Lanes

Trivial, reversible, single-repo. The lightest thing Ark does.

What happens

• No intent spec. The ticket and the MR are the whole artefact trail.
• A Builder makes the change in an isolated worktree, runs the relevant tests, and opens the MR.
• An independent Reviewer checks it before any person sees it.
• A person reviews and merges. Done.

Evidence and sharing: No probe evidence, no visual evidence. The share pack is an MR link, queued for owner approval.

From the board

• Editor Document/Customer card showed literal "null" and "N/A" (KP-742)
• Migrated bookings showing a placeholder instead of a real Location on Builder lines

If a fix starts sprawling (second repo, metadata changes, unclear root cause), it re-triages up the ladder. That is the escalation working, not the lane failing.

A bounded enhancement. Real spec, collapsed ceremony.

What happens

• A stripped intent spec with one success metric. A person still aligns it before any run.
• The Lead decomposes into a small task graph; Builders run in parallel worktrees.
• UI-touching tasks capture before/after screenshots automatically when the intent declares a bootable app.
• Review, person merge, and a verdict against the single metric.

Evidence and sharing: No probe gate. Before/after visuals on UI tasks. Share pack: summary, MR links, and images, queued for approval.

From the board

• Edge Pay custom-amount stale-write race (guests charged 1/100th of the intended amount)
• Cruise vertical rig: Traveltek connector plus a Cruise Test Console
• Lane policy itself: every lane defines its evidence and sharing defaults

Cross-repo, novel, risky, or anything touching payments, auth, secrets, or tenant isolation.

What happens

• The probe gate blocks decomposition until evidence exists: the defect photographed on a seeded org, or the capability proven, before any code is written.
• The environment doctor refuses the run if the machine lacks what the lane needs (scratch-org capability, Jira credentials, a local loop).
• TDD discipline: a red test that reproduces the problem, then the fix that turns it green, then the regression suites.
• The fix is verified on an org through the surface the customer reported, not just in a test run.
• os verify-pre-release runs the declared checks and writes the SIT report; the MR description carries the full QA handover.
• Person architecture sign-off when the work touches auth, secrets, payments, or tenant isolation.
• After release, the Verifier measures live signal before the intent can be called proven.

Evidence and sharing: Probe evidence required. UI capture plus UAT composites. Share pack: summary, MRs, and composites, queued for approval.

From the board

• Belmond 23.2.8 patch: five defects fixed with repro evidence, red/green tests, and visuals before the 10 Aug build
• Edge Connect facade and native supplier connectors
• The Ark OS itself: board API, live cost ledger, scheduled verification (built in its own full-lane intent)

An investigation. The outcome is a finding or a decision, not shipped code.

What happens

• The run IS the probe: read-only access to orgs, databases, logs, and configs. No build commitment.
• Agents gather evidence, reproduce behaviour, and cost options.
• The output is a memo in the evidence folder: what was found, what it would cost, what is recommended.
• The finding then mints or upgrades a build intent, or ends in a deliberate decision to do nothing.

Evidence and sharing: The probe is the evidence. Share pack is a finding summary, queued for approval.

From the board

• Carr Golf retention viability: a keep-or-exit position costed with evidence before another upgrade attempt
• Root-cause probes that decide whether a defect is Salesforce-side or KTAPI-side before anyone writes a fix

A standing loop over a source: a customer channel, a bug queue, an error budget.

What happens

• The intent stays in-flight for weeks. Each cycle sweeps the source, answers what it can, and appends one run-log entry.
• Carry-over discipline: open items are tracked cycle to cycle until resolved, with owners named.
• Anything too big for the loop re-triages OUT into its own build intent; the sweep keeps the customer conversation moving meanwhile.
• Quiet cycles are logged as quiet. The run log is the audit trail that nothing was dropped.
• Closes at a milestone: a go-live, a plan playback, a channel handover.

Evidence and sharing: Customer-facing packs are publish-gated: assembled by the pipeline, sent only after person approval.

From the board

• DuVine, Aurora, Bunnik, and GRJ delivery programmes: every ask in the channel answered, release scope shipped
• The periphery sweep: governance and build channels covered so nothing rots in a thread
• Aurora production cutover: Package Search plus Booking Wizard live, tracked to the day

A data-migration programme over a legacy dataset.

What happens

• Probe first: the legacy schema is explored and documented before any mapping is designed.
• Then design, then detailed field-level mapping, then verification cycles of increasing volume.
• Every load into a customer org is person-sign-off gated. No exceptions.
• The outcome is reconciliation: source and target agree, with the discrepancies explained.
• Closes at cutover.

Evidence and sharing: Probe evidence required before decomposition. Verification reports per cycle.

From the board

• Bunnik TravelStudio migration, verified layer by layer into Kaptio Salesforce

A go-to-market programme: a campaign, a segment, a deal close.

What happens

• Research first: the market motion is grounded in evidence before decomposition.
• Then plan, then execution cycles of increasing reach.
• Every external touch (an email to a customer, a proposal, a post) is person-sign-off gated. Agents draft; people send.
• Product claims must trace to proven Outcome Reports on the board. No invented capability claims.
• The outcome is a funnel outcome: signed paper, a booked meeting, a launched campaign. Closes at the milestone.

Evidence and sharing: Research evidence required before decomposition. Commercial numbers are owner-approved before anything leaves the building.

From the board

• Luxury Escapes Foundation SOW signed by 31 August
• Belmond signs Kaptio Engage Growth, on an enhancement-process story that stands even if they decline
• Rhino Africa headless re-platform play