Quant Platform Roadmap¶
This document is the macro product-progression view. The authoritative atomic dependency/execution model is CAPABILITY_DAG.md. Exact semantic authority remains in accepted ADRs/contracts; SCOPE.md activates bounded work.
Planning layers¶
ROADMAP.md = product progression
CAPABILITY_DAG.md = atomic capability dependencies + decision gates + verticals
CAPABILITY_MAP.md = compact current-state snapshot
OPEN_DECISIONS.md = unresolved decisions / atom-specific gate triggers
ADRs + contracts = semantic authority
SCOPE.md = active bounded authorization
A numbered phase is not an executable work packet. Parallel work is allowed when Capability DAG dependencies are satisfied, but the active scope may impose a stricter one-slice-at-a-time policy.
Macro phases 0–15¶
- Repository and governance foundation
- Data access and identity foundation: DataGateway, catalog-backed reads and provenance
- Representation foundation: representation identity, CandleDefinition and deterministic historical/live semantics
- Feature foundation: FeatureDefinition, FeatureSetDefinition, FeatureArtifact, provenance, providers and materialization
- Research and outcome foundation: Hypothesis/Event/Outcome identity, reproducibility, event studies and sweeps
- Validation and labeling: availability, warmup, walk-forward, purge, embargo, lockbox and censoring
- Strategy and decision foundation: StrategySpec, policies, DecisionIntent, risk and sizing
- Execution, replay and portfolio: orders, fills, positions, accounting and deterministic replay
- Unified Experiment Orchestration and Persistence: Study, Trial, Run and Artifact identity/persistence/comparison/reproduction
- Canonical API and job runtime
- Supervised ML
- Strategic RL
- Execution RL
- Clients: CLI, TUI and App UI
- Paper and shadow operation
- Live operation
These phases remain intentionally stable. The DAG determines actual dependency order inside and across them.
Completed foundation checkpoints¶
Contract Freeze Gate PASSED
Conformity Implementation Gate PASSED
Human / Golden Bybit BTCUSDT E2E PASS
Adversarial Acceptance A1-A9 PASS
Candle Ordering Compatibility PASS
Package Boundary / Modular Monolith Foundation v1 COMPLETE
Legacy Capability Harvest Audit v1 COMPLETE
ASS-01 Application ownership/enforcement COMPLETE
ASS-02 in-process Application service (C02 + C03) COMPLETE
K11 Governance-state consistency COMPLETE
E02 FeatureDefinition v1 semantic foundation COMPLETE
D06 FootprintDefinition v1 representation COMPLETE
B04 Non-contiguous coverage reads v1 COMPLETE
K06 RAW / source protection v1 COMPLETE
A10 Backfill / repair v1 COMPLETE
E04 FeatureArtifact v1 COMPLETE
E06 H01 canonical integration COMPLETE
F02 EventSpec/detection COMPLETE
F03 OutcomeSpec/Outcome COMPLETE
F04 Event studies and parameter sweep runtime v1 COMPLETE
F07 labels/censoring/lockbox COMPLETE
F08 DSR/PBO robust comparison COMPLETE
K08 backup / restore v1 COMPLETE
A11 Bybit live trades acquisition v1 COMPLETE
K02 live-ingest runtime identity v1 COMPLETE
K10 checkpoint / recovery v1 COMPLETE (real restart proof: PR #122)
Live Ingest Server Production Readiness v1 COMPLETE for first producer operating path
Wave 4 Strategy / Replay COMPLETE (G01-G04,H01-H05; V7 Golden PASS)
Wave 6 Live Consumer Data Plane & Storage Lifecycle COMPLETE (B06,D04,K07,K09; Golden PASS)
Wave 7 API & Platform Transport v1 COMPLETE (J02,J04,J05,J06; exceptional Golden waiver)
Omega external-audit stabilization line COMPLETE (not a wave; tracking #232; ADR-0054-0061 and
(implement/omega, PR #254,#256-#278) ADR-0037/ADR-0050 amendments; see Decision-gate model)
DG-J Remote Service Topology design gates RESOLVED (issues #282-#287/#288; ADR-0062-0067; G1-G6;
(implement/omega, PR #289-#294) J11 deferred; Wave 8 implementation inventory authorized)
Wave 8 Full Platform API & Remote Service COMPLETE (J03,J09,J10,J12-J15,K12,K13; operator closeout #327;
Topology v1 (PR #322, #326) J10/J12/J13 in-process only; no Golden proof claimed)
DG-K Process Topology v1 COMPLETE within recorded bounds (operator closeout #381; ADR-0068,
ADR-0069; P01-P06 promoted into main via #434/#437;
later composition credited by DG-K2, no Golden proof claimed)
DG-K2 Server/deck deployment composition COMPLETE within recorded bounds (operator closeout #409; ADR-0072 and Amendment 1;
S01-S07 on implement/dgk2; supervised proof: one happy path; S08 not authorized)
DG-M Canonical feature identity (implement/fid) COMPLETE within recorded bounds (closeout #475; ADR-0073 and Amendment 1; PR #466/#468/#473/#474);
M03 superseded; no real-host run claimed; promotion to main is separate
DG-N Multi-instrument Bybit linear trades (implement/fid) CLOSED within bounds (#483, #489, #527; ADR-0074 ACCEPTED #377, Amendment 1); T01,T04,T03,T02 COMPLETE (PR #486, #488, #492, #494-#502, #526); no real-provider run
DG-O Live publication (implement/fid) CLOSED within bounds (#507, #527; ADR-0075 ACCEPTED #504); O01,O02,O03 COMPLETE (PR #510, #524, #526); T02 closed with O03; no real-provider run
DG-R Rebuild of damaged legacy live days (implement/fid) AUTHORIZED (#518; ADR-0074 Amendment 2 ACCEPTED #515); R01,R02 not started; post-version
DG-C L2 historical v1 (implement/l2, #442) COMPLETE (A14,A14a-A14e; PR #375/#379/#401/#411/#417/#408;
golden #413/#430, 11/11 at 1cdd78f; explicit gaps);
A14f/A14g held MISSING; promotion to main is separate
Wave 1 implementation batch CONCLUDED
Wave 2 implementation batch COMPLETE
Wave 3 implementation batch COMPLETE
Broad independent Producer/Consumer expansion UNLOCKED
Wave 3 is implementation-complete through F08. The selected Live Ingest Vertical v1 authority set remains:
ADR-0039 K08 backup / restore v1
ADR-0040 A11 Bybit live trades acquisition v1
ADR-0041 K02 live-ingest runtime identity v1
ADR-0042 K10 checkpoint / recovery v1
PR #116 has integrated the Live Ingest implementation branch into main. K08, A11, K02 and K10 are now implementation COMPLETE: PR #122 (issues #109, #121) supplies K10's previously pending bounded real-server restart/deployed-checkpoint proof.
Completed checkpoints are credited and must not be re-proved absent a concrete invalidating change.
Roadmap vNext planning state¶
TOTAL_ATOMS = 132
CLASSIFIED_ATOMS = 132
UNCLASSIFIED_GAPS = 0
SEMANTIC_FROZEN_OR_RESOLVED = 120 / 132 = 90.9%
OPEN_BLOCKING = 3
OPEN_DEFERABLE = 9
ROADMAP_DEFINED = 120 + 3 + 9 = 132
ROADMAP_PLANNING_COMPLETENESS = 132 / 132 = 100%
IMPLEMENTATION_COMPLETE = 110 / 132 = 83.3%
DG-N/DG-O closeout (#527): T03, T02, O01, O02 and O03 are credited COMPLETE on implement/fid within their bounds (PR #492, #494 to #502, #510, #524, #526;
105 -> 110 of 132; implementation 79.5% -> 83.3%); no new atom and no decision-state change. The one-hour ETHUSDT real-provider run is not performed and not claimed.
Earlier, DG-R registration (#518): ADR-0074 Amendment 2 accepted; R01 and R02 added as FROZEN/MISSING (130 -> 132 atoms; semantic 90.8% -> 90.9%;
implementation 80.8% -> 79.5%); nothing is implemented.
Earlier, DG-O registration (#507): ADR-0075 accepted; O01-O03 added as FROZEN/MISSING (127 -> 130 atoms; semantic 90.6% -> 90.8%;
implementation 82.7% -> 80.8%); nothing is implemented.
Earlier, DG-N reconciliation (#489): T01 and T04 are credited COMPLETE on implement/fid (103 -> 105 of 127; implementation 81.1% -> 82.7%);
no new atom and no decision-state change. ADR-0074 Amendment 1 chose the Bybit public trade archive as the historical source.
Earlier, DG-N registration (#483): ADR-0074 accepted; T01-T04 added as FROZEN/MISSING (123 -> 127 atoms; semantic 90.2% -> 90.6%;
implementation 83.7% -> 81.1%); nothing is implemented.
Earlier, DG-M closeout (#475): M01, M02, M04, M05 are credited COMPLETE on implement/fid (99 -> 103 of 123; implementation
80.5% -> 83.7%); no new atom and no decision-state change.
Earlier, DG-M reconciliation (#458): ADR-0073 accepted; M01, M02, M04, M05 authorized (FROZEN/MISSING), M03 superseded
(124 -> 123 atoms; semantic 86.3% -> 90.2%; implementation 79.8% -> 80.5%); nothing is implemented.
Earlier, DG-M registration (#453): M01-M05 are added as OPEN_BLOCKING/MISSING (119 -> 124 atoms; semantic 89.9% -> 86.3%;
implementation 83.2% -> 79.8%); nothing is authorized.
Reconciliation #444 imports main at 693d765 into the L2 closeout line,
preserving both inventories and credits. The union is 119 atoms / 99 COMPLETE;
the executable verifier
checks both parents and all four current inventory blocks. L2 v1 (#442/#443)
credits A14 and A14a-A14e with golden #413/#430 (11/11); A14f/A14g stay held.
Earlier, DG-K2 closeout (#409): S01-S07 move from MISSING to COMPLETE (86 -> 93 of 112), each within the bound recorded in its row;
no decision state changes (semantic stays 88.4%).
Earlier, DG-K2 reconciliation (#389): ADR-0072 is ACCEPTED; S01-S07 move from OPEN_BLOCKING to FROZEN (semantic 82.1% -> 88.4%;
OPEN_BLOCKING 11 -> 4); their implementation state stays MISSING (86 / 112 = 76.8% unchanged).
Earlier, DG-K2 registration (#382): eight candidate atoms S01-S08 (OPEN_BLOCKING except the deferable S08) are added,
all MISSING. No existing atom changes state; the percentages fall because the inventory grows from 104 to 112
atoms (semantic 88.5% -> 82.1%; implementation 82.7% -> 76.8%), not because anything regressed.
Earlier, DG-K closeout (#381): P01, P02, P03, P04a, P04b, P05a, P05b and P06 move from
MISSING to COMPLETE (78 -> 86), each within the bound recorded in
CAPABILITY_DAG.md; no decision state changes (semantic score stays 88.5%).
Earlier, DG-K design-gate reconciliation (#319): ADR-0068 and ADR-0069 are accepted,
so P01-P06 moved from OPEN_BLOCKING to FROZEN (84 -> 92 decided);
implementation stayed MISSING until the closeout above.
Earlier, DG-K registration (issue #318) added eight candidate atoms (P01, P02,
P03, P04a, P04b, P05a, P05b, P06) as OPEN_BLOCKING/MISSING. No
existing atom changes state; the percentages fall because the inventory grows
from 96 to 104 atoms, not because anything regressed.
Earlier, the Wave 8 closeout (issue #327) moves J03, J09, J10, J12, J13, J14,
J15, K12 and K13 from MISSING to COMPLETE (69 -> 78), each within the
bound recorded in CAPABILITY_DAG.md; no decision state changes.
Earlier, recomputed mechanically from CAPABILITY_DAG.md's atom tables after the
DG-J design-gate reconciliation (issue #288): all six gates reached a
disposition, moving J03, J09, J10, J12, J13, J14, J15, K12 and
K13 from OPEN_BLOCKING to FROZEN under ADR-0062 through ADR-0067, and
J11 from OPEN_BLOCKING to OPEN_DEFERABLE under ADR-0065 §5's explicit
deferral. IMPLEMENTATION_COMPLETE is unchanged at 71.9% — none of these nine
atoms is implemented. The 100% planning score means every roadmap atom is
classified. It does not mean every future semantic choice is frozen.
The closed Live Ingest Server Production Readiness v1 scope also carried one scope-level DG-B proposition that is not a new atom and therefore is not included in these counts: remediation of a live gap that exceeds the provider's bounded reconciliation window. The accepted disposition remains NO_AUTHORITATIVE_REPAIR_PATH_PROVEN: the server may record the explicit non-complete gap and continue with a new governed live segment, but may not claim the missing interval filled/lossless without future attributable repair evidence.
Decision-gate model¶
DG-A Representation / Feature integration
DG-B Historical / Live data convergence
DG-C Market-data depth (L1/L2)
DG-D Application configuration / ASS-03 RESOLVED
DG-E Validation semantics RESOLVED / IMPLEMENTED
DG-F Strategy / Execution semantics RESOLVED / IMPLEMENTED
DG-G Experiment / RL / Jobs
DG-H Operational safety
DG-I API transport / client convergence RESOLVED / IMPLEMENTED
DG-J Remote Service Topology (Wave 8) RESOLVED / IMPLEMENTED
DG-K Process Topology v1 RESOLVED / IMPLEMENTED within recorded bounds (ADR-0068, ADR-0069; closeout #381)
DG-K2 Server/deck deployment composition RESOLVED / IMPLEMENTED within recorded bounds (ADR-0072, Amendment 1; closeout #409)
DG-M Canonical feature identity for deck results RESOLVED / IMPLEMENTED within recorded bounds (ADR-0073, Amendment 1; closeout #475)
DG-N Multi-instrument Bybit linear trades v1 CLOSED within bounds (ADR-0074 ACCEPTED #377 and Amendment 1; reconciliations #483, #489, #527); T01,T04,T03,T02 COMPLETE
DG-O Live publication: continuous session and append-only slices v1 CLOSED within bounds (ADR-0075 ACCEPTED #504; reconciliations #507, #527); O01,O02,O03 COMPLETE; the one-hour real-provider run is not claimed
DG-R Rebuild of damaged legacy live days v1 AUTHORIZED (ADR-0074 Amendment 2 ACCEPTED #515; reconciliation #518); implementation not started
Important current boundaries:
- DG-A:
D05's semantics are frozen under ADR-0059 (implementation stillMISSING); it remains independent from completed H01. - DG-B: A10/B04 historical repair is complete; A11 is frozen and implemented; B06 consumer-live cursor is frozen and implemented by Wave 6.
- DG-B long-gap remediation remains blocking only for claiming an unreconciled interruption filled/lossless. ADR-0044 closes the selected production-readiness path by retaining explicit-gap-only behavior until a future authoritative repair source/path is proven.
- DG-C: L1 does not activate L2; L3 waits for real feed evidence. L2 v1 historical ingest plus consumer read is COMPLETE on
implement/l2(#442), independent ofA13, under ADR-0070 and Amendments 1 (#403) and 2 (#417). UTC-day ownership is byexchange_ts, congruent withtrade-v1; explicit head/tail gaps remain. ADR-0049 Amendment 1 (#424) allows disposal of one redundant physical copy, never the last. #425/#426 is credited as an operational tool outside A14 and adds no K09/atom credit. Eight triggers, none authorized;A14f/A14gremain held. - DG-E: complete through F08.
- DG-F: G04 session/cooldown semantics are frozen by ADR-0045; H03 execution-conflict semantics are frozen by ADR-0046; G01-G04/H01-H05 are implementation-complete on
implement/wave-4. - DG-G: the supervised branch I03/I04/I05 is implementation-complete and Golden-proven by Wave 5; J03's job-runtime decision is frozen under ADR-0062 and its durable record is implemented (Wave 8; dispatch/worker belong to DG-K); remaining unresolved branches are I06 strategic RL and I07 execution RL.
- DG-H: K02/K03/K04/K05/K06/K07/K08/K09/K10 are complete, including K10's real restart proof (PR #122), the supervised live-ingest server operating path proof, and the Wave 6 storage lifecycle proof.
Explicit deferables — second-provider resolution, exact L3/MBO semantics, DatasetSnapshot shape, future schema evolution, generic provider extension, multi-asset execution, and the Replay Consumer-API seam (J11, not planned per owner decision 2026-10-10: replay runs on the deck over K12/K13) — remain open until their evidence trigger exists. Concrete API transport was resolved by ADR-0050 in Wave 7; authentication/TLS for non-loopback deployment is frozen under ADR-0063 and implemented by J09 (Wave 8).
- DG-J: all six gates resolved (ADR-0062 through ADR-0067); see
docs/architecture/OPEN_DECISIONS.mdfor the full G1-G6 disposition text.J03,J09,J10,J12-J15,K12,K13are frozen and implemented (Wave 8 closeout #327, within recorded bounds);J11remainsOPEN_DEFERABLE. - DG-K: Process Topology v1 is resolved (#318 registration, #319 design gate: ADR-0068 process topology v1, ADR-0069 J02 carriage of consumer seams and deck handoff v1).
P01-P06areFROZEN/COMPLETEwithin recorded bounds (operator closeout #381), merged intoimplement/dgk, not composed in any executable and without a Golden proof; owner inputs fix identity-preserving canonical profiles, hosts intools/, one J03 worker, a free server with data-local J03 handlers only, heavy compute on the deck on its own initiative, andJ11not planned. SeeOPEN_DECISIONS.md(DG-K). - DG-K2: registered by #382 after the DG-K closeout; its design gate #383 produced ADR-0072, accepted on 2026-10-16, and #389 authorizes
S01-S07(K12 authority, K13 composition, API host with TLS/mTLS, credential and trust operations, deck workload module, deployment units, two-part topology proof) on the integration branchimplement/dgk2(name proposed, to be confirmed by the maintainer). Closed by operator closeout #409: all seven are implemented onimplement/dgk2within the bounds recorded inCAPABILITY_DAG.md; the split is deployable only within the bounds of ADR-0072 Amendment 1; Omega adoption is out of scope. Default proposed placement: before Wave 9, because the split is the precondition for any remote consumer; the order remains the maintainer's. - DG-M: registered by #453 after the DG-K2 closeout; the maintainer named feature traceability as a requirement on 2026-10-18. Its design gate #454 produced ADR-0073, accepted on 2026-10-18 (#457), #458 authorized
M01(bar-flow representation),M02(failed-aggression feature),M04(server registry anddefinitionscarriage) andM05(deck declaration rule and proof) on the integration branchimplement/fid;M03is superseded. Closed by operator closeout #475: all four are implemented and merged (PR #466, #468, #473, #474) within the bounds recorded inCAPABILITY_DAG.md; no real-host run is claimed, and promotion intomainis the operator's separate PR. Follow-on #477 / PR #478: the deck protocol client carriesdefinitions(not an atom). Independent of the FEI line. Default proposed placement: after DG-K2 and before Wave 9; the order remains the maintainer's. - DG-N: the maintainer holds Parquet trades for other Bybit linear USDT instruments and wants them read, delivered and ingested live (2026-10-19). The design gate #377 produced ADR-0074 (accepted, #482) and #483 authorized
T01(read and delivery),T04(historical import),T03(explicit fees) andT02(live ingest per instrument) onimplement/fid, in the current wave; the wave is promoted tomainat its end. Amendment 1 (#487, accepted 2026-10-20) chose the Bybit public trade archive as the only historical source, because the existing files declare a source profile no commit produced and S13 refuses;T01andT04areCOMPLETE(#489);T03andT02areCOMPLETEwithin their bounds (closeout #527). A second venue, spot, inverse and L2 for other instruments are out of scope. - DG-O: measured on the maintainer's server on 2026-10-21 (a read-only catalog query and the Bybit public archive): the live publication supersedes each cycle with the next, so a day reads as under 0.4% of its published trades, and the loop closes its WebSocket between cycles, so about 27% of a day's trades are never acquired. The design gate #504 produced ADR-0075 (accepted) and #507 authorizes
O01(append-only slices),O02(continuous session) andO03(configuration, runbook, real-provider proof) onimplement/fid, before the rest of the wave;T02closed withO03; all three areCOMPLETEwithin their bounds (closeout #527) and the one-hour real-provider run is not performed. Consolidation of slices is outside the first version; the already damaged live days are not repaired by DG-O (see DG-R). - DG-R: the days published by the legacy live path (31 candidate days, 2026-09-23 to 2026-10-23) read as their last batch and miss trades never acquired. ADR-0074 Amendment 2 (accepted 2026-10-24) rebuilds explicitly selected, closed days: retire the damaged revision with an ordinary reassessment, import the archive day under a free key, verify. #518 authorizes
R01(reassessment writer) andR02(plan, retire, import, verify) onimplement/fid; they are post-version and do not blockO02orO03. The atomic replacement of the first draft is not adopted; the run on the server is the operator's, from a reviewed plan.
Operational dependency direction¶
K06 source protection -> K08 backup/restore proof -> A11 live acquisition
A11 live acquisition -> K10 checkpoint/recovery implementation
K06 source protection -> K07 relocation
K08 backup/restore proof -> K09 deletion authority
K08/A11/K02/K10 implementation and real-server evidence are now credited, including K10's real restart proof (PR #122). Backup/restore does not require tier relocation.
Vertical milestones¶
V1 Source -> canonical -> catalog -> DataGateway COMPLETE
V2 DataGateway -> Application service COMPLETE
V3 DataGateway -> historical Candle COMPLETE
V4 Representation -> Feature -> canonical H01 COMPLETE
V5 Feature -> Research COMPLETE
V6 Research -> Validation COMPLETE
V7 Strategy -> deterministic Replay COMPLETE
V8 Historical -> Live COMPLETE for the bounded data-plane/storage lifecycle path; broader live product gates remain open
V9 Application -> API -> Client COMPLETE for bounded WebSocket transport and thin clients
V10 Paper/Shadow -> Live BLOCKED by runtime/operational gates
V8 is complete for the bounded data-plane/storage lifecycle path selected through Wave 6: the first live-ingest producer path and the live consumer/candle/storage lifecycle path now converge through K08/A11/K02/K10, B06/D04 and K07/K09 evidence, ADR-0043/ADR-0044/ADR-0047/ADR-0048/ADR-0049, and the Wave 6 Golden proof. This does not complete J07/J08, broker/live execution, clients, second venue or market-depth branches; any long gap beyond bounded provider reconciliation must remain explicit until authoritative repair evidence exists.
Current execution frontier¶
Wave 7 API & Platform Transport is closed on implement/wave-7: ADR-0050 and
PR #225/#226/#227/#228 complete J02/J04/J05/J06. Issue #222 was closed by
explicit operator exception during expedited closeout, so this roadmap state
does not claim an additional dedicated Golden E2E artifact.
The post-Wave-7 implement/omega stabilization line (tracking #232) is
reconciled: eight design gates (ADR-0054-0061) and nine implementation/
hardening slices (PR #254, #256-#278), plus amendments to ADR-0037 and
ADR-0050. It resolves D05's semantics (implementation stays MISSING) and
closes the audited findings; it is not a wave and selects no new runtime/
product frontier by itself.
Wave 8 — Full Platform API & Remote Service Topology v1 is closed by operator
closeout (issue #327): J03, J09, J10, J12, J13, J14, J15, K12
and K13 are implemented and promoted to main (PR #322, #326). J10/J12/
J13 are credited as in-process application seams only: no J02 wire message or
remote scope carries them yet. No Wave 8 Golden E2E proof artifact is claimed.
Owner-accepted extras #313/#314 are on main through implement/wave-8.1.
DG-K P01-P06 (#381) and DG-K2 S01-S07 (#409) are COMPLETE within their recorded bounds and promoted into main (#434/#437). DG-K2 supplies the later executable composition, credentials/runbook and deck workload module; its real proof remains one happy path on the smallest partition, within ADR-0072 Amendment 1 bounds and the recorded findings. S08 is not authorized.
L2 v1 is CLOSED on implement/l2 (#442/#443), A14 and A14a-A14e COMPLETE, golden #413/#430 accepted 11/11. Reconciliation #444 combines both histories before a separate final L2 wave promotion. A14f/A14g stay held. The recorded limits and eight unauthorized triggers remain: no live L2, ob500, population guarantee or gap-free UTC-midnight read.
No new runtime/product frontier is authorized by this closeout, the Wave 8
closeout or the stabilization line. H06, I06-I07, J07/J08, RL, live broker execution, second
venue, L1/L3 and L2 beyond the bounded historical v1 credit remain outside all of them. J07 remains
MISSING.
Roadmap state is still not concurrent authorization.
Execution waves¶
Waves are dependency/value groupings, not a new linear phase numbering, and are distinct from temporary implement/wave-* integration branches.
Wave 0 Architecture foundation COMPLETE
Wave 1 First canonical computation/application slices COMPLETE
Wave 2 Representation / Feature COMPLETE
Wave 3 Research / Validation COMPLETE
Wave 4 Strategy / Replay COMPLETE
Wave 5 Experiment / Supervised ML COMPLETE (I01-I05; Golden supervised E2E PASS)
Wave 6 Live Data Plane COMPLETE (B06,D04,K07,K09 plus credited K02,K03,K04,K05,K06,K08,K10,A10,A11,A16; Golden PASS)
Wave 7 Runtime / Clients COMPLETE (J02 -> J04/J05/J06; exceptional Golden waiver)
Wave 8 Full Platform API & Remote Service Topology v1 COMPLETE (J03,J09,J10,J12,J13,J14,J15,K12,K13 within
recorded bounds; J10/J12/J13 in-process only; J11
deferred; no Golden proof claimed; operator closeout
#327). J02 carriage of J10/J12/J13 moves to DG-K P05.
DG-K Process Topology v1 (design-gate line, not a wave) COMPLETE within recorded bounds (operator closeout #381); before Wave 9
DG-K2 Server/deck deployment composition COMPLETE within recorded bounds (operator closeout #409); deployable only within ADR-0072 Amendment 1 bounds
DG-C/L2 L2 historical v1 (#442) COMPLETE on implement/l2: A14,A14a-A14e; golden #413/#430 11/11;
A14f/A14g held MISSING; separate L2 promotion to main
Wave 9 Reinforcement Learning I06 strategic RL, I07 execution RL; harvest-audit
evidence sources: ml_core (already the Wave 5 legacy
baseline), JJJerome/mbt_gym, sadighian/crypto-rl
Wave 10 Paper / Live product J07 -> J08 after required data/execution/ops gates;
implementation intends to draw architectural
inspiration from nautilus_trader as read-only
reference evidence only, never a runtime dependency
The earlier closed Live Ingest scope was a bounded path through part of Wave 6, not authorization to implement Wave 6 generally. Wave 6 itself is now complete only for the bounded live data-plane/storage lifecycle path described above.
Post-Wave-7 sequencing: "platform v1.0"¶
Waves 0-7 are the accepted foundation: canonical data plane, representation, research/validation, strategy/replay, supervised ML, live consumer data plane and storage lifecycle, and now API transport with thin clients. The operator has set an explicit priority order for what comes next, driven by usability of what is already built rather than by new capability breadth:
- Wave 8 first — the already-complete Strategy/Replay/Validation/
Training engines (Waves 2-5) are only reachable today by writing a Python
script against the domain packages directly; there is no Consumer API or
client path for any of them. Wave 8 closes that gap the same way Wave 7
closed it for market data: a
C02/C03-equivalent seam per domain. Wave 8 delivered those seams in-process (J10/J12/J13); carrying them over J02 and making them reachable from J04/J05/J06 was reassigned by the owner on 2026-10-08 to DG-K Process Topology v1, atom P05 (#318). DG-K before Wave 9 (default recorded by #318; the order remains the maintainer's): DG-K Process Topology v1 finishes what Wave 8 left in-process (J02 carriage of J10/J12/J13, J03 dispatch and one worker, a PostgreSQL runtime store, the K12/K13 deck handoff over J02). Heavy compute stays on the deck on its own initiative and the server stays data-local; noJ11gate follows DG-K (owner decision 2026-10-10). - Wave 9 second —
I06/I07remainOPEN_BLOCKINGunderDG-Gand are independent of Waves 8 and 10 in the capability DAG; a legacy-harvest audit againstml_core,JJJerome/mbt_gymandsadighian/crypto-rlprecedes any implementation issue, following the same disposition disciplineADOPTION_LEDGER.mdalready established for Wave 5's supervised-ML harvest. - Wave 10 last —
J07's own DAG dependencies (A11,H05,I05,J02,K03) are already satisfied, so it could technically start earlier; it is deliberately sequenced last because it delivers a new capability rather than unlocking use of ones already built.nautilus_traderis noted here as the intended architectural reference for this wave's eventual design gate, under the same read-only-evidence discipline as every other third-party/legacy reference this repository has used.
This is a priority order, not authorization: each wave still requires its own
SCOPE.md and, where a decision remains open, its own design gate before any
implementation issue may start.
Producer / storage interpretation¶
Capacity observation, application work, historical representation work, backup design and other independent capabilities may advance in parallel when their declared dependencies are met, subject to the active scope's mutation policy.
Backfill/repair does not require the entire storage-tiering program. K08 backup/restore, A11 live acquisition, B06 live consumer access, D04 live candles, K07 tier relocation and K09 deletion authority are complete for the selected bounded Wave 6 path. API transport does not precede the completed in-process ASS-02 service.
Planning rule¶
Before opening implementation:
- verify the current authoritative branch/ref;
- read
SCOPE.mdand the selected atom's accepted ADR; - verify its transitive
Requirespath; - credit existing evidence;
- define only the exact missing implementation/proof proposition;
- keep unrelated frontier atoms inactive;
- reconcile governance only after integrated evidence changes state.
Roadmap state never authorizes production/runtime mutation by itself.