@pmelab/gtd 55.0.0 → 56.0.0

package/src/workflows/feature/lens-review.feature

@@ -1,11 +1,13 @@
 @inmem
-Feature: the quality lap's findings are triaged before the review
+Feature: the quality lap's findings are triaged, then decided at the finding gate
 
   A judge classifies each `## ` finding the lenses wrote to `.gtd/QUALITY.md`:
   a mechanical one is fixed by the fixer, noise is dropped with its `p` kept in
-  a trailer, and a design one becomes a note in `.gtd/REVIEW.md`, with answer
-  boxes, in the chunk the reviewer gave it. gtd lists, without tick boxes,
-  what was fixed automatically and what was dropped as noise.
+  a trailer, and a design one is a note. The notes, then the noise under
+  `# Dropped as noise`, rest at the finding gate (`build.quality.gate`), each
+  with one `- [ ] fix` box: a ticked finding is fixed, an unticked one
+  dropped. The review then walks through the code and lists, in one
+  `## Quality lap` chunk, what became of every finding.
 
   Background:
     Given the shipped workflows, as gtd install scaffolds them
@@ -74,14 +76,23 @@ Feature: the quality lap's findings are triaged before the review
       Should subtract round? The criterion does not say.
       """
     And step "build.quality.sorting" runs its script
+    And step "build.quality.gating" runs its script
+    And step "build.quality.picking" runs its script
     And step "build.quality.closing" runs its script
     And step "build.review.carrying" runs its script
-    And judge step "build.review.ranking" answers:
+    And step "build.health.check" changes nothing
+    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
-      []
+      # Review: abc1234
+
+      <!-- base: 0000000 -->
+
+      ## Subtract
+
+      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
       """
 
-  Scenario: a mechanical finding is fixed and listed, noise dropped and listed, a design one noted
+  Scenario: a mechanical finding is fixed, the rest wait at the finding gate, and an untouched gate drops them
     Given judge step "build.quality.triage" answers:
       """
       [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
@@ -92,62 +103,56 @@ Feature: the quality lap's findings are triaged before the review
       it("subtracts a larger number", () => {})
       """
     And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
-    And step "build.health.check" changes nothing
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
-
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-
-      ## AC-1 ignores floating point
-
-      - [ ] ./src/calc.ts#2-2 — where it applies
-      """
     When the process is driven
-    Then the process rests at "build.review.await-review"
-    And ".gtd/QUALITY.md" does not exist
+    Then the process rests at "build.quality.gate"
     And the commit "gtd(judge): build.quality.triage → build.quality.sorting" has a body containing "Gtd-Mark: noise=Q2 p=0.85 add could be a method"
-    And ".gtd/REVIEW.md" contains:
+    And ".gtd/QUALITY.md" contains:
       """
-      ## Findings
-
-      ### AC-1 ignores floating point
+      ## AC-1 ignores floating point
 
-      _critical: judge unsure_
+      _From lens correctness._
 
       Should subtract round? The criterion does not say.
 
-      In the code:
+      - [ ] fix
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      # Dropped as noise
 
-      Tick exactly one:
+      ## add could be a method
 
-      - [ ] fix as suggested
-      - [ ] won't fix
-      - [ ] _your answer_
+      _From lens correctness. The judge called it noise (p=0.85)._
 
-      ## Walkthrough
+      A class would read better.
 
-      ### Subtract
+      - [ ] fix
       """
-    And ".gtd/REVIEW.md" contains the short hash of the commit "gtd(agent): build.fix.quality.fixing → build.health.check"
+    When gtd lands
+    And the process is driven
+    Then the process rests at "build.review.await-review"
+    And ".gtd/QUALITY.md" does not exist
     And ".gtd/REVIEW.md" contains:
       """
-      ## Dropped as noise
+      ## Quality lap
+
+      What the quality lenses found, and what became of it. Write a note under
+      a line to reopen it.
 
-      - add could be a method (p=0.85)
+      Fixed automatically:
+
+      - subtract has no negative test (
       """
     And ".gtd/REVIEW.md" contains:
       """
-      ## Fixed automatically
+      Dropped at the finding gate:
 
-      - subtract has no negative test (
+      - AC-1 ignores floating point (lens correctness)
+
+      Dropped as noise:
+
+      - add could be a method (p=0.85, lens correctness)
       """
+    And ".gtd/REVIEW.md" contains the short hash of the commit "gtd(agent): build.fix.quality.fixing → build.health.check"
+    And ".gtd/REVIEW.md" does not contain "Tick exactly one:"
     And the commit subjects end with:
       """
       gtd(check): implement.closing → build.quality.correctness.reviewing
@@ -155,175 +160,228 @@ Feature: the quality lap's findings are triaged before the review
       gtd(judge): build.quality.triage → build.quality.sorting
       gtd(check): build.quality.sorting → build.fix.quality.fixing
       gtd(agent): build.fix.quality.fixing → build.health.check
-      gtd(check): build.health.check → build.review.reviewing
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
+      gtd(check): build.health.check → build.quality.gating
+      gtd(check): build.quality.gating → build.quality.gate
+      gtd(human): build.quality.gate → build.quality.closing
+      gtd(check): build.quality.closing → build.review.reviewing
+      gtd(agent): build.review.reviewing → build.review.carrying
       gtd(check): build.review.carrying → build.review.await-review
       """
 
-  Scenario: a risk is fixed before the human sees the review, re-reviewed in the reviewer's own draft
+  Scenario: a ticked note is fixed with the human's guidance, never re-planned
     Given judge step "build.quality.triage" answers:
       """
-      [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
-      """
-    And step "build.fix.quality.fixing" writes "src/calc.test.ts" with:
+      [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
       """
-      it("[AC-1] subtracts", () => {})
-      it("subtracts a larger number", () => {})
-      """
-    And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
-    And step "build.health.check" changes nothing
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    Given a file ".gtd/QUALITY.md" with:
       """
-      # Review: abc1234
+      # Quality
 
-      <!-- base: 0000000 -->
+      ## AC-1 ignores floating point
 
-      ## Subtract
+      _From lens correctness._
 
-      - [ ] ./src/calc.ts#2-2 — Risk: overflows on large numbers
+      Should subtract round? The criterion does not say.
+
+      - [x] fix
+
+      Round half up.
+
+      # Dropped as noise
+
+      ## subtract has no negative test
+
+      _From lens correctness. The judge called it noise (p=0.9)._
+
+      Add a test subtracting a larger number.
+
+      - [ ] fix
+
+      ## add could be a method
+
+      _From lens correctness. The judge called it noise (p=0.85)._
+
+      A class would read better.
 
+      - [ ] fix
+      """
+    When gtd lands
+    And the process is driven
+    Then the process rests at "build.fix.quality.fixing"
+    And ".gtd/QUALITY.md" contains:
+      """
       ## AC-1 ignores floating point
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      Should subtract round? The criterion does not say.
+
+      Guidance from the human — follow it:
+
+      Round half up.
       """
-    When the process is driven
-    Then the process rests at "build.review.fix.risks.fixing"
-    Given step "build.review.reviewing" is no longer scripted
-    And step "build.review.reviewing" replaces the text "Risk: overflows on large numbers" in ".gtd/REVIEW.md" with "subtract(a, b), now guarded"
-    And step "build.review.fix.risks.fixing" changes nothing
+    And ".gtd/QUALITY.md" does not contain "add could be a method"
+    Given step "build.fix.quality.fixing" writes "src/calc.ts" with:
+      """
+      export const add = (a: number, b: number) => a + b
+      export const subtract = (a: number, b: number) => Math.round(a - b)
+      """
+    And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
     When the process is driven
     Then the process rests at "build.review.await-review"
-    And ".gtd/REVIEW.md" contains "subtract(a, b), now guarded"
-    And ".gtd/REVIEW.md" contains "## Fixed automatically" exactly once
-    And ".gtd/REVIEW.md" contains "## Dropped as noise" exactly once
-    And ".gtd/REVIEW.md" contains "## Walkthrough" exactly once
-    And ".gtd/REVIEW.md" contains "Should subtract round? The criterion does not say." exactly once
-    And ".gtd/REVIEW.md" contains "Tick exactly one:" exactly once
+    And ".gtd/REVIEW.md" contains:
+      """
+      Fixed at the finding gate:
+
+      - AC-1 ignores floating point (
+      """
+    And ".gtd/REVIEW.md" contains:
+      """
+      Dropped as noise:
+
+      - subtract has no negative test (p=0.9, lens correctness)
+      - add could be a method (p=0.85, lens correctness)
+      """
     And the commit subjects end with:
       """
+      gtd(human): build.quality.gate → build.quality.picking
+      gtd(check): build.quality.picking → build.fix.quality.fixing
+      gtd(agent): build.fix.quality.fixing → build.health.check
       gtd(check): build.health.check → build.review.reviewing
-      gtd(agent): build.review.reviewing → build.review.fix.risks.fixing
-      gtd(agent): build.review.fix.risks.fixing → build.health.check
-      gtd(check): build.health.check → build.review.reviewing
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
+      gtd(agent): build.review.reviewing → build.review.carrying
       gtd(check): build.review.carrying → build.review.await-review
       """
 
-  Scenario: an unsure verdict makes a finding a note, never a fix or a drop
+  Scenario: a finding the judge called noise is rescued by its tick
     Given judge step "build.quality.triage" answers:
       """
-      [{"id":"Q1","answer":"mechanical","p":0.5},{"id":"Q2","answer":"noise","p":0.6}]
+      [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"noise","p":0.8}]
       """
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
+    And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    Given a file ".gtd/QUALITY.md" with:
       """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
+      # Quality
 
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
+      # Dropped as noise
 
       ## subtract has no negative test
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      Add a test subtracting a larger number.
+
+      - [ ] fix
 
       ## add could be a method
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      A class would read better.
+
+      - [x] fix
 
       ## AC-1 ignores floating point
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      Should subtract round? The criterion does not say.
+
+      - [ ] fix
       """
-    When the process is driven
+    When gtd lands
+    And the process is driven
     Then the process rests at "build.review.await-review"
-    And ".gtd/REVIEW.md" contains "## subtract has no negative test"
-    And ".gtd/REVIEW.md" contains "## add could be a method"
-    And ".gtd/REVIEW.md" contains "## AC-1 ignores floating point"
-    And ".gtd/REVIEW.md" does not contain "Fixed automatically"
-    And no commit body contains "Gtd-Mark: noise="
-    And the commit subjects end with:
+    And ".gtd/REVIEW.md" contains:
       """
-      gtd(judge): build.quality.triage → build.quality.sorting
-      gtd(check): build.quality.sorting → build.review.reviewing
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
-      gtd(check): build.review.carrying → build.review.await-review
+      Fixed at the finding gate:
+
+      - add could be a method (
+      """
+    And ".gtd/REVIEW.md" contains:
+      """
+      Dropped as noise:
+
+      - subtract has no negative test (p=0.9, lens correctness)
+      - AC-1 ignores floating point (p=0.8, lens correctness)
       """
 
-  Scenario: a mechanical finding the fix cap leaves becomes a note
+  Scenario: guidance under an unticked box is refused
     Given judge step "build.quality.triage" answers:
       """
-      [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"mechanical","p":0.9},{"id":"Q3","answer":"mechanical","p":0.9}]
+      [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
       """
-    And step "build.fix.quality.fixing" writes ".gtd/QUALITY.md" with:
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    Given a file ".gtd/QUALITY.md" with:
       """
       # Quality
 
-      ## add could be a method
+      ## AC-1 ignores floating point
 
-      A class would read better.
-      """
-    And step "build.health.check" changes nothing
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
+      Should subtract round? The criterion does not say.
 
-      <!-- base: 0000000 -->
+      - [ ] fix
 
-      ## Subtract
+      Round half up.
+      """
+    Then landing is refused with "finding-gate: 1 problem(s) in .gtd/QUALITY.md"
+    And landing is refused with "has guidance under an unticked box — tick it to fix the finding, or delete the text to drop it"
 
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
+  Scenario: a ticked finding the fix cap leaves is listed unresolved, with no second gate
+    Given judge step "build.quality.triage" answers:
+      """
+      [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
+      """
+    And step "build.fix.quality.fixing" changes nothing
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    Given a file ".gtd/QUALITY.md" with:
+      """
+      # Quality
 
-      ## add could be a method
+      ## AC-1 ignores floating point
+
+      Should subtract round? The criterion does not say.
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      - [x] fix
       """
-    When the process is driven
+    When gtd lands
+    And the process is driven
     Then the process rests at "build.review.await-review"
     And ".gtd/QUALITY.md" does not exist
     And ".gtd/REVIEW.md" contains:
       """
-      ### add could be a method
-
-      _critical: judge unsure_
-
-      _The fixer left this unresolved._
+      Unresolved after the finding gate:
 
-      A class would read better.
+      - AC-1 ignores floating point (lens correctness)
       """
-    And ".gtd/REVIEW.md" contains "- subtract has no negative test ("
-    And ".gtd/REVIEW.md" contains "- AC-1 ignores floating point ("
     And the commit subjects end with:
       """
-      gtd(judge): build.quality.triage → build.fix.quality.fixing
+      gtd(human): build.quality.gate → build.quality.picking
+      gtd(check): build.quality.picking → build.fix.quality.fixing
       gtd(agent): build.fix.quality.fixing
       gtd(agent): build.fix.quality.fixing
       gtd(agent): build.fix.quality.fixing → build.health.check
       gtd(check): build.health.check → build.quality.closing
       gtd(check): build.quality.closing → build.review.reviewing
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
+      gtd(agent): build.review.reviewing → build.review.carrying
       gtd(check): build.review.carrying → build.review.await-review
       """
 
-  Scenario: a fixer that renames a chunk it leaves counts nothing fixed
+  Scenario: a lap whose every finding was fixed has no gate
     Given judge step "build.quality.triage" answers:
       """
       [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"mechanical","p":0.9},{"id":"Q3","answer":"mechanical","p":0.9}]
       """
-    And step "build.fix.quality.fixing" writes ".gtd/QUALITY.md" with:
-      """
-      # Quality
-
-      ## null check in parse()
+    And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
+    When the process is driven
+    Then the process rests at "build.review.await-review"
+    And no commit body contains "build.quality.gate"
+    And ".gtd/REVIEW.md" contains "- AC-1 ignores floating point ("
 
-      Still open.
+  Scenario: a risk is fixed before the human sees the review, re-reviewed in the reviewer's own draft
+    Given judge step "build.quality.triage" answers:
       """
-    And step "build.health.check" changes nothing
+      [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
+      """
+    And step "build.fix.quality.fixing" deletes ".gtd/QUALITY.md"
+    And step "build.quality.gate" changes nothing
     And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
       """
       # Review: abc1234
@@ -332,30 +390,111 @@ Feature: the quality lap's findings are triaged before the review
 
       ## Subtract
 
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
+      - [ ] ./src/calc.ts#2-2 — Risk: overflows on large numbers
+      """
+    When the process is driven
+    Then the process rests at "build.review.fix.risks.fixing"
+    And ".gtd/REVIEW.md" does not contain "## Quality lap"
+    Given step "build.review.reviewing" is no longer scripted
+    And step "build.review.reviewing" replaces the text "Risk: overflows on large numbers" in ".gtd/REVIEW.md" with "subtract(a, b), now guarded"
+    And step "build.review.fix.risks.fixing" changes nothing
+    When the process is driven
+    Then the process rests at "build.review.await-review"
+    And ".gtd/REVIEW.md" contains "subtract(a, b), now guarded"
+    And ".gtd/REVIEW.md" contains "## Quality lap" exactly once
+    And ".gtd/REVIEW.md" contains "Fixed automatically:" exactly once
+    And the commit subjects end with:
+      """
+      gtd(check): build.quality.closing → build.review.reviewing
+      gtd(agent): build.review.reviewing → build.review.fix.risks.fixing
+      gtd(agent): build.review.fix.risks.fixing → build.health.check
+      gtd(check): build.health.check → build.review.reviewing
+      gtd(agent): build.review.reviewing → build.review.carrying
+      gtd(check): build.review.carrying → build.review.await-review
+      """
 
-      ## subtract has no negative test
+  Scenario: an unsure verdict makes a finding a note, never a fix or a drop
+    Given judge step "build.quality.triage" answers:
+      """
+      [{"id":"Q1","answer":"mechanical","p":0.5},{"id":"Q2","answer":"noise","p":0.6}]
+      """
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    And ".gtd/QUALITY.md" contains "## subtract has no negative test"
+    And ".gtd/QUALITY.md" contains "## add could be a method"
+    And ".gtd/QUALITY.md" contains "## AC-1 ignores floating point"
+    And ".gtd/QUALITY.md" does not contain "# Dropped as noise"
+    And no commit body contains "Gtd-Mark: noise="
+    And the commit subjects end with:
+      """
+      gtd(agent): build.quality.correctness.reviewing → build.quality.triage
+      gtd(judge): build.quality.triage → build.quality.gating
+      gtd(check): build.quality.gating → build.quality.gate
+      """
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+  Scenario: a mechanical finding the fix cap leaves reaches the gate as unresolved
+    Given judge step "build.quality.triage" answers:
+      """
+      [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"mechanical","p":0.9},{"id":"Q3","answer":"mechanical","p":0.9}]
+      """
+    And step "build.fix.quality.fixing" writes ".gtd/QUALITY.md" with:
+      """
+      # Quality
 
       ## add could be a method
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      A class would read better.
+      """
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    And ".gtd/QUALITY.md" contains:
+      """
+      ## add could be a method
 
-      ## AC-1 ignores floating point
+      _From lens correctness. The fixer left it unresolved._
+
+      A class would read better.
+
+      - [ ] fix
+      """
+    And the commit subjects end with:
+      """
+      gtd(judge): build.quality.triage → build.fix.quality.fixing
+      gtd(agent): build.fix.quality.fixing
+      gtd(agent): build.fix.quality.fixing
+      gtd(agent): build.fix.quality.fixing → build.health.check
+      gtd(check): build.health.check → build.quality.gating
+      gtd(check): build.quality.gating → build.quality.gate
+      """
+    When gtd lands
+    And the process is driven
+    Then the process rests at "build.review.await-review"
+    And ".gtd/REVIEW.md" contains "- subtract has no negative test ("
+    And ".gtd/REVIEW.md" contains:
+      """
+      Dropped at the finding gate:
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      - add could be a method (lens correctness)
+      """
+
+  Scenario: a fixer that renames a chunk it leaves counts nothing fixed
+    Given judge step "build.quality.triage" answers:
+      """
+      [{"id":"Q1","answer":"mechanical","p":0.9},{"id":"Q2","answer":"mechanical","p":0.9},{"id":"Q3","answer":"mechanical","p":0.9}]
+      """
+    And step "build.fix.quality.fixing" writes ".gtd/QUALITY.md" with:
+      """
+      # Quality
 
       ## null check in parse()
 
-      - [ ] ./src/calc.ts#2-2 — where it applies
+      Still open.
       """
     When the process is driven
-    Then the process rests at "build.review.await-review"
-    And ".gtd/REVIEW.md" contains "## null check in parse()"
-    And ".gtd/REVIEW.md" contains "## subtract has no negative test"
-    And ".gtd/REVIEW.md" contains "check whether it was fixed"
-    And ".gtd/REVIEW.md" does not contain "Fixed automatically"
+    Then the process rests at "build.quality.gate"
+    And ".gtd/QUALITY.md" contains "## null check in parse()"
+    And ".gtd/QUALITY.md" contains "## subtract has no negative test"
+    And ".gtd/QUALITY.md" contains "check whether it was fixed"
 
   Scenario: prose the fixer leaves outside every chunk does not hide its fixes
     Given judge step "build.quality.triage" answers:
@@ -368,38 +507,18 @@ Feature: the quality lap's findings are triaged before the review
 
       All fixed.
       """
-    And step "build.health.check" changes nothing
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
-
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-      """
     When the process is driven
     Then the process rests at "build.review.await-review"
     And ".gtd/REVIEW.md" contains "- subtract has no negative test ("
     And ".gtd/REVIEW.md" does not contain "check whether it was fixed"
 
-  Scenario: signing off a review with no notes ends the process
+  Scenario: signing off the review ends the process
     Given judge step "build.quality.triage" answers:
       """
       [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.9},{"id":"Q3","answer":"noise","p":0.9}]
       """
+    And step "build.quality.gate" changes nothing
     And step "build.review.closing" runs its script
-    And step "build.review.reviewing" writes ".gtd/REVIEW.md" with:
-      """
-      # Review: abc1234
-
-      <!-- base: 0000000 -->
-
-      ## Subtract
-
-      - [ ] ./src/calc.ts#2-2 — subtract(a, b)
-      """
     When the process is driven
     Then the process rests at "build.review.await-review"
     When gtd lands
@@ -408,9 +527,42 @@ Feature: the quality lap's findings are triaged before the review
     And ".gtd/REVIEW.md" does not exist
     And the commit subjects end with:
       """
-      gtd(agent): build.review.reviewing → build.review.ranking
-      gtd(judge): build.review.ranking → build.review.carrying
+      gtd(agent): build.review.reviewing → build.review.carrying
       gtd(check): build.review.carrying → build.review.await-review
       gtd(human): build.review.await-review → build.review.closing
       gtd(check): build.review.closing → idle
       """
+
+  Scenario: a finding whose entry the human deleted is listed as dropped
+    Given judge step "build.quality.triage" answers:
+      """
+      [{"id":"Q1","answer":"noise","p":0.9},{"id":"Q2","answer":"noise","p":0.85},{"id":"Q3","answer":"design","p":0.8}]
+      """
+    When the process is driven
+    Then the process rests at "build.quality.gate"
+    Given a file ".gtd/QUALITY.md" with:
+      """
+      # Quality
+
+      # Dropped as noise
+
+      ## add could be a method
+
+      A class would read better.
+
+      - [ ] fix
+      """
+    When gtd lands
+    And the process is driven
+    Then the process rests at "build.review.await-review"
+    And ".gtd/REVIEW.md" contains:
+      """
+      Dropped at the finding gate:
+
+      - AC-1 ignores floating point (lens correctness)
+
+      Dropped as noise:
+
+      - add could be a method (p=0.85, lens correctness)
+      - subtract has no negative test (p=0.9, lens correctness)
+      """