Session Handoff — Open Credit-Gate Scoping — version 0.1

Version. 0.1 Date. 2026-05-30 For. A fresh Claude.ai session opening the pipeline-spend credit-gate scoping. From. The session that shipped source identification and scoped-ahead the credit gate. Operator. Marvin Percival.

Plain-language summary

You are opening the scoping for the pipeline-spend credit gate — step two of the Memory-completion path, the major solvency-weighted build. Step one (source identification) is shipped, merged, and recorded; you do not need it except as done-context. Everything you need to scope the credit gate has been verified against live code and captured in a constraints note. Your job this session is to turn that constraints note into a scoping note (the scoping-note-before-Change-Request discipline). There is one ground-truth verification to run first — it shapes the scope and must not be assumed. Read the orientation documents, run the verification, then scope.

First-read (do this before anything else)

In order:

The constraints note — loomworks-record/scoping-notes/loomworks-pipeline-spend-credit-gate-scoping-constraints-v0_2.md. This is your primary orientation. It carries: the governing principle (§0 — the credit system is a prepaid, hard-gated solvency control, not a metering feature; the deficit risk to DUNIN7; the invariant; the hard precondition on opening the pipeline), the corrected scope (three model-rooms, not four), the five corrected pieces, the two Operator requirements (credits-for-sale forward-compat; FORAY accountability), and the one convergent substrate gap (grant→consumption lineage + funded reservoir). Read §0 first and let it govern — this is a solvency control; the gate is primary and pre-spend; reconciliation is corrective.
The seed — highest-versioned loomworks-candidate-seed-vN_M in candidate-seeds/loomworks/ (was v0.9). Standing first-read.
The current-status manifest — highest-versioned current-status-manifest-vN_M in current-status/ (was v0.44; Entry 106 records source identification shipped). Standing first-read.

The two read-only inspections that produced the constraints note are referenced inside it; you do not need to re-run them. One inspection artifact sits untracked at loomworks-engine docs/inspections/credit-substrate-four-room-gate-inspection-v0_1.md if you want the full file:line detail — but the constraints note already distills it.

Live baseline (verified this session)

Engine main at daee809, alembic head 0074 (source identification merged; the development database is migrated to 0074; schema and main agree).
Operator Layer main at 1ce2542.
Record main at 722a436 (constraints note v0.2 filed in scoping-notes/).
Both repos carry tag source-identification-v0_1; the source-identification branches are merged and left in place as the record.
Re-verify these at session entry through Claude Code rather than trusting this handoff — state may have moved.

The mandatory first action — spender-threading verification

Before scoping the async-room metering, answer this open question (constraints note §5), because it shapes the scope and was unconfirmed in inspection:

> Is person_id threaded through the Shaping and Rendering async background jobs?

Credit balances are person-scoped (party_id = person_id). Manifestation is synchronous (the person is in the request). But Shaping and Rendering run as async background jobs (dispatch → run later), and inspection could not confirm the spender's identity reaches the running job. If it does not, threading it is a prerequisite piece, not a detail — there is no one to debit otherwise.

Run this as a read-only Claude Code inspection at session entry: trace the Shaping and Rendering job dispatch (the shaping_jobs / render_jobs enqueue and executor.py / agents/runner.py run path) and report whether person_id (or a resolvable equivalent) is carried into the running job. Report present / absent / partially. The answer changes the scope.

The session's job

Turn the constraints note into a scoping note (loomworks-record/scoping-notes/, versioned). The scoping note decides what the constraints note only framed:

The five pieces, sequenced (constraints note §1): (1) usage-returning LLM client; (2) system-key path for the rooms — prefer converging onto the canonical resolve_engagement_api_key resolver rather than three divergent inline ones; (3) spender-threading (informed by the first-action verification); (4) per-room consumption-flow writer; (5) gate + exhaustion per execution model (inline for synchronous Manifestation; dispatch-time / job-outcome for async Shaping and Rendering).
The grant→consumption lineage design (constraints note §3, Gap 1 — the crux). Both Operator requirements converge here: give consumption flows a grant linkage (which grant / lot a spend draws against — a FIFO or draw-down structure). Cheap to build into piece 4 now; expensive to retrofit. Decide the structure.
The funded-reservoir decision (Gap 2). The reservoir is currently an unbounded mirror. Decide whether the funded/accountable-reservoir model is built with this work or deferred — but decide it consciously, against the solvency invariant (§0): credits-for-sale needs a funded reservoir to be solvent.
Honour the constraints (constraints note §4): tier-identified flows (never collapsed), grant-linked, coherently-reservoired, FORAY-attestable-complete and seam-marked (the seam exists already, no-op, since Phase 63 — widen the payload to the full row shape at integration time). Reject every shortcut that collapses tiers, omits grant linkage, or records consumption without coherent reservoir accounting.

Keep the gate primary and pre-spend throughout — reconciliation is corrective, never the primary control. And keep the hard precondition in view: the pipeline is not opened to other Operators until the gate is real.

Working method (unchanged)

Claude.ai scopes, drafts, and produces the scoping note and (later) the Change Request. Claude Code executes read-only verification this session (the spender-threading check) and, in a later build session, the build. Claude Code halts at checkpoints; never pushes without explicit Operator authorization. The Operator communicates tersely and expects ground-truth verification before scoping decisions, not assumption. Documents are versioned name-vN_M.ext, Markdown-primary for technical consumers; the HTML-companion gap is tracked record-wide (do not hand-produce companions per-document — it waits for the render path). Full forms in the record (no abbreviations: Change Request, Operator Layer, Claude Code spelled out).

Tracked items (non-blocking, recorded — do not let them derail scoping)

Two pre-existing engine baseline failures on main (test_api_assertion_retract, test_phase_44_evaluator dedup) — unrelated, pinned, accepted. Their own errand.
The conftest sequence-reset gap (engagement_display_identifier_seq not reset by TRUNCATE … RESTART IDENTITY; needs manual ALTER SEQUENCE … RESTART for a clean engine baseline).
Stale model-tier asset map (loomworks_credit_opus → claude-opus-4-7, behind opus-4-8) — small separate follow-up.
Record-wide HTML-companion gap — waits for the render path.

What this session does NOT do

It does not build (no code, no migration). It does not write the Change Request yet — scoping note first, per discipline. It does not decide the funded-reservoir timing without surfacing it to the Operator. It produces a scoping note and the action path to the Change Request.

DUNIN7 — Done In Seven LLC — Miami, Florida Session Handoff — Open Credit-Gate Scoping — version 0.1 — 2026-05-30