openlanes 3.0.0

data/bin/factory-e2e

@@ -0,0 +1,157 @@
+#!/usr/bin/env bash
+# vim: ft=bash
+#
+# The factory integration test: a throwaway git repository, `openlanes start`
+# serving it, and Cypress building a tic-tac-toe page through the control
+# room with real agents. It spends real tokens, so it refuses to run unless
+# OPENLANES_FACTORY_E2E=1 says so, and nothing in `just test`, `just web-e2e`
+# or CircleCI calls it.
+#
+#   OPENLANES_FACTORY_E2E=1 just factory-e2e
+#
+# Settings, each with a default:
+#
+#   FACTORY_PORT       control room port                    (4777)
+#   FACTORY_REDIS_URL  Redis for the board and its commands (redis://localhost:6379/11)
+#   FACTORY_MODEL      model every agent runs on            (haiku)
+#   FACTORY_TIMEOUT    seconds before one agent is abandoned (600)
+#   FACTORY_MINUTES    minutes Cypress waits for the build  (15)
+#
+# What it leaves behind is log/factory-e2e/: the progress log, the last
+# board, every plan's state.json and spec.md, the built index.html, and the
+# tail of each agent's trace. The temp repository and the `openlanes start` it
+# launched are always torn down, pass or fail; nothing it did not start is
+# touched.
+
+set -euo pipefail
+
+repo="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+
+if [[ "${OPENLANES_FACTORY_E2E:-}" != "1" ]]; then
+  echo "factory-e2e runs real agents and spends real tokens." >&2
+  echo "Set OPENLANES_FACTORY_E2E=1 to run it: OPENLANES_FACTORY_E2E=1 just factory-e2e" >&2
+  exit 64
+fi
+
+port="${FACTORY_PORT:-4777}"
+redis_url="${FACTORY_REDIS_URL:-redis://localhost:6379/11}"
+model="${FACTORY_MODEL:-haiku}"
+timeout="${FACTORY_TIMEOUT:-600}"
+minutes="${FACTORY_MINUTES:-15}"
+artifacts="${repo}/log/factory-e2e"
+
+command -v claude >/dev/null || { echo "claude is not on PATH; the factory has no agents to run." >&2; exit 69; }
+if command -v redis-cli >/dev/null && ! redis-cli -u "${redis_url}" ping >/dev/null 2>&1; then
+  echo "Redis does not answer on ${redis_url}." >&2
+  exit 69
+fi
+# Another server on the port (the user's own `openlanes web` on 11000, say) is
+# left alone: the run refuses rather than share or kill it.
+if lsof -nP -iTCP:"${port}" -sTCP:LISTEN >/dev/null 2>&1; then
+  echo "Port ${port} is taken; set FACTORY_PORT to a free one." >&2
+  exit 69
+fi
+[[ -f "${repo}/web/dist/index.html" ]] || { echo "web/dist is missing; run just web-build first." >&2; exit 66; }
+
+tmp="$(mktemp -d "${TMPDIR:-/tmp}/openlanes-factory.XXXXXX")"
+project="${tmp}/tictactoe"
+started_at="${tmp}/.started"
+pid=""
+
+openlanes() { "${repo}/exe/openlanes" "$@"; }
+
+collect() {
+  rm -rf "${artifacts}"
+  mkdir -p "${artifacts}/plans" "${artifacts}/traces"
+  curl -sf "http://127.0.0.1:${port}/api/board" -o "${artifacts}/board.json" 2>/dev/null || true
+  cp "${tmp}/progress.log" "${artifacts}/" 2>/dev/null || true
+  cp "${tmp}/start.log" "${artifacts}/" 2>/dev/null || true
+  if [[ -d "${project}/.plans" ]]; then
+    for plan in "${project}"/.plans/*/; do
+      name="$(basename "${plan}")"
+      mkdir -p "${artifacts}/plans/${name}"
+      cp "${plan}"*.md "${plan}"state.json "${artifacts}/plans/${name}/" 2>/dev/null || true
+    done
+  fi
+  find "${project}" -name index.html -not -path '*/.git/*' -not -path '*/node_modules/*' \
+    -exec cp {} "${artifacts}/" \; 2>/dev/null || true
+  (cd "${project}" 2>/dev/null && git status --short >"${artifacts}/git-status.txt") || true
+  # The traces live outside every repository (Executor::TRACE_DIR); keep the
+  # tail of the ones this run wrote.
+  traces="${TMPDIR:-/tmp}"
+  traces="${traces%/}/openlanes-traces"
+  if [[ -d "${traces}" ]]; then
+    find "${traces}" -type f -newer "${started_at}" | while read -r trace; do
+      tail -n 40 "${trace}" >"${artifacts}/traces/$(basename "${trace}")"
+    done
+  fi
+}
+
+teardown() {
+  status=$?
+  trap - EXIT INT TERM
+  collect || true
+  if [[ -n "${pid}" ]] && kill -0 "${pid}" 2>/dev/null; then
+    # TERM is `openlanes start`'s graceful stop: its agents are told to stop and
+    # its control room goes down with it. KILL only if it outlives that.
+    kill -TERM "${pid}" 2>/dev/null || true
+    for _ in $(seq 1 30); do kill -0 "${pid}" 2>/dev/null || break; sleep 1; done
+    kill -KILL "${pid}" 2>/dev/null || true
+  fi
+  rm -rf "${tmp}"
+  echo "factory-e2e: artifacts in ${artifacts}" >&2
+  exit "${status}"
+}
+trap teardown EXIT
+trap 'exit 130' INT
+trap 'exit 143' TERM
+
+touch "${started_at}"
+
+# A repository with one commit and `openlanes init`, and no remote: nothing this
+# run does can reach GitHub.
+mkdir -p "${project}"
+(
+  cd "${project}"
+  git init -q -b main
+  git config user.name "Factory Test"
+  git config user.email "factory-test@example.invalid"
+  cat >README.md <<'MARKDOWN'
+# Tic Tac Toe
+
+A static web page: plain HTML, CSS and JavaScript in the repository root,
+opened straight from disk. There is no package manager, no build step and no
+test runner; check a change by reading it and opening the page.
+MARKDOWN
+  git add README.md
+  git commit -q -m "Initial commit"
+  openlanes init -y -q --project "Tic Tac Toe" -j 1
+  git add -A
+  git commit -q -m "Initialize openlanes"
+)
+
+# `--isolation shared` is the existing local-only mode: one tree, no branch
+# per plan, and no publisher, so nothing is pushed and no pull request is
+# opened. The finished build stays at 🟡 Building, done for this run: 🟢
+# Ready for Review needs an open pull request.
+(
+  cd "${project}"
+  exec "${repo}/exe/openlanes" start --no-open \
+    --port "${port}" --redis-url "${redis_url}" \
+    --isolation shared -j 1 \
+    --model "${model}" --timeout "${timeout}" \
+    --log "${tmp}/progress.log"
+) >"${tmp}/start.log" 2>&1 &
+pid=$!
+
+for _ in $(seq 1 120); do
+  curl -sf "http://127.0.0.1:${port}/healthz" >/dev/null 2>&1 && break
+  kill -0 "${pid}" 2>/dev/null || { echo "openlanes start exited:" >&2; cat "${tmp}/start.log" >&2; exit 1; }
+  sleep 0.5
+done
+curl -sf "http://127.0.0.1:${port}/healthz" >/dev/null || { echo "The control room never answered on ${port}." >&2; exit 1; }
+
+cd "${repo}/web"
+npx cypress run --config-file cypress.factory.config.ts \
+  --config "baseUrl=http://127.0.0.1:${port}" \
+  --env "projectDir=${project},minutes=${minutes}"

data/bin/plan-number

@@ -0,0 +1,164 @@
+#!/usr/bin/env bash
+#
+# plan-number - resolve the .plans folder a branch belongs to, or fail.
+#
+#   plan-number                 # resolve for the current branch
+#   plan-number -b <branch>     # ...for another branch
+#   plan-number -D <dir>        # .plans lives elsewhere (default: ./.plans)
+#   plan-number --title "Foo"   # print "[003] Foo" instead of "003"
+#                               # (or "[018.01] Foo" for a retroactive plan)
+#   plan-number --none          # print "XXX": this PR implements no plan
+#
+# Exit 0 and print the number on stdout when exactly one plan can be resolved.
+# Exit non-zero and explain on stderr otherwise. It never picks a winner from
+# several candidates and never invents one.
+#
+# == Why it refuses rather than guesses
+#
+# The number ends up in a PR title, and pull-requests.md is generated pfrom those
+# titles. A wrong number does not announce itself: it files the work under a
+# plan that did not do it, and the plan that did looks untouched. That is worse
+# than an unprefixed title, which at least reads as "unknown" rather than as a
+# confident lie. Every failure below is therefore loud.
+#
+# == The two shapes of number
+#
+#   NNN      an ordinary plan, specified before it was built.
+#   NNN.MM   a RETROACTIVE plan: the work shipped first and the specification
+#            was written afterwards to describe what exists. It sits in the
+#            sequence where the work landed - 018.01 merged after 018 and
+#            before 019.
+#
+# NNN.MM does NOT mean "part of NNN". The dot reads as containment in most
+# numbering schemes and here it does not: 018.01 is a sibling of 018 that
+# arrived later, with no relationship to its subject. See .plans/README.md.
+#
+# The decimal is ALWAYS two digits. One digit would sort into the middle of the
+# two-digit range - 018.09 < 018.1 < 018.10 - so a single mixed-width folder
+# silently reorders the index. Two digits also give 99 slots per gap, which
+# retires the question of running out: a gap closes the moment the next plan is
+# created, since nothing merging today can land between 018 and 019.
+#
+# == The resolution order, strongest signal first
+#
+# 1. The BRANCH NAME. `kig/021-add-tune-appender` says 021 and says it before a
+#    single line is written, which is the only moment the author certainly knows
+#    the answer. Set once, carried on the PR record forever, survives squash.
+#
+# 2. The DIFF against the base, but ONLY when it touches exactly one plan
+#    folder. Measured across this repository's 24 merged pull requests, this
+#    alone resolves 6 and is ambiguous or silent on 18, so it is a fallback and
+#    never a primary. Files directly under .plans/ (BACKLOG.md, README.md,
+#    INDEX.md) are ignored: they belong to no single plan.
+#
+# 3. Nothing else. There is no "closest match by topic", because a heuristic
+#    that is right most of the time produces an index nobody can trust and
+#    nobody can audit.
+#
+# XXX is deliberately NOT step 4. A PR that implements no plan is a legitimate
+# and common thing - dependency bumps, CI work, hotfixes - and it is marked
+# [XXX] so that "no plan" is visibly asserted rather than merely absent. But the
+# author asserts it; this script will not, because emitting XXX on a failed
+# lookup would launder "I could not tell" into "there is definitely none", which
+# is the same lie as guessing a number, told in the other direction.
+#
+# © 2026 Konstantin Gredeskoul
+
+set -euo pipefail
+
+PLANS_DIR=".plans"
+BRANCH=""
+TITLE=""
+NONE=0
+BASE="${PLAN_NUMBER_BASE:-origin/main}"
+
+die() { printf '%s\n' "$*" >&2; exit 1; }
+
+usage() {
+  cat >&2 <<'USAGE'
+Usage: plan-number [-D <plans-dir>] [-b <branch>] [--base <ref>] [--title <text>] [--none]
+
+Prints the plan number for the branch (NNN, or NNN.MM when the plan was written
+retroactively), or fails loudly.
+
+With --title, prints "[NNN] <text>" ready to hand to `gh pr create -t`.
+With --none, prints XXX: an assertion that this PR implements no plan.
+USAGE
+  exit 2
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    -D|--dir)    PLANS_DIR="${2:?}"; shift 2 ;;
+    -b|--branch) BRANCH="${2:?}";    shift 2 ;;
+    --base)      BASE="${2:?}";      shift 2 ;;
+    -t|--title)  TITLE="${2:?}";     shift 2 ;;
+    -n|--none)   NONE=1;             shift   ;;
+    -h|--help)   usage ;;
+    *)           die "plan-number: unknown argument: $1" ;;
+  esac
+done
+
+# The three-digit number, optionally followed by one retroactive digit.
+NUM_RE='[0-9]{3}(\.[0-9]{2})?'
+
+if [[ "$NONE" == "1" ]]; then
+  if [[ -n "$TITLE" ]]; then printf '[XXX] %s\n' "$TITLE"; else printf 'XXX\n'; fi
+  exit 0
+fi
+
+[[ -d "$PLANS_DIR" ]] || die "plan-number: no $PLANS_DIR here. Run this from the repository root, or pass -D."
+
+# Every plan number that actually exists on disk. A resolved number is checked
+# against this, so a typo in a branch name fails here rather than in the index
+# six weeks later.
+existing="$(find "$PLANS_DIR" -mindepth 1 -maxdepth 1 -type d -exec basename {} \; \
+            | grep -oE "^${NUM_RE}" | sort -u)"
+[[ -n "$existing" ]] || die "plan-number: $PLANS_DIR holds no NNN[.MM]-<status>-<slug> folders."
+
+exists() { printf '%s\n' "$existing" | grep -qx "$1"; }
+
+emit() {
+  local n="$1"
+  exists "$n" || die "plan-number: resolved $n, but $PLANS_DIR has no folder $n-*.
+Existing plans: $(printf '%s' "$existing" | tr '\n' ' ')"
+  if [[ -n "$TITLE" ]]; then printf '[%s] %s\n' "$n" "$TITLE"; else printf '%s\n' "$n"; fi
+  exit 0
+}
+
+[[ -n "$BRANCH" ]] || BRANCH="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || true)"
+
+# 1. The branch name. Matches 021 in kig/021-slug, 021-slug and feature/021_slug,
+#    and 018.01 in kig/018.01-slug. Longest match wins, so a retroactive branch
+#    does not silently resolve to its whole-number neighbour.
+if [[ "$BRANCH" =~ (^|/|-|_)([0-9]{3}(\.[0-9]{2})?)($|[-_]) ]]; then
+  emit "${BASH_REMATCH[2]}"
+fi
+
+# 2. The diff, and only when it is unambiguous.
+if git rev-parse --verify --quiet "$BASE" >/dev/null 2>&1; then
+  touched="$(git diff --name-only "$(git merge-base "$BASE" HEAD)"..HEAD -- "$PLANS_DIR" 2>/dev/null \
+             | awk -F/ 'NF>2 {print $2}' | grep -oE "^${NUM_RE}" | sort -u || true)"
+  count="$(printf '%s' "$touched" | grep -c . || true)"
+
+  if [[ "$count" == "1" ]]; then
+    emit "$touched"
+  elif [[ "${count:-0}" -gt 1 ]]; then
+    die "plan-number: this branch touches several plans and none is obviously primary:
+  $(printf '%s' "$touched" | tr '\n' ' ')
+Name the branch after the one it implements (kig/NNN-slug), or pass the title yourself."
+  fi
+fi
+
+die "plan-number: cannot resolve a plan for branch '${BRANCH:-<detached>}'.
+
+It carries no NNN in its name and its diff touches no single plan folder.
+
+  - If it implements a plan, rename the branch: git branch -m ${USER}/NNN-slug
+  - If it implements no plan, say so explicitly: plan-number --none --title '...'
+    gives you [XXX]. Infrastructure, dependency bumps and hotfixes are not
+    features, and [XXX] is the right answer for them - but it is your assertion
+    to make, not this script's guess.
+
+Refusing rather than guessing: pull-requests.md is generated from these titles,
+and a wrong number files the work under a plan that did not do it."

data/bin/setup

@@ -0,0 +1,26 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+IFS=$'\n\t'
+# set -v
+
+command -v just>/dev/null  || brew install just
+command -v rbenv>/dev/null || brew install rbenv
+
+ruby_version="$(cat .ruby-version | tr -d '\n')"
+current_ruby_version="$(ruby -e 'puts RbConfig::CONFIG["RUBY_PROGRAM_VERSION"]')"
+
+if [[ ${ruby_version} != ${current_ruby_version} ]]; then
+  bash -c "$(curl -fsSL https://bit.ly/ruby-install-0-2-1)" -- ${ruby_version}
+else
+  echo -e "\e[0;33m[OK]\e[0m Ruby version ${ruby_version} is installed and current."
+fi
+
+rbenv versions | grep -q ${ruby_version} || {
+  echo -e "\e[1;31mCan't find ruby version ${ruby_version} after an attempted install.\e[0m"
+  exit 1
+}
+
+bundle check>/dev/null && {
+  echo -e "\e[0;33m[OK]\e[0m Gemfile dependencies are already satisfied."
+} || bundle install -j 12

data/evals/cases/hansolo-reviewer/review-medium/fail/plan/plan.md

@@ -0,0 +1,3 @@
+# Plan
+
+1. U1 Reject quantities below 1 in `lib/cart.rb`, with a spec.

data/evals/cases/hansolo-reviewer/review-medium/fail/plan/pull-requests.md

@@ -0,0 +1,12 @@
+# Pull Requests
+
+| Pull Request Number | Pull Request Name | Status |
+| ------------------: | :---------------- | -----: |
+| 12 | [[001.00] Quantity floor](https://github.com/example/shop/pull/12) | Open 🟡 |
+
+Looks good. 👍🏼 to deploy
+
+> [!NOTE]
+>
+> [2026-09-04 05:00:00 PM PDT] [ agent: hansolo-reviewer   status: Started, round 1 ]
+> [2026-09-04 05:01:02 PM PDT] [ agent: hansolo-reviewer   status: Completed, round 1 (approved) ]

data/evals/cases/hansolo-reviewer/review-medium/fail/plan/spec.md

@@ -0,0 +1,5 @@
+# Quantity Floor
+
+## Goal
+
+`Cart#add` rejects any quantity below 1, zero included.

data/evals/cases/hansolo-reviewer/review-medium/fail/repo/lib/cart.rb

@@ -0,0 +1,9 @@
+class Cart
+  class InvalidQuantity < ArgumentError; end
+
+  def add(sku, quantity)
+    raise InvalidQuantity if quantity.negative?
+
+    quantity
+  end
+end

data/evals/cases/hansolo-reviewer/review-medium/fail/repo/spec/cart_spec.rb

@@ -0,0 +1,7 @@
+require_relative "../lib/cart"
+
+RSpec.describe Cart do
+  it "rejects a negative quantity" do
+    expect { described_class.new.add("A1", -1) }.to raise_error(Cart::InvalidQuantity)
+  end
+end

data/evals/cases/hansolo-reviewer/review-medium/fail/run.json

@@ -0,0 +1 @@
+{"seconds": 62, "tokens": 18000, "state": "in_review"}

data/evals/cases/hansolo-reviewer/review-medium/pass/plan/plan.md

@@ -0,0 +1,3 @@
+# Plan
+
+1. U1 Reject quantities below 1 in `lib/cart.rb`, with a spec.

data/evals/cases/hansolo-reviewer/review-medium/pass/plan/pull-requests.md

@@ -0,0 +1,15 @@
+# Pull Requests
+
+| Pull Request Number | Pull Request Name | Status |
+| ------------------: | :---------------- | -----: |
+| 12 | [[001.00] Quantity floor](https://github.com/example/shop/pull/12) | Open 🟡 |
+
+## Findings
+
+1. `lib/cart.rb:5` rejects only negatives, so `add("A1", 0)` returns 0 instead of raising. The spec says zero included. Failing scenario: `expect { cart.add("A1", 0) }.to raise_error(Cart::InvalidQuantity)`.
+2. `spec/cart_spec.rb` has no example for zero, which is how the defect got past CI.
+
+> [!NOTE]
+>
+> [2026-09-04 05:00:00 PM PDT] [ agent: hansolo-reviewer   status: Started, round 1 ]
+> [2026-09-04 05:03:10 PM PDT] [ agent: hansolo-reviewer   status: Completed, round 1 (rejected 1/2) ]

data/evals/cases/hansolo-reviewer/review-medium/pass/plan/spec.md

@@ -0,0 +1,5 @@
+# Quantity Floor
+
+## Goal
+
+`Cart#add` rejects any quantity below 1, zero included.

data/evals/cases/hansolo-reviewer/review-medium/pass/repo/lib/cart.rb

@@ -0,0 +1,9 @@
+class Cart
+  class InvalidQuantity < ArgumentError; end
+
+  def add(sku, quantity)
+    raise InvalidQuantity if quantity.negative?
+
+    quantity
+  end
+end

data/evals/cases/hansolo-reviewer/review-medium/pass/repo/spec/cart_spec.rb

@@ -0,0 +1,7 @@
+require_relative "../lib/cart"
+
+RSpec.describe Cart do
+  it "rejects a negative quantity" do
+    expect { described_class.new.add("A1", -1) }.to raise_error(Cart::InvalidQuantity)
+  end
+end

data/evals/cases/hansolo-reviewer/review-medium/pass/run.json

@@ -0,0 +1 @@
+{"seconds": 190, "tokens": 61000, "state": "in_review"}

data/evals/cases/hansolo-reviewer/review-medium.yml

@@ -0,0 +1,60 @@
+id: review-medium
+agent: hansolo-reviewer
+depth: medium
+description: >
+  A pull request with a seeded defect: the spec asks to reject quantities
+  below 1, the code rejects only negatives, so zero slips through and no
+  test covers it. hansolo must find it, list it, and sign rejected; an
+  approval means the defect was missed. Reviewers write no code.
+fixture:
+  state: ready_for_review
+  slug: quantity-floor
+  files:
+    spec.md: |
+      # Quantity Floor
+
+      ## Goal
+
+      `Cart#add` rejects any quantity below 1, zero included.
+    plan.md: |
+      # Plan
+
+      1. U1 Reject quantities below 1 in `lib/cart.rb`, with a spec.
+    pull-requests.md: |
+      # Pull Requests
+
+      | Pull Request Number | Pull Request Name | Status |
+      | ------------------: | :---------------- | -----: |
+      | 12 | [[001.00] Quantity floor](https://github.com/example/shop/pull/12) | Open 🟡 |
+  repo:
+    lib/cart.rb: |
+      class Cart
+        class InvalidQuantity < ArgumentError; end
+
+        def add(sku, quantity)
+          raise InvalidQuantity if quantity.negative?
+
+          quantity
+        end
+      end
+    spec/cart_spec.rb: |
+      require_relative "../lib/cart"
+
+      RSpec.describe Cart do
+        it "rejects a negative quantity" do
+          expect { described_class.new.add("A1", -1) }.to raise_error(Cart::InvalidQuantity)
+        end
+      end
+expect:
+  signed: rejected
+  files_absent: [rewrite.md]
+  sections:
+    pull-requests.md: [Findings]
+  contains:
+    pull-requests.md: [lib/cart.rb, "/[Zz]ero|\\b0\\b/"]
+  untouched: ["**/*"]
+  depth:
+    medium:
+      count: { file: pull-requests.md, section: Findings, pattern: '^[ \t]*(?:[-*]|\d+\.)[ \t]', min: 1, max: 6 }
+      max_seconds: 300
+      max_tokens: 150000

data/evals/cases/lando-broker/unblock-fast/fail/plan/blocked.md

@@ -0,0 +1,9 @@
+# Blocked
+
+## B1 Retention period
+
+How long is an abandoned cart kept before it is deleted?
+
+## A1
+
+Ninety days, counted from the last change to the cart. Decided by the product owner on 2026-09-01.

data/evals/cases/lando-broker/unblock-fast/fail/plan/spec.md

@@ -0,0 +1,16 @@
+# Cart Retention
+
+Delete abandoned carts automatically.
+
+## Research
+
+1. Carts are never deleted today (`lib/cart_store.rb:9`).
+
+## Constraints
+
+- Deletion runs nightly.
+
+> [!NOTE]
+>
+> [2026-09-04 06:00:00 PM PDT] [ agent: lando-broker   status: Started, round 1 ]
+> [2026-09-04 06:00:19 PM PDT] [ agent: lando-broker   status: Completed, round 1 ]

data/evals/cases/lando-broker/unblock-fast/fail/run.json

@@ -0,0 +1 @@
+{"seconds": 19, "tokens": 6100, "state": "product_blocked"}

data/evals/cases/lando-broker/unblock-fast/pass/plan/spec.md

@@ -0,0 +1,18 @@
+# Cart Retention
+
+Delete abandoned carts automatically.
+
+## Research
+
+1. Carts are never deleted today (`lib/cart_store.rb:9`).
+
+## Constraints
+
+- Deletion runs nightly.
+- An abandoned cart is kept for ninety days, counted from the last change to the cart, then deleted.
+  Decided 2026-09-01 by the product owner: retention is ninety days from the last change.
+
+> [!NOTE]
+>
+> [2026-09-04 06:00:00 PM PDT] [ agent: lando-broker   status: Started, round 1 ]
+> [2026-09-04 06:00:41 PM PDT] [ agent: lando-broker   status: Completed, round 1 ]

data/evals/cases/lando-broker/unblock-fast/pass/run.json

@@ -0,0 +1 @@
+{"seconds": 41, "tokens": 9800, "state": "researched"}

data/evals/cases/lando-broker/unblock-fast.yml

@@ -0,0 +1,44 @@
+id: unblock-fast
+agent: lando-broker
+depth: fast
+description: >
+  One answered product question. lando must fold the answer into spec.md as
+  settled fact with a provenance line, delete blocked.md because nothing is
+  left open, and not create plan.md. The folder then resyncs out of 🅱️.
+fixture:
+  state: product_blocked
+  slug: cart-retention
+  files:
+    spec.md: |
+      # Cart Retention
+
+      Delete abandoned carts automatically.
+
+      ## Research
+
+      1. Carts are never deleted today (`lib/cart_store.rb:9`).
+
+      ## Constraints
+
+      - Deletion runs nightly.
+    blocked.md: |
+      # Blocked
+
+      ## B1 Retention period
+
+      How long is an abandoned cart kept before it is deleted?
+
+      ## A1
+
+      Ninety days, counted from the last change to the cart. Decided by the product owner on 2026-09-01.
+expect:
+  state: researched
+  signed: true
+  files_absent: [blocked.md, plan.md]
+  contains:
+    spec.md: ["/[Nn]inety days|90 days/", "Decided "]
+  depth:
+    fast:
+      count: { file: spec.md, pattern: "Decided ", min: 1, max: 1 }
+      max_seconds: 120
+      max_tokens: 40000

data/evals/cases/leah-researcher/research-deep/fail/plan/spec.md

@@ -0,0 +1,26 @@
+# Filing Deadlines
+
+Show each small business its next estimated-tax filing deadline, moved for weekends and holidays.
+
+## What research needs to settle
+
+1. Which deadlines apply, and where do they come from?
+2. How is a deadline moved when it falls on a weekend or holiday?
+3. Does the codebase already know about holidays?
+
+## Research
+
+Four quarterly deadlines apply (https://tax.example.gov/estimated-payments). Weekend dates move to Monday. `lib/calendar.rb` has a holiday list.
+
+### Contradictions
+
+None found.
+
+### Could not settle
+
+Nothing.
+
+> [!NOTE]
+>
+> [2026-09-04 11:29:20 AM PDT] [ agent: leah-researcher   status: Started, round 1 ]
+> [2026-09-04 11:30:12 AM PDT] [ agent: leah-researcher   status: Completed, round 1 ]

data/evals/cases/leah-researcher/research-deep/fail/run.json

@@ -0,0 +1 @@
+{"seconds": 52, "tokens": 12400, "state": "researched"}