previews · sprint reports · Lawrence Experience · Sprint 6

Sprint 6 — Demo Tail, Release Polish & the Drawing Board

No goals page was written this sprint: the cycle opened with Sprint 5's extension to the "Hacking Law Firm" demo on Fri 29 May; the rest was a release-readiness push — polish, bug bashing — and the deepest planning corpus of any sprint yet.

✅ Final Cycle 6 · May 28 → Jun 10 2026 Generated 2026-06-10 Notion source · LEX Sprint 6 Report Linear · team/LEX Previous · Sprint 5 report
TL;DR

Sprint 6 ran in three acts: the Sprint 5 demo tail landed in the first 48 hours and the demo ran Friday 29 May; the bulk of the cycle was a release-readiness push — email-draft integrity, a canvas & docking rework, citation-highlight recovery, forms quality, a rebuilt matter overview; and throughout, the squad produced 15+ planning artifacts covering Forms 2.0, legal-data partnerships, matter memory, testing & load capacity, and the Lawrence 2.0 infrastructure roadmap.

Engineering delivery: 37 issues / 69 pts landed by Linear status — ~41 issues / ~77 pts once validated against git — inside the sustainable 70–100 pt band, with two days of demo crunch and a deliberate tilt toward planning. Demo/hackathon/spike content tickets are excluded from these totals (see Sprint health).

The polish has a destination: releasing Lawrence 2 as the default chat experience for every user on v3. It currently sits behind a feature flag — not all lawyers have access yet, and the legacy chat experience is still live alongside it. This sprint's email, canvas, citations and forms work is what stands between the flag and the flip.

🎬 The Hacking Law Firm demo (Fri 29 May) went well. For how it landed and what happens next with the firm, get in touch with Sedona.

How the sprint ran — three acts

Act 1 · Demo tailMay 28–29
Sprint 5's final 48 hours.

Agent form-editing end-to-end, forms-in-canvas, fill-form cards and citation fixes landed on the wire; the demo ran Friday 29 May.

Act 2 · Release polishThe bulk
Bug bash toward default-on.

Email-draft integrity fixed at the root, canvas & docking Phases 1–3, citation-highlight recovery, forms quality, matter overview rebuilt.

Act 3 · Drawing boardThroughout
Thinking through next stages.

15+ artifacts: Forms 2.0, OpenLaws & legal data, matter memory, testing strategy, load & capacity, compliant evals, the Lawrence 2.0 roadmap.

Sprint health

14 content tickets (66 pts) — the LCO spike family, hackathon QA and demo personas — are excluded from the totals below: real, finished work, but research/content output that would inflate the engineering velocity read. It appears on the drawing board instead. That leaves 83 issues / 121 pointed pts in scope.

121pts
In scope
83 issues · content tickets excluded
69pts
Landed
37 issues deployed or merged · 57% of pointed
~77pts
Git-validated landed
canvas stack on trunk, tickets lagging · ~41 issues
15+
Planning artifacts
Forms 2.0 → matter memory → proactivity roadmap

Points distribution by status

Deployed (53%) Merged (4%) In review (13%) In progress (25%) Todo · 29 issues, mostly unpointed polish backlog

Burndown note: Linear's completed-scope line jumped 53 → 132 over the final two days — mostly a retroactive ticket-housekeeping wave, not a deploy crunch. Hygiene: the canvas docking stack (LEX-478/424/477/480, 8 pts) hit trunk 9 Jun via the Graphite merge queue but the tickets still read In Review — hence the 69 vs ~77 gap. 17 deployed issues carry no estimate, so landed points understate landed work; 49 of 97 issues were unpointed overall.

May 28
Demo · May 29
Jun 2
Jun 6
Canvas stack · Jun 9
Closed · Jun 10

Two-week cycle. The Hacking Law Firm demo ran Fri May 29, two days in; the cycle then shifted to release polish and planning, with the canvas docking stack landing Jun 9, the evening before close.

Act 1 — Landing the demo (May 28–29)

The form-fill backbone from Sprint 5 got its last pieces in the final 48 hours before the demo:

Act 2 — Release polish & bug bash

The through-line: getting Lawrence 2 ready to be switched on as the default chat for all v3 users.

Highlight

⭐ Matter overview rework

Landed
The surface Lawrence lives in · 17-commit batch

Jaime rebuilt the matter surface around Lawrence: the matter overview card, a new matter pill, a Lawrence summary card on the overview, de-sprung panel/sidebar animations, consistent shell padding and gutters, and flag-gated pinning infrastructure. Untracked under LEX, but it reshaped the context every other polish item sits in.

Email

Email draft integrity

Landed
LEX-468 / LEX-458 / LEX-381

The empty-email-draft family of bugs fixed at the root rather than patched.

What landed
  • Content framework is now the sole autosave writer with a correct dirty baseline (no more autosave-on-open)
  • Recipients persist across close/reopen; Send disabled on an empty body; headings render as bold body text in emails
  • Backed by an engineering writeup of the inconsistent save handling across content-framework consumers (1 Jun)
Canvas

Canvas & docking — Phases 1–3

Landed · prod pending
LEX-478/424/477/480/423/464/430 · 8-PR stack, Jun 9

Files, PDFs and content drafts open in the canvas, with Lawrence docking instead of floating over fullscreen.

What landed
  • z-index token system putting Lawrence below modals; docked-canvas safe areas
  • Cited documents open in canvas tabs; canvas no longer misbehaves when tabbing between floating threads; per-field "review all fields" prompt removed
  • Merged to develop the evening before close — prod deploy pending
Citations

Citations quality

Landed
Highlight recovery + UX cleanup

The demo's citation backbone hardened for everyday use.

What landed
  • Missed PDF citation highlights recovered with a gapped text-anchor fallback; tabular citation highlights
  • ?formId= deep links through the Lawrence canvas
  • "Couldn't locate this passage" banner removed (LEX-463); previews capped at 4 lines (LEX-360); incorrect-citation flagging (AICAP-52)
Forms

Forms quality

Landed
Fix the demo's rough edges · prep the next one

Accuracy and admin tooling caught up with the demo push.

What landed
  • Checkbox-group label mis-binding fixed (LEX-466); multi-upload no longer shows stale PDFs
  • Forms admin deactivate endpoint + UI
  • Georgia Divorce forms sourced and unflattened (LEX-488/489) — prep for the next demo storyline
Threads & bar

Threads, search & bar polish

Landed
Exact-string thread search live

Search across threads shipped; the Lawrence bar got its rough edges filed off.

What landed
  • Exact-string search across threads (agents #605 + local typesense setup)
  • Hidden-chats indicator → "All chats" button (LEX-470); tooltip-after-minimise and font-change fixes (LEX-425/453); matter-pill padding
  • Matter-tab data caching with parallelised account fetches

Adjacent ship worth a line: Personalise Lawrence (AI Capabilities) — flag-gated personalisation settings UI landed on v3.

Act 3 — The drawing board

What "thinking through next stages" actually produced — 15+ artifacts in 14 days. Each links to the Notion source.

ThemeArtifacts
Forms next actForms 2.0 Proposal · Phase 1 — Rich Form Ingestion design · Phase 3 — Agentic Filling (Interface B) (all 29 May) · Options for Completing: Forms and Skills (3 Jun — capacity & review strategy)
Legal dataOpenLaws Integration brainstorm + Vendor Brief (29 May) · Free Law Project Data Services Agreement redline (LEX-416, in review) · US/UK legal data tools in build (LEX-511/512) · CourtListener partnership (LEX-386)
Memory & personalisationMatter Memory draft PRD + M2 Tech Plan + memory-candidate detection research plan (8–9 Jun)
Quality & scale readinessLawrence Testing Strategy (1 Jun) · Load Test & Capacity Plan for Broad Rollout (6 Jun) · Compliant Evals PRD (29 May)
Product & platform directionLawrence 2.0 — Next Steps and Strategies (9 Jun): proactivity (triggers + background execution), durable streams, agentic form filling, Global/Anywhere Lawrence, agent-API generalisation, framework/chat-agent untangling · Proactive End-to-End Work Delivery proposal · Lawrence Design Principles (7 Jun)
Closed planning ticketsMatter-level threads FE + BE tech plans (LEX-338/339) · context-window management research spike (LEX-341) · confidence calibration v0 (LEX-450) · LCO generation spike family (plan → notebook → report shipped; PRD in progress) · Structured Vision Doc in review (LEX-259)

Side quests

Untracked-by-design personal work, surfaced for visibility — no retroactive tickets.

Matter-tab performance (data caching, parallelised fetches, hover prefetch in review) · ingestion hardening (8 PRs: CVE bumps, retry-loop fixes) · v3 mailroom auto-deploy · the org-wide agentic coding knowledge base project · deployment-staleness alerts and plugin/infra work.

Bugs squashed

AreaWhatRef
EmailAutosave dirty baseline fixed at the root — no more phantom drafts or empty-body sends; recipients persistLEX-468
EmailDraft body survives close/reopen from a previous threadLEX-458
CanvasCanvas misbehaving when tabbing between floating threads; form modal close-button overlapLEX-423 · #11058
CitationsMissed PDF highlights recovered (gapped text anchors); tabular citation highlights#10799 · #10923
FormsCheckbox-group label mis-binding when AcroForm order differs from visual orderLEX-466
FormsMulti-upload no longer shows stale PDFs#10895
BarTooltip reappearing after minimise; font-change behaviourLEX-425 · LEX-453
ChatBackslash formatting in chat; web-search results stripped of control characters#10579 · agents #628

In flight — carried into Sprint 7

Awaiting deploy / review
On or near trunk.
Canvas docking stack → prod (merged to develop Jun 9)LEX-478/424Wills
Free Law Project DSA redlineLEX-416Harry
Structured Vision DocLEX-259Harry
Open PR stacks
Parallel builds in flight.
Forms-processing service — ~39 PRs across platform-v3 + ai-platformPeter
ask_user_question human-in-the-loop primitive — 7 PRsPeter
OpenLaws legal-sources stack + agent-mode tool partsHarry
G-28 form-fill accuracy stackLEX-471/472/482Gareth
Form-fill progress indicator + ActivityBlock/streaming reworkLEX-483Pratik
Todo29 · mostly unpointed
Polish backlog + next demo.
Polish chip backlog — z-index, modals, scroll, tooltipsLEX-421→486
Legacy chat deprecation after Canvas rolloutLEX-500
Georgia Divorce demo storylinenext demo prep

Lessons learnt

What went well

  • The squad pivoted from demo crunch straight into release-readiness without a goals page and stayed coherent — the implicit goal (Lawrence 2 default-ready) was understood by everyone.
  • Planning debt got paid down substantially — six workstreams now have written proposals or plans before their build starts.
  • Engineering delivery stayed inside the sustainable band despite the demo, the planning tilt and a shorter effective window.

What to improve

  • Ticket hygiene again: the canvas stack landed but four tickets read In Review; 49 of 97 issues were unpointed. Velocity is only legible after git archaeology.
  • Content vs engineering pointing: spike/demo content tickets distorted the raw number (this report excludes them) — worth a label or a no-points convention for content work.
  • Flag-flip readiness needs a checklist: the load-test & capacity plan exists; executing it (plus legacy-chat deprecation, LEX-500) should gate the default-on moment.

Sprint 7 preview

Cycle 7 runs Jun 11–24. The likely shape, from the planning corpus: Lawrence 2 default rollout readiness (load test, capacity, legacy-chat deprecation path), forms-processing service + Forms 2.0 phases, OpenLaws / legal data tools, matter memory M1, the proactivity & infrastructure roadmap from Next Steps and Strategies, and the Georgia Divorce demo storyline.