Tiers
Six tiers that the build items group into, in dependency order. Substrate is Tier 1 — the foundation everything downstream builds on. Skills are substrate-aware from day one. Managed Agents and substrate-dependent Intelligence come after. Each tier delivers standalone value while setting up the next. Each item annotation flags what changes if 4K does the simplifications — items get marked SIMPLIFIED (work gets easier) or REMOVED (work disappears) when relevant. Click any item ID to jump to its full detail on the Capabilities page.
Tier 0 — Foundations
Setup and configuration work that has to happen before anything else. Some of this is invisible to 4K (Newfangled-side alignment posture); some is highly visible (workspace setup, INSTRUCTIONS files). This tier is mostly Configuration / Setup / Process — not tools-we-build.
- A working Claude Teams workspace with official MCPs installed for ClickUp, Notion, Slack, and Google Workspace
- An org INSTRUCTIONS file that grounds every conversation in 4K's operating philosophy, terminology, and tool-selection rules
- Per-engagement INSTRUCTIONS headers for the top ~10 priority engagements (rest can be added incrementally by 4K)
- Newfangled's documented NIST 800-53 Low alignment posture in parallel (not visible to 4K's team, but visible to their auditors as needed — see Compliance)
Every later tier depends on the INSTRUCTIONS file and the MCP installations. The team can't get value from any skill or agent until Claude Teams is grounded in 4K context. Without this tier, even the simplest skills produce generic-AI output.
Tier 1 — Substrate
The CKA tenant + modules + prompt tuning. Foundation for every tier below. Promoted from old Tier 3 because under the new framing it isn't a heavy build — it's modules and prompts on top of Frontier's abstracted CKA framework, which is built outside this project as a Newfangled-side investment.
Deploy 4K's tenant of the CKA platform. Build 4 data-source modules (ClickUp, Notion, dual-channel Slack, tagged GDrive) on the abstracted framework. Tune extraction, summary, synthesis, and strategic prompts for 4K's data and operation. Initial indexing + validation against real 4K queries.
- The CKA substrate as a working foundation, indexing ClickUp + Notion (engagement DB + client-success scoring DB + scorecards DB + partners DB + meeting notes + working agreements + knowledge bases + VTO space) + dual-channel Slack + tagged GDrive corpus
- 5–6 high-value synthesis dimensions at launch (per-engagement strategy summary, decision capture, sentiment markers, response patterns, engagement context for onboarding). The full 15 dimensions layer in over time.
- Per-engagement multi-tenant scoping across the initial ~50 engagements
- Substrate tools queryable by skills — the next tier's 12 skills can be substrate-aware from day one rather than shallow-now-deep-later
Substrate is what makes every downstream item produce full-depth output instead of shallow surface answers. Landing it early means the rest of the engagement isn't a "lightweight first, rebuild later" pattern — skills and intelligence items can query a working substrate from day one.
Expectation-setting: Tier 1 doesn't deliver direct daily user-facing value on its own. The substrate is invisible until Tier 2 skills query it. The quick-wins-in-parallel option (see Tier 2) addresses this by shipping ~7 substrate-independent skills during Tier 1 to give 4K's team value while substrate work happens.
The CKA's core abstraction work — multi-tenant infrastructure, abstracted data-source ingestion, modular data-source framework, prompt-override system — is Newfangled-side investment, built outside the 4K project. That's why this tier is bounded scope rather than the largest single piece of engineering. See #28 and the Pricing page for the full framing.
Simplification 02 (canonical engagement IDs) SIMPLIFIES #28 — per-engagement scoping doesn't require us to reverse-engineer it from naming drift.
Simplification 04 (Drive folder cleanup) SIMPLIFIES #28 — tagged-subset ingestion gets clean inputs rather than fuzzy matches.
Simplification 05 (Notion DB linkage) SIMPLIFIES #28 — ingestion gets explicit cross-DB relationships rather than us inferring them.
Simplification 06 (resolve confidentiality / Zoom policy) gates how deep #28's synthesis can go on shared-Slack-channel content.
Tier 2 — Skills
12 SOP skills plus 2 MCP connectors. Substrate-aware from day one — every skill can call into Tier 1 substrate tools where the workflow benefits. This is where 4K's team starts feeling the engagement's value at full depth, not a lightweight first cut.
Two MCP connectors deploy alongside the skills so the workflow integrations have programmatic access. Both reuse existing Newfangled assets — #16 is a redeployment from the WWP project; #17 is an internal refactor behind a pluggable adapter protocol so the eventual Harvest swap is config, not rewrite.
- ~12 skills in active use across 4K's team, each one replacing a specific manual workflow named in the audit
- 4K Pipedrive accessible via MCP — first time the team has programmatic Pipedrive access
- Harvest pluggable layer deployed and ready for the eventual replacement swap
- Substrate-backed retrieval where it adds value — rate-card lookup, SOW Q&A, win/loss capture, Roundtable agenda generation all benefit from substrate context
Roughly 7 of the 12 skills don't need the substrate to be valuable: #02 sentiment scoring, #04 VTO, #08 SOW Q&A, #09 Joanna's aggregator, #10 Shanice's burndown, #11 brand voice writer, #14 late-invoice review. If 4K wants daily user-facing value during Tier 1 substrate work, these 7 can ship in parallel. The other 5 substrate-dependent skills ship later at full depth. Doesn't change total scope or price — just sequencing.
Simplification 03 (documenting implicit rules) directly improves #01, #02, #07, and #09 — each one's calibration target gets more concrete. Simplification 04 (Drive folder cleanup) directly improves #05 and #08 — Drive search becomes reliable rather than fuzzy.
Simplification 07 (engineer one-ticket-per-Harvest-entry SOP) SIMPLIFIES #09 meaningfully — the multi-ticket disambiguation rule becomes unnecessary.
Tier 3 — Automations
The Managed Agents that automate weekly cycles. This is where Beto's, Joanna's, and Peke/Jade's weekly ETL pain goes away. Each agent consumes skills + connectors that came online in earlier tiers and substrate context where it matters; this tier wires them into scheduled, autonomous runs.
Two supporting builds: #21 (watched-folder for Forecast CSVs, needed by #24) and #25 (custom ClickUp connector, recommended proactively to avoid official-MCP rate limits constraining #23).
- Beto's scorecard ETL automated — the "Updating Fucking Scorecards" pain goes away. Pre-L10 cadence; mid-week double-update becomes a single automated sync.
- AR monitoring goes always-on — Mon/Thu manual cycle replaced by daily automation that drafts emails for admin review.
- Client reports auto-generated per-PM templates preserved; multi-ticket disambiguation handled programmatically.
- Pipeline upstream consolidation runs weekly — Beto's manual transfer of PM projections eliminated; Finance HQ discovery sprint produces Phase 2 sizing artifact.
- Weekly briefings delivered Monday mornings — leadership gets pipeline + activity summary without manual review.
These items all use the frontier-agents Lambda platform (which already runs in production for Newfangled and for WWP). They consume Skills + Connectors from Tier 2 and substrate context from Tier 1 where useful (e.g., the briefing agent can reference substrate-extracted health signals). This is where 4K's most-named operational pains (Beto's scorecards, Peke/Jade's AR, Joanna's reports) get retired.
Simplification 01 (unified PSA when replacing Harvest) has the biggest effect here:
- #21 REMOVED — if 4K's Harvest replacement is a unified PSA with documented API, Forecast goes away entirely and the watched-folder pipeline is unnecessary
- #24 SIMPLIFIED — most of Beto's master-sheet reconciliation formulas exist because data lives in too many places; unified PSA collapses much of that
- #22 SIMPLIFIED — sales-to-delivery handoff becomes one record's stage change, not a manual transfer
Simplification 02 (canonical engagement IDs) SIMPLIFIES #24's reconciliation work and removes the name-format-drift tax from every cross-system lookup.
Simplification 07 (PM projection-update SOP) addresses the root cause of #22's pain — automation catches drift, but only behavioral change fixes the source.
Tier 4 — Intelligence
The substrate-dependent intelligence items that compose on top of the working substrate + skill layer. Previously bundled with substrate as the "scary big finale"; now separated and downgraded because under the new framing they're smaller items that consume Tier 1 rather than build it.
Both depend on the substrate's synthesis layer. #26 composes substrate-extracted sentiment, decision logs, risk signals, and engagement-age signals into a per-account health view with hybrid alert logic. #27 reimplements SiBorg as a CKA-querying estimation skill with comparable retrieval and Harvest-actuals feedback — Si's "all-logo data wish" becomes reality.
- SiBorg reimplemented against the substrate. Comparable proposal retrieval, rate-card lookup, win/loss memory migration with parallel-running validation against Si's current texture.
- Client Health Agent live with the "rules handle WHEN, AI handles WHAT" hybrid alert model, multi-channel delivery (email + Slack, configurable per account), leadership-only access
- Specific alert triggers: sentiment drop, recurring late deliverables, engagement-age threshold crossed, yellow→red state transition
- Two integration surfaces for CHA: Projects Roundtable agenda (composed view lands weekly) + pipeline conversation (risk status surfaces alongside pipeline weighting where it currently doesn't appear)
These items depend on the substrate being mature enough for high-quality retrieval (for #27) and high-quality synthesis (for #26). Sequencing them after substrate + skills + automations means: substrate has been queried by 12 skills + 5 agents long enough to surface quality issues, and threshold-tuning for #26's deterministic alert rules can be calibrated against real engagement data rather than a cold start.
Both items were previously rated HARD; now MODERATE. The substrate they depend on is Tier 1 foundation, not a separate net-new build. See #27 and #26 for the downgrade reasoning.
Simplification 06 (resolve confidentiality / Zoom policy) gates how deep #26's synthesis can go on shared-Slack-channel content. Until resolved, CHA reads 4K-side signals only.
Simplification 02 (canonical engagement IDs) makes both items' cross-engagement comparable retrieval cleaner — Si gets richer comparables, CHA gets reliable per-account scoping.
Tier 5 — Extensions
Items gated on 4K decisions, vendor surfaces, or policy interpretations. None of these block the earlier tiers; each unlocks when its specific blocker resolves. This tier is most accurately described as "what we'd add once these unblock," not "what comes after Tier 4 sequentially." Treated as per-item add-on in Pricing.
- #29 Finance HQ Phase 2 — gated on the discovery sprint output from #22 (Tier 3). Unsizeable until we know Finance HQ's surface.
- #30 Harvest replacement integration — gated on 4K's tool decision (Productive, Runn, Scoro, etc.). Per Simplification 01, this decision should pick a unified PSA.
- #31 Zoom transcript pipeline — gated on 4K's 24-hour retention policy interpretation (do transcripts get the same treatment as recordings, or different?). Per Simplification 06.
- #32 Shared-Slack-channel deep synthesis — gated on 4K's MSA confidentiality review. Per Simplification 06.
- Finance HQ integration — discrepancy investigation surface; sales-to-finance flow fully automated; per-service-line profitability becomes visible
- Harvest replacement — clean swap to the chosen PSA via the pluggable adapter pattern set up in Tier 2
- Zoom transcript pipeline — extends CKA with sentiment-from-voice; enables deeper Client Health Agent dimensions
- Shared-channel deep synthesis — full sentiment / decision-capture on client communications in shared Slack channels (today: only 4K-side signals)
All four items in this tier are directly unblocked by Simplification 06 (resolve policy questions) and Simplification 01 (Harvest replacement decision). If 4K acts on those simplifications, this tier moves from "blocked" to "active scope."
Tier-by-tier composition
Quick scannable overview of what each tier contains and what shapes the work.
| Tier | Items | Tool types involved | Difficulty mix |
|---|---|---|---|
| Tier 0 — Foundations | 4 items: #15, #12, #13, #20 | Config / Setup / Process (Newfangled-internal NIST work parallel) | 3× Easy + 1× Moderate |
| Tier 1 — Substrate | 1 item: #28 | Custom App (CKA tenant + modules + prompts on Frontier-side abstracted framework) | 1× Moderate (downgraded from Hard) |
| Tier 2 — Skills | 14 items: #01, #02, #03, #04, #05, #06, #07, #08, #09, #10, #11, #14, #16, #17 | Skills + Custom MCP Connectors (substrate-aware from day one) | All Easy |
| Tier 3 — Automations | 7 items: #18, #19, #21, #22, #23, #24, #25 | Managed Agents + Custom App (small) + Custom MCP Connector | 2× Easy + 5× Moderate |
| Tier 4 — Intelligence | 2 items: #27, #26 | Skill + Managed Agent (+ small leadership-only UI for #26) | 2× Moderate (both downgraded from Hard) |
| Tier 5 — Extensions | 4 items: #29, #30, #31, #32 | TBD (per item) — gated | All Blocked until 4K decisions resolve |
Decisions still to make
Several open questions affect what ends up in which tier, or whether some items ship at all. Resolution before commercial conversations.
~7 of Tier 2's skills don't need substrate to be valuable. Two approaches: (a) strict sequence — Tier 1 substrate fully ships before any Tier 2 skill starts, (b) parallel — quick-wins ship during Tier 1 so 4K's team sees user-facing value while substrate is being built. (b) is the recommended default; explicit decision.
#14 (human-assisted late-invoice review) is a stopgap before #19 (full AR automation). Worth deciding upfront whether to ship both or skip the stopgap. If we skip, #14 leaves Tier 2 and we lean on #19 in Tier 3.
Notion touches 8 items in scope. Same logic as #25 for ClickUp — official MCP works but our custom wrapper gives control, RunReport, read-only enforcement, response normalization. Currently not in any tier — flagged as decision point. If yes, slots into Tier 2 alongside #16/#17.
Tier 3 has 7 items. Some sequence naturally (#21 before #24 because #24 consumes #21's output). Others are independent (#18, #19, #22). Worth deciding whether Tier 3 ships as one bundle or as sub-phases.
Tier 5 items are gated on external resolution. Two approaches: (a) quote them alongside Tiers 0–4 with "if unblocked by X date" conditions, (b) leave out of initial scope and quote separately when blockers resolve. (b) is cleaner and matches the per-item add-on framing in Pricing; (a) gives 4K visibility into total potential scope.
Whether tiers correspond to commercial milestones (4K pays per tier delivered), or whether the engagement is priced as a whole with tier sequencing internal. Separate question from technical sequencing. Worth a conversation before quoting.