LL2 · LLL review board

Get Leads: repair → consolidate → grow

SKETCH · REVIEW ONLY

Review this short sketch and answer ten choices before any executor launches.

Execution: not launchedRun: get-leads-ll2-chain-20261006Updated: 2026-10-06T18:23:52.305435+00:00

Current path to the deliverables

CURRENT STATE

Current state

A working devbox pipeline and lead table already exist. Seven campaign backups and a greeting/domain handoff are prepared. No new executor in this chain has launched.

INPUTS

Available inputs

  • Existing system · confirmed — Keep get-leads2.cfd-staging.com and the proven devbox scripts, paid caches and review edits.
  • Campaign handoff · confirmed — Dated 6 October audit: 7 campaigns; 1,790 no-reply / fewer-than-three-sent records selected. Refresh before any write.
  • Mac-independent handoff · confirmed — Original prompt, five session extracts with gists, skills and historical evidence are copied to devbox. Initial bundle:784 files /41.2MB, all hashes matched. Final decisions will be added before launch.
  • Devbox capacity · confirmed — 6 October18:13UTC: about5GB disk and1.1GB RAM available; swap full. Existing producer active. Tune/verify capacity before eight workers; no new producer launched.
  • Ten decisions · open — Set transition authority, campaign handling, lookup effort, inventory scope, target baseline, budget and review cadence.
DESIRED STATE

Desired state

One existing web app explains the inventory, historical runs, running jobs, methods, spend and skills. Three sequential devbox executors repair campaigns, consolidate the workflow and generate the agreed additional 2,000 leads for review.

OBSERVABLE DELIVERABLES

Deliverables

  • Sol 6.1 xhigh: improve greetings and domain articles while preserving sending history.
  • Astra High: update unused inventory and tidy the existing web app, scripts and skills.
  • Astra High: launch at least eight parallel processes, then assess once after two hours.
Deliverable details

1 · Repair the selected campaign leads

Current problem: The selected campaign records need better names; six campaign copies also reference a domain field missing from their payloads.

Proposed change: Use the prepared audit, enrich supported names, populate the required article/domain field and verify all seven sequences. Final deliverable: prepare task 2 with LL2 and hand it to a fresh Astra High executor under the agreed transition rule.

Visible at: Existing Instantly campaigns, separate campaign tables and one before/after receipt.

2 · Make the proven workflow easy to reuse

Current problem: Working methods and historical results are spread across runs and handoffs.

Proposed change: Read the exact original request; simplify this board; apply the proven enrichment to unused leads. Show counts, methods, running jobs, costs, exact skill definitions and when to use them in the existing app. Final deliverable: prepare task 3 with LL2 and hand it to a fresh Astra High executor.

Visible at: The existing lead web app and a compact canonical skill/document set.

3 · Deliver 2,000 additional import-ready leads

Current problem: The new lead target needs a defined baseline and a reliable autonomous production run.

Proposed change: Read the original request; simplify this board; reuse proven source and personalization scripts with shared deduplication and a budget ledger. Assess actual yield after two hours and refine if needed. Produce leads for final review, not automatic import.

Visible at: One run/filter in the existing table, matching CSV and source/cost breakdown.

Optional operational questions

Annotate a choice only if it changes the outcome or priority. No option is preselected.

1. After your initial Launch, how should the three stages advance?

Recommended: A. Automatic handoffs keep the chain working while you are away.

  • A — Automatically launch the next named executor after the previous stage passes its checks.
  • B — Pause after campaign repair, then run stages 2 and 3 together after my approval.
  • C — Ask me before each new executor launches.

2. How should the selected campaign records be updated?

Recommended: A. Keeping the same provider IDs is the strongest available route to preserve sending progress.

  • A — Update existing lead records in place.
  • B — Reimport only never-contacted leads; update partially contacted leads in place.
  • C — Require removal and reimport, but proceed only if exact progress restoration is proven.

3. What should happen to active campaigns during the repair?

Recommended: A. A brief pause prevents emails leaving while their variables are being repaired.

  • A — Temporarily pause the affected active campaigns, then restore their prior status after verification.
  • B — Keep them active and update each passing row as soon as it is ready.
  • C — Pause affected active campaigns and leave them paused for my review.

4. Six campaigns reference a domain field missing from 1,839 records. How broadly should that field be repaired?

Recommended: A. This fixes the demonstrated field gap without widening the name-enrichment cohort.

  • A — Fill the domain field wherever missing in those six campaigns; enrich names only for the no-reply / under-three-sent cohort.
  • B — Change only the selected no-reply / under-three-sent cohort.
  • C — Fix never-contacted records first; leave other records for a separate pass.

5. How much effort should go into reaching roughly 70% named greetings?

Recommended: A. Reuse and short ordered lookups offer the best starting balance; report real coverage.

  • A — Saved site evidence first, then a bounded public-directory lookup chain; use the generic greeting when unresolved.
  • B — Keep researching unresolved companies toward 70%, within the shared spending cap.
  • C — Use saved evidence only for the first pass; show unresolved names as a separate later opportunity.

6. What should change across all unused database leads?

Recommended: A. You said the current personalizations were good; preserve that work and your edits.

  • A — Apply the new name/greeting/domain methods; preserve existing P1 unless a concrete error is demonstrated.
  • B — Also regenerate unreviewed P1 values that fail the refined checks.
  • C — Regenerate all machine-written P1 values, but always preserve my manual edits.

7. What does the third stage’s 2,000 new leads mean?

Recommended: A. This preserves your earlier confirmed target without discarding useful work already paid for.

  • A — Complete 2,000 available leads beyond the 666 imports; current eligible unimported rows in that run count.
  • B — Produce 2,000 more beyond every lead already in the database at the new launch.
  • C — Produce 2,000 newly sourced companies; do not count the existing unused reservoir.

8. What spending ceiling should the three-stage chain use?

Recommended: B. A fresh shared allowance covers name research, verification and acquisition; past spend remains visible.

  • A — Use only the unspent balance of the existing $100 run cap.
  • B — Allow up to $100 of new spending across all three stages, separate from recorded past spend.
  • C — Allow up to $150 of new spending across all three stages.
  • D — Use a different total that I specify.

9. After the required two-hour check, how much further agent supervision do you want?

Recommended: A. It gives the scripts ownership of routine processing and limits model overhead.

  • A — One assessment and refinement, then scripts continue without scheduled AI checks; report completion or a terminal failure.
  • B — Continue two-hour assessments until the target or spending cap is reached.
  • C — Allow one more assessment the following morning, then stop scheduled AI checks.

10. How should the eight parallel workers balance proven sources and experiments?

Recommended: A. Previous results give us useful starting sources; more parallelism alone is not proof of more output.

  • A — Concentrate on proven sources, adding a new source only when usable supply or yield falls short.
  • B — Keep six workers on proven sources and two on bounded new-source trials.
  • C — Start eight distinct source trials, then concentrate on the best at the two-hour check.

Choose the next request

Request a revised sketch at the same detail level: annotate changes, and I will update this same board.
Request the full contract: ask me to expand this same board with criteria, sourced constraints, and evidence expectations.

Neither request starts execution. A later explicit Launch is still required.

Preview (illustrative, not test proof)

Illustrative trigger → action → outcome · not test proof

TRIGGER

Settle 10 choices

No executor launched

ACTION

Repair campaign leads

Sol 6.1 xhigh

ACTION

Tidy the proven system

Fresh Astra High

OUTCOME

2,000 leads for review

Fresh Astra High + scripts