CWI Store

In September 2026, our indie label submitted three tracks to a free playlist-pitching service. All three were accepted onto "New Rap Hits" (673 followers at the time), and a full 105-track scan confirmed them at #21, #30, and #31. Total spend: $0.

This kit is the complete field report from that campaign — not a guru course, not a guarantee:

• The submission method — how we picked the tracks, wrote the pitches, and used free services before spending a cent.
• The verification method — the full-scan checklist we use before we believe any placement. It's also how we caught two "we added you" claims we couldn't verify — and didn't announce.
• The outreach tracker — the exact log format we use to run the campaign, so every pitch, follow-up, and outcome is on record.
• What we said no to — pay-per-add offers, paid "opportunities" with no upside, and announcing placements before verification.

Who this is for: independent artists and small labels doing their own playlist outreach who want a real, checkable process instead of promises.

Who this is NOT for: anyone looking for guaranteed placements, stream counts, or revenue. This kit documents one campaign and one verified scan. Playlists change, curators change, results vary. There are no guarantees here — only a method you can verify yourself.

Curators promise. Screenshots lie. This report doesn't.

The Playlist Evidence Report is generated by CWI's Intelligence Engine from a full Spotify catalog scan of the playlists you claim. Every placement is confirmed by track-ID matching — not by a curator's DM, not by a screenshot.

Each report includes:

• Verified placements table (playlist, track, position, follower count, verification date)
• Full evidence log with sources and observation dates
• A "claims NOT verified" section — because what we couldn't confirm matters as much as what we could
• Methodology note you can hand to anyone

You send us your Spotify artist link and the playlists you believe you're on. Within 3 business days you get a PDF report. If we can't verify a single placement, you get a null-result report and a 50% refund — we'd rather tell you the truth than sell you a fantasy.

Built by Cumulative Web Inc, the indie label that grades every fact verified / claimed / pending / rejected before it reports a win.

Music supervisors don't license vibes. They license cleared, documented, verifiable tracks — and they move on fast when paperwork is missing.

This is the complete sync-readiness system from Cumulative Web Inc:

• 24 evidence-graded sync one-sheets (That Boy Hi Hat catalog) — every claim carries its evidence tier, source, and observation date
• The split-sheet template CWI requires signed before any license
• The ISRC/UPC audit worksheet for the full catalog
• CWI's sync pitch guide + evidence-tier legend

Use it two ways: (1) license That Boy Hi Hat tracks directly from us (hp@cumulativeweb.com), or (2) copy the system for your own catalog.

This pack documents readiness; it is not a license. Licensing is negotiated separately. BPM values are estimates, not audio analysis; identifiers and splits are marked NOT PROVIDED where they must come from distributor records.

Born from the reasoning method CWI runs on every project: twenty named perspectives — skeptic, economist, security reviewer, Black's chair, and sixteen more — each delivering a one-sentence verdict and one risk, then a synthesis with the dissent recorded.

You bring the models. The engine brings the structure: distinctness checks so no two minds parrot each other, mandatory dissent, per-mind model routing (run the skeptic on one model and the economist on another), and machine-readable reports in Markdown and JSON.

• CLI, local-first — Python 3, zero dependencies, no telemetry, no accounts
• Three backends: manual (paste verdicts, $0, no setup), local Ollama, or your own API key
• Kill-criterion tracking across runs — the protocol knows when to stop deliberating

Honest by design: the engine structures reasoning; it does not make models smarter. A report is only as good as the minds behind it.

Built by Cumulative Web Inc. Real software, never a simulation.

Agent ecosystems run on trust claims, and most trust scores are vibes. Verdict CLI is arithmetic: feed it evidence JSON — ERC-8004 identity signals, Needle Drop verified history, First Spin published verdicts — and get back a score with a published band and a full evidence trace, or a refusal when the evidence isn't there. It never invents a score.

The scoring core is the canonical CWI Verdict Engine v1.0.0, vendored verbatim and pinned — weights, thresholds, and gates are spec-versioned, never edited ad hoc. Sybil damping (issuer caps, identity clustering, self-assertion discount) is applied openly, and every output carries an input_sha256 so any score reproduces byte-identically.

• CLI, local-first — Python 3, zero dependencies, no network, no telemetry, no accounts
• Five commands: score, check (pipeline one-liner), example, spec, pin (audit the engine provenance)
• Machine-readable report.json + human report.md explaining exactly why the verdict is what it is

Honest by design: insufficient-data is the correct answer for new subjects, not a failure. There is deliberately no "scored" demo — a scored example would need fabricated evidence.

Built by Cumulative Web Inc. Real software, never a simulation.

What it is: a one-command scan over your installed skills — each skill directory checked against 25 versioned signature rules for prompt-injection text ("ignore previous instructions" and friends), exfiltration lines (pipe-to-shell, environment or key material sent to URLs), approval-disabling instructions ("never ask for confirmation", "disable guardrails"), and embedded secret material (AWS-key and API-key patterns, flagged as possible). Stdlib Python only: no dependencies, no network, no telemetry. You get a severity-triaged report with an explicit PASS/FAIL verdict, in human-readable text or machine-readable JSON for CI gates.

What it is not: it never executes skill code, runs no sandbox, and does no runtime analysis. Signature-based detection can miss novel attacks and can false-positive on benign text. A PASS means no known-bad pattern matched — it is signature triage, never a certificate of safety. We say that on every report because a trusted "clean" badge on a poisoned skill would be worse than no scan at all.

What it is: verification verdicts are snapshots, and snapshots go stale. Verdict Ledger records each verdict with sha256 snapshots of its evidence in an append-only local ledger, then re-checks that evidence on your schedule. The moment the evidence drifts or the TTL expires, it appends a superseding STALE envelope with the exact reason — evidence-drift:<source> or ttl-expired — and exits 1 for cron and CI gates. Unreachable evidence is an error, never a silent pass. Stdlib Python only: no dependencies, no telemetry, the only network traffic is the evidence fetches you point it at. Exit codes: 0 all current / 1 stale findings / 2 error. Default TTL is 30 days, matching the STALE rule CWI runs on its own Placement Wall.

What it is not: it never proves a verdict correct — CURRENT means the evidence has not moved since recording, nothing more, and every report says so. It is not tamper-proof: the ledger file is append-only by convention, not cryptography. The build-week receipts demonstrate two real stale catches (an evidence-drift catch over live HTTP and a TTL-expiry catch) plus one honest refusal, documented verbatim in the repo — check them before you buy.