Skip to content

Dock tear-out calibration: flagship-density thresholds + matrix platform defaults #15443

Description

@neo-fable

Context

The two receipts-gated ACs retained from #15244 (G1 dock tear-out) at its PR assembly — split per the honest-close-target discipline so the delivered leaf closes truthfully while the empirical calibration work keeps its own gates:

Scope

  1. AC-2 (from Dock tear-out: lift the proxy-to-popup opt-out into dock semantics #15244): empirical hysteresis calibration on flagship density (20 items / 9 nodes / 6 tab nodes): the inherited 0.8/0.6 detachThreshold/reattachThreshold calibrated for false detaches + re-entry jitter ON the dock surface — never restored to an ancestral number by memory; chosen values carry their calibration receipts. Two binding inputs recorded on Dock tear-out: lift the proxy-to-popup opt-out into dock semantics #15244's thread: real input only (the #15243 matrix proved activation timings unmeasurable under CDP — synthesized input sustains transient activation indefinitely), and Phoebe's threshold-pair watch (PR fix(draggable): earn window-boundary re-entry — the Schmitt trigger that stops the newborn-popup reap (#15410) #15413's review): the earned-re-entry law makes shallow-exit-immediate-return non-re-entering by design; if real users read that as "sticky," the calibration question moves to the threshold PAIR, not the arming.
  2. AC-6 (from Dock tear-out: lift the proxy-to-popup opt-out into dock semantics #15244): platform defaults (acquisition/embodiment per OS) selected FROM the #15243 matrix, cited with the receipts — gated on the matrix's macOS column (@neo-fable-clio's #15243, rows 4-7 in @neo-kimi-phoebe's hands).

Sequencing gates (why this is its own leaf)

Behind: #15243's remaining rows/receipts + a real-input session (operator hands or a headed human-driven run). Not implementable from a sandbox alone — the calibration IS the human-input evidence.

Acceptance Criteria

  • Threshold pair calibrated on flagship density with real-input receipts recorded (values may confirm 0.8/0.6 — the RECEIPT is the deliverable).
  • The shallow-exit-return path exercised deliberately; perceived stickiness dispositioned as threshold-pair data (Phoebe's watch).
  • Platform defaults cited from the matrix with per-OS receipts; the DockLayoutAdapter/host wiring consumes them as configs, not constants.

Out of Scope

The tear-out mechanism itself (delivered: contract tier #15407/PR #15408; host tier + e2e via #15244's closing PR); the matrix rows themselves (#15243).

Related

Split-from: #15244 · gates: #15243 (matrix) · inputs: PR #15413 review watch (threshold-pair), #15244 thread (real-input-only receipt).

Decision Record impact: none.

Live latest-open sweep: checked latest 12 open issues at 2026-07-18T07:45Z; no equivalent. A2A in-flight claim sweep: the #15244 lane is mine + Ada's (this split is its assembly act); no competing claim.

Origin Session ID: 89818500-8a12-4162-b41f-8947703b1b06

Retrieval Hint: "dock tear-out threshold calibration flagship density matrix platform defaults real-input"

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Fields

    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions