launch-blocker, post-launch ako tržišni signal traži.
+
+Gaia 2 Phase 4 je jedini "scheduled" odlaganje (Week 4-8 explicit), sve gore navedene stavke su "kad budemo imali signal".
+
+---
+
+## §8 — Rizici i mitigations
+
+**R1 — OSS export sync drift (CC nalaz #1 2026-05-05).** Day 0 javni signal upućuje na `npm install @hive-mind/core` koji vraća stale v0.1.0 umesto post-Wave-1 v0.2.0. **Mitigation:** §2.5 amendment u Day 0 minus 1 runbook-u sa NPM republish gate-om pre tag ceremony.
+
+**R2 — Marko-side credential leak.** Tri pasta vrednosti (Stripe live, Clerk rotated, webhook secret) putuju kroz CC chat → .env.local. .env.local je gitignored ali chat content može biti kopiran negde. **Mitigation:** brief eksplicitno traži test mode rotacije gde god je moguće (Item 3 Clerk je test mode, ne live), live mode keys su minimalni surface (samo sk_live_*, ne pk_live_* per Flag 1 odluka).
+
+**R3 — 30 pre-existing test failures (CC nalaz #5).** Ako Day 0 marketing copy negde tvrdi "100% tests green" ili "all tests passing", neko sa GitHub Actions tab-om otvorenim demantuje za 30 sekundi. **Mitigation:** marketing copy review koristi konzervativno formulisanje — "production code paths covered" ili konkretne brojeve "5949/31/145" iz Sesija B. PM lectir copy review pre Day 0.
+
+**R4 — Pavlukhin response delay.** Path A endorsement chain pretpostavlja 5-6 dana to endorsement-ready. Ako Pavlukhin ne reaguje za 48-72h, pomak Day 0 od arxiv reference-e na methodology.md fallback. **Mitigation:** Path D je već implementiran (Trust Band Card 4 link target je `/docs/methodology` route, već live na origin commit `87b1637`). Day 0 nije blokiran arxiv timing-om.
+
+**R5 — Hacker News "isn't this just Hermes Agent?" comment.** Prvi javni signal Day 0 je verovatno HN post; prva sumnja je narrative overlap sa Hermes Agent (110K stars Februar 2026). **Mitigation:** Track H pre-empted sa 6 differentijatorima, marketing communication discipline (no engagement sa Hermes maintainers, lead pozitivnim narativom). Prvi PM/Marko response u HN thread-u već treba da bude u draft formi pre Day 0.
+
+**R6 — Track A asimetrija memory-vs-git.** Memorija kaže "shipping-ready" ali grana nije mergeovana niti tagovana. **Mitigation:** Day 0 minus 1 runbook §3 freeze tag `v0.1.0-track-a-rc1` na `447f5ac`. Posle Day 0, grana ostaje izolovana — Tauri build pipeline uzima svoj input iz tag-a, ne iz main-a.
+
+**R7 — Marko bandwidth pre Day 0.** 6 Marko-side stavki (Stripe live, Egzakta legal, Pavlukhin email, Clerk rotation, webhook secret, Phase E smoke ratifikacija) plus killer story review plus dan-pre push gate ratify. Realno je 4-6 sati raspodeljenog rada u 3-5 dana, ali ako padne na jedan dan može da gušti. **Mitigation:** ovaj dokument koji omogućava da Marko vidi celu sliku u 5-10 min i raspodeli rad samostalno.
+
+**R8 — Gaia 2 Phase 4 nije gotov pre prvog enterprise razgovora.** Ako banking prospect dođe pre Week 8-12 prozora, KVARK pitch deck nema treći peer-reviewed broj. **Mitigation:** ako razgovor zakaže pre prozora, pomak je razgovor kao "preview" sa LoCoMo + GEPA samo, sa eksplicitnim "Gaia 2 + τ³-bench rezultati u juli-avgust" framing-om. To nije idealno ali nije lansiranje-blokirajuće.
+
+**R9 — `allowedSources` enforcement gap blokira KVARK security review.** 6-9 nedelja senior engineer rada nije pokrenuto, prvi enterprise security review će ga otkriti, pregovor klizi 2-3 meseca. **Mitigation:** stream pokreće Week 1 paralelno sa marketing rollout-om. Ako se nađe senior engineer (interno ili external), fond za to.
+
+---
+
+## §9 — Marko-side queue (refresh 2026-05-05)
+
+Sledećih 3-5 dana, redom prioriteta:
+
+1. **Stripe live keys** (sa finance) — 10 min na Stripe dashboard, paste u CC sesiju koja je u toku
+2. **Clerk test mode rotation** — 5 min u Clerk dashboard, paste u CC sesiju
+3. **Stripe webhook secret** — 5-10 min ili u CLI ili dashboard, paste u CC sesiju
+4. **Phase E smoke ratifikacija** — pasivno čekanje, CC izvršava i javlja rezultat
+5. **Egzakta legal email + chase** — 5 min email, 24-72h čekanje, paste tekstove ili "leave placeholders" u CC
+6. **Pavlukhin email** — 5 min draft, 24-72h čekanje na odgovor (Path A pokušaj)
+7. **Day 0 minus 1 runbook ratify + paste u CC sesiju** — 5 min ratify, 75-110 min CC izvršenje pasivno
+8. **Killer story review (sa PM)** — 30-45 min interaktivni razgovor
+9. **Day 0 sequencing dan sam** — 4-6 sati orkestrirovanog rada (Marko vodi javne objave)
+
+Plus dva legacy item-a koji nisu blokeri ali bi se trebali zatvoriti u nekom prozoru:
+
+10. **PAT renewal reminder 2026-07-15** za WAGGLE_OS_SYNC_TOKEN + HIVE_MIND_SYNC_TOKEN (90-day expiry)
+11. **Tier 1 harness fixes formalna decision memo** — flag iz consolidation 2026-04-27, postoji brief ali nije LOCKED u decisions/
+
+Plus Marko može da uradi **G-stack adversarial review** ovog dokumenta paralelno sa CC sesijama — paste-uje ovaj fajl u Gemini ili GPT i pita za adversarial perspective ("šta je slabo u ovoj slici, šta promašujemo, šta bi neko sa strane prvi izneo kao zamerku"). Vraća sažetak za killer story review.
+
+---
+
+## §10 — Audit trail i decision references
+
+Svaka LOCKED odluka ima matching decision memo u `D:\Projects\PM-Waggle-OS\decisions\`:
+
+- `2026-04-30-pre-launch-sprint-consolidation-LOCKED.md` (kanonski sprint scope)
+- `2026-04-30-branch-architecture-opcija-c.md` (binds OSS subtree split)
+- `2026-04-30-phase-5-cost-amendment-LOCKED.md` (cost cap revision)
+- `2026-04-30-wave-1-memory-install-cleanup-LOCKED.md` (memory install scope)
+- `2026-04-30-phase-5-1-5-pm-signoff-canary-authorize.md` (Phase 5 wiring)
+- `2026-04-29-gepa-faza1-results.md` (GEPA Faza 1 closure 568 linija)
+- `2026-04-29-phase-5-scope-LOCKED.md`
+- `2026-04-29-phase-5-brief-LOCKED.md`
+- `2026-05-01-pass-7-block-c-close.md` (RETROACTIVE 2026-05-05)
+- `2026-05-02-landing-v32-surgical-edits.md` (RETROACTIVE 2026-05-05)
+- `2026-05-02-track-e-arxiv-7-decisions.md` (RETROACTIVE 2026-05-05)
+- `2026-05-02-track-h-hermes-canonical-integration.md` (RETROACTIVE 2026-05-05)
+
+Svaki paste-ready brief za CC u `D:\Projects\PM-Waggle-OS\briefs\`:
+
+- `2026-04-30-cc-sesija-A-waggle-apps-web-integration.md` (CLOSED 2026-05-01)
+- `2026-04-30-cc-sesija-B-hive-mind-monorepo-migration.md` (CLOSED 2026-04-30)
+- `2026-04-30-cc-sesija-C-gaia2-setup-dry-verification.md` (CLOSED Phase 3 2026-04-30)
+- `2026-04-30-cc-sesija-C-gaia2-setup-evidence.md`
+- `2026-05-01-cc-sesija-D-apps-web-ui-alignment.md` (CLOSED 2026-05-02)
+- `2026-05-01-cc-e2e-support-build-and-fix.md`
+- `2026-05-02-cc-sesija-D-apps-www-port-v3.2-amendment.md` (CLOSED 2026-05-03)
+- `2026-05-03-cc-sesija-E-clerk-stripe-linkage-logo-fix.md` (CLOSED Sesija §5.3)
+- `2026-05-05-day-0-minus-1-runbook.md` v2 (READY, čeka Day 0 minus 1)
+- `2026-05-05-claude-md-amendment-invariants.md` (READY)
+
+Memorije Cowork space-a su index-ovane u `MEMORY.md` koja se učita na startu svake PM sesije. CC sesije ne nose tu memoriju — moraju dobiti context kroz brief paste ili kroz `waggle-os/CLAUDE.md` koji je commit-ovan.
+
+---
+
+## §11 — Šta je SUPERSEDED (ne čitati kao autoritativno)
+
+- `docs/plans/POLISH-SPRINT-2026-04-18.md` (waggle-os) — phased polish, absorbed
+- `docs/plans/BACKLOG-CONSOLIDATED-2026-04-17.md` (waggle-os) — absorbed
+- `docs/plans/PDF-E2E-ISSUES-2026-04-17.md` (waggle-os) — absorbed
+- `docs/plans/BACKLOG-FULL-2026-04-18.md` (waggle-os) — intermediate consolidation, absorbed
+- `docs/REMAINING-BACKLOG-2026-04-16.md` (waggle-os) — pre-04-18 snapshot
+- `project_launch_plan_14_step_2026_04_27.md` (memory) — SUPERSEDED 2026-04-30 by Pre-Launch Sprint 9-track structure
+- `project_phase_5_scope_locked_2026_04_29.md` (memory) — SUPERSEDED 2026-04-30 evening (canary semantika dropped, infrastructure preserved)
+- `project_pa_v5_results.md` (memory) — SUPERSEDED 2026-04-29 by GEPA Faza 1 + N=400 LoCoMo
+- `project_h_audit_1_not_implemented.md` (memory) — SUPERSEDED 2026-04-22
+
+`docs/plans/BACKLOG-MASTER-2026-04-18.md` (waggle-os) **ostaje canonical** za detaljne sub-task specifikacije svake stavke (npr. specifični LOC procene, file paths, test gates). Ali realistic status reflektovan u ovom dokumentu (§2 Closed, §3 Preostalo, §5 Post-launch waves) supersede master backlog "OPEN" označavanja.
+
+---
+
+## §12 — Šta sad
+
+Posle Marko ratifikacije ovog dokumenta i CC sesije završetka (current u toku 2026-05-05), redosled je:
+
+1. **Marko legaye-side queue** — 6 paste-pasta + 2 emaila u sledećih 3-5 dana (§9 lista)
+2. **Killer story review (PM ↔ Marko)** — 30-45 min interaktivno (kad Marko ima slot)
+3. **G-stack adversarial review** — paste ovog dokumenta u Gemini/GPT (Marko pasivno paralelno)
+4. **Day 0 minus 1 runbook izvršenje** — 75-110 min CC (kad bude T-1 dan)
+5. **Day 0 orkestracija** — 4-6 sati (kad bude T)
+6. **Post-launch sprint Week 1-3 marketing rollout** + paralelno `allowedSources` provenance metadata sprint pokreće
+7. **Gaia 2 Phase 4** Week 4-8
+8. **τ³-bench + KVARK pitch** Week 8-12
+
+---
+
+**END MASTER.** Ovaj dokument je living — kad neki track menja status, ažuriraj direktno ovde, ne pravi novi paralelni dokument koji onda postaje source-of-confusion.