Progress¶

Session definitions: docs/BUILD_PLAN.md section 10. Milestones: section 16.4.

Summary (2026-10-08)¶

Done: every MVP session (S1-S22) and milestone (M0-M4); Step 1.5 (docs site, admin console, workspace, browser strategies); Step 2 hardening (duplicate client order IDs, snapshots, journal archiving); the S14 realism report; the order-latency tail investigation; allocation and GC work (ADR 0023); leaderboard names for every kind of account. 23 ADRs, full make test and make e2e green, docs site link-checked.

Remaining to call Prayog finished, in suggested order (each "done when" is the acceptance check):

Browser check of strategy names: start a browser strategy; done when its user/algo-... row shows on the leaderboard.
Volatility clustering: time-varying sigma or self-exciting jumps in the simulated traders; done when a longer realism run shows positive autocorrelation of absolute returns in docs/realism.md.
Demo video: 2 minutes, following docs/demo.md; done when linked from the README.
UI tests: component tests per panel, then Playwright journeys (Playwright needs approval); done when CI runs them.
Releases: tag R0 (MVP) now; R1 after items 2-4; done when GitHub releases exist with notes.
Hosting: decide where; done when a stranger can open the market without installing anything.
Bot arena v1 (offline submissions, shared with Galois) as release R3.
License: the public repo has no LICENSE file, which legally means "all rights reserved". Pick one (MIT is common for showcase projects); done when LICENSE is in the repo and GitHub shows it.
GitHub repo page: set the homepage to https://rj8228.github.io/prayog/ and add topics (for example matching-engine, lmax-disruptor, kafka, event-sourcing, java, spring-boot).
MVP definition of done, demo part: BUILD_PLAN section 11 and S22 ask for a recorded demo. S22 is ticked for the script; the MVP is formally complete only when item 3 is done.
Keycloak integration test: the testing strategy (BUILD_PLAN section 12) lists Testcontainers for Keycloak; tokens are covered by the live stack (make smoke, make e2e) but no Testcontainers test starts Keycloak. Decide: add one, or record why the e2e coverage is enough (ADR).
Public scaling write-up (optional): the 1-million-user design is in the private interview notes; a short public version in docs/ would show it to readers of the repo.
Stale planning docs (optional): the Claude Docs "Prayog Step 2 Handoff" (2026-10-04) and the build-plan artifact predate this state; mark them superseded so they don't mislead later.

Parked: JEV-model bot (needs an API key).

M0 Setup¶

Tool installs (Docker Desktop, maven, uv, gh, pnpm, Python 3.12)
Git init and GitHub repo (public: github.com/rj8228/prayog)
S1 Repo skeleton and tooling (5 h)

M1 First module¶

S3 Contracts and domain types (5 h) — ADR 0002
S4 Order book and limit matching (6 h) — ADR 0003

M2 Engine complete (done 2026-10-04)¶

S5 Market, cancel, modify (6 h) — ADR 0004
S6 Price bands, sessions, self-trade prevention (5 h) — ADR 0005
S7 Sequencer and single-writer pipeline (6 h) — ADR 0006
S8 Journal and deterministic replay (6 h) — ADR 0007

M3 Runnable app¶

Step 1.5 (2026-10-05)¶

Docs site: MkDocs Material on GitHub Pages, link-checked (commits f5deacd, 02d89e5)
Admin user and console: self-test, simulation control, accounts, event replay, order journey — ADR 0011
Trading workspace: drag, resize, swap, presets, new panels, one-click, shortcuts, explain mode — ADR 0012
Strategies in the browser: six strategies with risk limits, own accounts, live P&L and logs — ADR 0012

Step 2 additions (2026-10-07)¶

Duplicate client order IDs rejected per account per trading day — ADR 0014
Engine snapshots (recovery replays from the latest snapshot) — ADR 0016
Journal archiving (gzip finished segments, readers see one history) — ADR 0022, 2026-10-08
Boxing-free order index; GC settings measured — ADR 0023, 2026-10-08

M4 Hardening layers¶

S9 Benchmarks (4 h) — docs/benchmarks.md
S14 Realism checks (4 h) — docs/realism.md 2026-10-08; spread and return autocorrelation realistic, fat tails weak, volatility clustering not reproduced (docs/learnings/S14-realism.md)
S20 Exchange rules as Cucumber scenarios (5 h) — ADR 0018
S21 Observability (4 h) — ADR 0019; Grafana showed a live session and Kafka lag spikes 2026-10-07 (docs/media/05-grafana.png)
S22 Docs and demo (5 h) — README, failure-modes.md, demo.md (recording: follow docs/demo.md)
Multi-arch images — ADR 0021 (make images, CI job)
Phase 2 items pulled into scope (agent rate-limit tier) — ADR 0020

Backlog (from the handoff, focus plan and ideas docs, 2026-10-07)¶

MVP close-out - [x] S14 report: run the realism report on the live stack, commit docs/realism.md and docs/learnings/S14 - [ ] Volatility clustering in the simulator (time-varying sigma or self-exciting jumps), then a longer realism run - [x] Learnings for S18 and S19 (ops page, blotter, server P&L), linked from docs/learnings/README.md - [ ] Record the demo video (docs/demo.md script; raw capture in docs/media/raw) and link it from the README - [x] Full make test and make e2e on the repaired stack (2026-10-08) - [x] Update the Step 2 handoff doc — replaced by a Step 3 handoff in the private interview notes, 2026-10-08 - [x] Investigate new-order p99 in Grafana: Serial GC pauses slowed about 6x (likely host memory pressure); steady-state p99 14 ms; ZGC worse in a 768 MiB container (docs/benchmarks.md, 2026-10-08)

Hardening - [ ] UI tests: component tests per panel and Playwright journeys (Playwright needs approval) — deferred 2026-10-08 - [x] Journal segment archiving — ADR 0022 - [x] Leaderboard: simulated traders show names (confirmed 2026-10-08); browser strategy accounts did not, fixed in 3e107ba

Showcase (focus plan definition of done) - [ ] Host it so a stranger can use it, with a 2-minute demo video (hosting not decided) - [x] JMH baseline, then GC tuning and lock-free changes, each with before and after numbers — ADR 0023, docs/benchmarks.md 2026-10-08 (allocation -12 to -21%; GC settings change pauses, not the laptop's tail; matching already lock-free) - [x] Written design for 1 million users (shard by market, hot standby replaying the journal, market data fan-out with conflation and backpressure) — in the private interview notes, 2026-10-08 - [ ] JEV-model bot (TypeSafe AI's Jev decision API) against a random baseline — skipped 2026-10-08, needs an API key - [ ] Releases R0 to R3 tagged, ending with bot arena v1 (offline submissions, shared with Galois) - [x] Interview companion entries for Prayog: pitch, numbers, failures, decisions — in the private interview notes, 2026-10-08