S1-L1 · OPUS OWNER
Client demand science
Client-intelligence and ProductSpec evidence packs joined to all 17 industries.
- Commercial discovery/product-spec survey
- OSS discovery/spec agents
- Industry priors and first-principles question policy
Three sprints · five Opus owners per sprint · coordinator convergence between sprints
Research in parallel. Converge with evidence.
This graph compresses the 15 moving parts into three sprints and makes the missing integration job explicit: CENA verifies and joins Sprint 1 before any dependent framework synthesis begins.
3bounded sprints
5parallel Opus owners per sprint
15permanent knowledge nodes
Research can run in parallelFramework/build consumes verified convergence
S1
Independent evidence harvest
Research the domains that do not require a finished cross-system contract. Five persistent Opus owners work in parallel.
Gate · Five verified evidence packets; no synthesis contracts promoted.
S1-L2 · OPUS OWNER
Capability supply graph
Top-100 candidate shelf with top-10 dossiers, overlap graph and industry joins.
- Repository capability decomposition
- Composite-repo overlap matrix
- User-provided SaaS/repository lists retained as a separate source lane
S1-L3 · OPUS OWNER
Experience science
Component taxonomy, preference-science report and shell/archetype option map.
- 21st refresh/dedupe/gallery research
- Choice modelling and stopping rules
- ISSO, SISOCRM, Bykonz and competitor shell comparisons
S1-L4 · OPUS OWNER
Host foundation
Data, identity/settings/navigation and connector evidence packs.
- Workload-based data study
- Host absorption and donor chrome removal
- Tenant-safe connector architecture
S1-L5 · OPUS OWNER
Editor, runtime and learning
Private/OSS editor census, runtime profiles and production-learning model.
- Open Design and comparable editor census
- Sandbox/release/rollback precedents
- Evidence feedback and asset reranking
S1-GATE · CENA coordinator
Sprint 1 evidence convergence
Turn five independent lane returns into one verified architectural handoff. Research owners supply evidence; the coordinator reconciles it; an independent verifier attacks it; Shaan decides consequential trade-offs.
Status · waiting for five lane callbacks
5lane callbacks
→12part packets
→4handoff artifacts
→5Sprint 2 edges
S1-L1
Client evidence, industry priors, question policy and candidate ProductSpec.
S1-L2
Capability shelf, top-10 dossiers, overlap graph and industry joins.
S1-L3
Component taxonomy, preference experiments and shell/archetype options.
S1-L4
Data ownership, host identity/settings/navigation and connector constraints.
S1-L5
Editor operations, runtime/release profiles and learning evidence.
Coordinator outputs
- S1-CONVERGENCE.md — Readable synthesis of shared findings, contradictions, unknowns and architectural implications.
- s1-contract-handoff.json — Machine-readable candidate inputs, outputs, invariants and unresolved fields for every moving part.
- cross-lane-dependency-matrix.json — Exact evidence and contract dependencies between Sprint 1 findings and Sprint 2 owners.
- s1-decision-gates.json — Resolved, provisional, contradictory, experiment-required, operator-required and deferred decisions.
research/workstreams/2026-08-27-sprint-1/convergence/
Promotion gate
All five lane callbacks exist, twelve part packets pass structural and claim verification, four convergence artifacts exist, contradictions are classified, and operator-required decisions are reviewed. Until then Sprint 2 remains blocked.
- Coordinator verifies packet counts, source identities, hashes, evidence classes and links.
- Claims are deduplicated and normalized without rewriting historical receipts.
- Contradictions and missing evidence remain explicit; no majority-vote synthesis.
- A fresh-context Opus verifier attempts to falsify promoted claims but owns no product decision.
- Shaan reviews decisions that change product direction, scope or irreversible architecture.
Exact handoff into Sprint 2
| Sprint 2 owner | Consumes verified parts | Produces |
|---|---|---|
| S2-L1 | P03 · P09 · P10 · P14 | P04 repo-to-block framework |
| S2-L2 | P05 · P06 | P07 design harmonization contract |
| S2-L3 | P05 · P06 · P08 · P13 | Formal P08 shell and P13 bounded-editor contracts |
| S2-L4 | P01 · P02 · P03 · P05 · P06 · P08 · P09 · P10 · P11 · P13 · P14 · P15 | P12 deterministic composer and compatibility schema |
| S2-L5 | P09 · P10 · P11 · P14 · P15 | Formal runtime, qualification, release, rollback and learning contracts |
S2
Framework and contract synthesis
Use Sprint 1 evidence to design the three missing cross-domain systems and formalize the experience/runtime contracts.
Gate · All 15 node contracts reconcile; solver dry-run returns no hidden dependency.
S2-L1 · OPUS OWNER
Repo-to-block framework
Source-shape protocol, normalization surgery map and three worked conversion traces.
- Consumes P03, P09, P10 and P14
- Service, transplant and package examples
- Branding/onboarding/settings/token/update mechanics
S2-L2 · OPUS OWNER
Design harmonization
Token extraction/mapping contract and visual-coherence experiments.
- Consumes P05 and P06 evidence
- Tests three visually different donors
- Produces semantic token and scoped bridge contracts
S2-L3 · OPUS OWNER
Shell and editing framework
Archetype/shell contract plus an upgrade-safe visual change model.
- Formalizes Sprint 1 option research
- 20/80 host/app pane and alternative shells
- Replace/add/remove/theme/text edit taxonomy
S2-L4 · OPUS OWNER
Deterministic composer
Compatibility schema, solver, UNDERDETERMINED protocol and synthetic fixtures.
- Consumes every Sprint 1 contract
- Model receives feasible sets only
- Measures glue budget and clarification needs
S2-L5 · OPUS OWNER
Runtime and learning contracts
Runtime profile, qualification, release, rollback and feedback contracts.
- Joins package/module/service execution
- Defines accepted-build and maintained-client metrics
- Independent review of framework operability
S3
Convergence and pilot blueprint
Pressure-test the complete graph against one industry workflow and produce a build-ready, independently reviewed pilot packet.
Gate · One chosen pilot, complete contracts, falsifiable evals and no unresolved critical edge.
S3-L1 · OPUS OWNER
Pilot demand and workflow
Industry choice, client workflow, archetype and acceptance fixtures.
- Scores all 17 industries
- Requires an accessible client path
- Keeps one generic dashboard as control only
S3-L2 · OPUS OWNER
Pilot supply pack
Exact service, block, components and visual bindings for the pilot.
- Pins reuse shapes
- Records normalization surgery
- No substitute or padding when supply is weak
S3-L3 · OPUS OWNER
Pilot host pack
Data, identity, settings, connectors, runtime and rollback plan.
- One owner per state resource
- Tenant-safe external actions
- Preview and production profiles
S3-L4 · OPUS OWNER
Pilot assembly and editor
Deterministic AssemblyPlan, bounded change model and complete eval sequence.
- No open-ended codegen assumption
- Compares custom build, direct generation and assembly
- Measures tokens, repairs and glue size
S3-L5 · OPUS OWNER
Independent adversary
Cross-source verification, contradiction ledger and proceed/redesign/stop verdict.
- Owns no design artifact
- Attempts to falsify every promoted claim
- Checks all 15 pages and source receipts
How every agent gets enough context
Agents receive a compact index and exact source routes—not 500 MB of raw files stuffed into one prompt.
Required context
- AGENTS.md and CURRENT_STATE.md
- knowledge/00-MASTER-SYNTHESIS.md
- knowledge/02-ASSUMPTION-LEDGER.md
- knowledge/03-EVIDENCE-MAP.md
- 17-industry specifications and niche→atom→block join
- The dedicated page and linked resources for every owned part
Required return packet
- research-report.md — source-backed findings and synthesis
- source-register.jsonl — every commercial, OSS and local source
- top-companies.jsonl and top-repos.jsonl — explicit denominators and top-10 dossiers
- innovation-register.jsonl — wide ideas, scores and retained top 10
- first-principles.md — irreducible problem, assumptions, contradictions and design
- decision-ledger.json — decisions, confidence, falsifiers and dependencies
- lane-state.json — counts, hashes, blockers, verification and callback
Compute discipline
- One persistent Opus owner per lane; workers never self-fan
- Checkpoint after prior audit, commercial survey, OSS survey and first-principles synthesis
- Exact non-overlapping write ownership by part ID
- Compact callback; full evidence remains in artifacts
- Coordinator convergence plus fresh-context verifier before Sprint 2
Location and promotion contract
Write location
research/workstreams/<part-id>-<slug>/runs/<run-id>/
Each lane may own multiple part directories, but no two live agents write the same part.
Promotion
CENA validates counts, links, source classes, contradictions and hashes, creates the four convergence artifacts, and only then unlocks Sprint 2. Historical evidence is never rewritten.