Unlock Content Brain · One audit, one roadmap

Content Brain — Audit & Roadmap

A single view of where the brain is vs where it should be, across two levels that turn out to be the same problem: the content + belief model that lives in it, and the schema + architecture that holds it. Built from a live read of the vault (Projects/Unlock-Content-Brain/), the /brain MCP, doc 544 (Belief Map), the 138-file index matrix, and the V2 design proposal. Ends in one sequenced roadmap.

For Tom + Werner Operating model Tom curates the vault · Werner implements the DB Date 5 June 2026
Decisions + notes save in this browser automatically.

The brain is not a failed build, it is a partial one. The atomic content substrate migrated cleanly and two of the hardest schema moves landed. What is missing clusters into one root cause: the belief model that drives everything was flattened in migration, and the content + schema that depend on it never followed.

~190
Entities migrated
2
Whole layers missing
3
Verify w/ Werner
8
Roadmap steps

The one issue underneath both audits

Doc 544 is your real belief model, a signal map for diagnosing a live investor. The database replaced it with a generic 5-stage awareness funnel (BS1–BS5). That single flattening explains the content-level gaps (no belief-signal system, no routing table, no closing assets) and the schema-level gaps (the belief_stages table is the marketing funnel, not your model). Fix the model, and most of the rest sequences from it.

Level 1 — Content & model

  • Belief-signal operating system absent (544–551)
  • Conversion / closing assets (100-series) absent
  • C4 compliance gate has no home

Level 2 — Schema & architecture

  • belief_stages = funnel, not the 544 signals
  • lifecycle, backing relationships, quality surfacing pending
  • naming + ownership uneven

How the two layers relate (so nobody "fixes" the wrong thing)

The vault and the DB hold different fields on purpose: you curate in the vault, Werner implements in the DB, intelligence flows one way. So differences between the layers are the handoff in flight, not drift. This document is that handoff, packaged. The Roadmap tab turns it into sequenced work with owners.

Tabs: Belief Drift is the root cause. What's Built credits what works. Content Gaps and Schema Gaps are the two levels. Roadmap is the single plan. Decision pills on every gap + step persist in this browser; Export pulls them into a session summary.

The belief model was flattened

When you say "beliefs" you mean doc 544. The database stored BS1–BS5 instead. Different ontologies, and the richer one, the one that lets a transcript be scored and a call be routed, is the one that did not make it in as structured data.

544 — your model (left behind)

~18 belief signals + states + routing + gates
  • Qualification: QT (tax rate), QL (liquidity)
  • Core C1–C4: problem real / unsolved / EIS understood / risk accepted
  • Clusters: G1–G3 Growth Seeker · L1–L2 Legacy Builder · P2–P3 Preserver
  • Situational S1–S6: credible / considering / valuation / team / terms / timing
  • States: Green / Amber / Grey / Red / N/A
  • Gates: C4 compliance gate · Pack 1 gate (S2 + C4 + score ≥70)
  • Routing: belief → exact content asset
→

Brain — what's there now

5-stage awareness funnel
  • BS1 Unaware → BS2 Problem → BS3 Solution → BS4 Committed → BS5 Advocate
  • One linear marketing axis
  • No states, no gates, no per-belief routing
  • Cannot answer "what does this investor believe now"
  • Cannot enforce the C4 compliance gate

Continuity worth noting: the G / L / P clusters did survive, as the persona E-axis (E1 Preserver, E2 Growth Seeker, E3 Legacy Builder, E4 Undetermined) across the 10×4 persona matrix. So the persona half of 544 is intact; the belief-signal half is missing. And every rendering was re-tagged to BS1–BS3 only, so the per-asset belief code (330=C2, 380=G1, 490=S6, 430=S1) was dropped: 0 of 19 renderings carry a 544 belief code.

Migrated well

What came across cleanly

Credit where due, the foundations are solid at both levels.

Content substrate

atoms · renderings · personas · pillars · channels · recipes · compliance
▶
Present
01_acus 24 · 02_renderings 19 (330–550, original IDs kept) 03_personas 51 (P1–P10 × E1–E4 + bases, G/L/P preserved as E-axis) 04_pillars 8 · 05_channels 9 · 08_recipes 28 · 07_compliance 8
Verdict
The atomic layer migrated faithfully. The gap is the operating system on top, not the substrate.

Schema move: surface + moment routing

V2 Move 2 (the "High risk" one) — live in the DB
▶
Evidence
rendering rows carry surface / primary_moment / secondary_moments / rendering_matrix_id tools: query_renderings_by_surface / _by_moment / list_surfaces / list_moments channel retained for backward-compat
Verdict
The channel→surface+moment split is implemented at the query layer. (File-side caveat in Schema Gaps.)

Schema move: campaigns first-class

CAMP-001 + 11_campaigns folder
▶
Evidence
CAMP-001 "EIS October 2026": funnel_rules · compliance_overrides · metrics · owner + collaborators all 12 numbered folders exist + populated (09–12 included)
Verdict
Fully structured, present in both layers, arguably ahead of the V2 doc which still frames it as a proposal.

Quality already captured in files

craft_result + uat_result on 19/20 renderings
▶
Evidence
02_renderings: craft_result: PASS (19/20) · uat_result: PASS (19/20) not surfaced through the MCP, not named "quality"
Verdict
The "quality layer" the V2 doc proposes already exists as data. The work is naming + surfacing, not authoring (see Schema Gaps / Roadmap).
Content & model gaps

Whole layers the brain was built without

Referenced by docs that did migrate, but the things themselves are absent. Not naming-drift artefacts.

Critical

The belief-signal operating system (544–551, 235, 530)

The instrument that makes signals work
▶
Missing from the brain
545 Belief Map Framework V4 — definitions + routing philosophy + gap register 546 Signal System Spec — Werner's data model + gate logic (the bridge doc) 547 Question Sequence (25 Qs) — each mapped to a belief signal 548 Demo Run Order · 549 Three-Call System · 551 Post-Call Protocol 235 Belief-Map Email Templates · 530 Agent Quick Reference
Why it matters
This is exactly what couldn't be answered earlier, "what beliefs must an investor hold, scored from a transcript." The definitions, the question→signal map, the states and the routing table are not in the brain. 546 is the natural Tom→Werner bridge: it is the build spec for the data model.
Critical

The C4 compliance gate (doc 140 Access Explainer)

A non-negotiable gate with no home in the brain
▶
What 544 says
"C4 compliance gate: If C4 is not green, send 140 — Access Explainer ONLY. Nothing else. No Pack 1. No valuation content. No founding round terms."
In the brain
Doc 140 is not a rendering, and the gate is not a constant in 07_compliance. The brain has FCA-perimeter + capital-at-risk constants, but not the behavioural gate governing what may be sent before risk is understood.
High

The investor-pack / conversion layer (100–199)

Only top-of-funnel content exists; closing assets absent
▶
Missing renderings
100 One-pager · 110 Three-pager · 120 Pack 1 · 130 Pack 2 IIM 150 Secret Weapon · 155 EIS Guide · 160 Case Studies 170 IHT Planning · 180 Fee Comparison · 181 Stress Test · 182 BPR Explainer confirmed: NONE of the 100-series present in 02_renderings
The practical break
rendering belief_stage spread: BS1 ×6 · BS2 ×6 · BS3 ×7 · BS4 ×0 · BS5 ×0
Why it matters
544's routing sends 140/150/120/160/170/180/181/182 at specific belief states, and the brain can serve none of them. Everything sits in the top three stages: an awareness engine with no closing kit.
Medium

Behavioural + sequence pieces (450, 540, 470)

Signal-bearing assets dropped from the renderings set
▶
In matrix, not in brain
450 Social Proof E03 — S1 credibility (testimonial) 540 Clicked-Not-Booked — the behavioural re-engagement branch 470 Voicemail Drop VM1 · 440 Rate Response template brain kept 460 + 550 from the same sequence but dropped 450/540
Why it matters
450 carries the S1 signal; 540 is the behavioural branch the email-sequence logic depends on. Partial sequence migration leaves holes mid-arc.
Correctly excluded

Not gaps — these belong elsewhere

Governance / CRM / roadmap / build-spec docs (~55 files)

OS-vault material, not content-brain material
▶
Rightly absent
000–012 governance · 200/210 Pipedrive+Aircall · 600/601 team specs 950–996 roadmap / CRM maps · 964/965 migration · ERD + DB specs
Boundary calls (yes/no each)
701 Compliance Language V4 (full register vs the brain's 8 constants), 520 19-Persona Guide (vs the 10×4 matrix), 655 Glossary (→ GLO Tier-B class). Not assumed.
Schema & architecture gaps

From the V2 proposal, measured against the live schema

These are the structural moves the V2 doc proposes that have not yet landed. Several are decisions, not missing data.

Critical

belief_stages holds the funnel, not the 544 signals

The schema face of the headline drift
▶
State
06_belief_stages: BS1_Unaware … BS5_Advocate (5 marketing stages) no table for C1–C4 / G·L·P / S1–S6 / states / gates
Decision
Add a belief_signals entity type (code, definition, state enum, persona-cluster map) and decide whether BS1–BS5 is retired or kept as a coarse overlay. Highest-leverage schema change, it re-enables transcript scoring.
Decision

lifecycle_stage absent

V2 Move 4 — confirmed not built (files + DB)
▶
State
entities carry status (LOCKED/APPROVED/…) + is_current/superseded_by lifecycle_stage: found only inside the V2 proposal HTML, 0 content rows
Decision
Separate deployment-state from immutability (a LOCKED ACU can still be retired). Doc rates it low-risk + reversible + recommended → likely yes. Joint LIF-series sign-off.
Decision

backed_by relationships not populated

V2 Move 3 — registry built, links empty
▶
State
Tier B registry: 7 classes live (DEC/RES/REG/CSE/GLO/AUD/PER) + backing_role backed_by present on: 10_methodology + _review docs only · 0 atoms, 0 renderings
Decision
How to populate: manual per entity, or generate from citations already inside atom source strings? And the minimum set that makes impact-analysis useful. Joint BCK-series.
Decision

Quality fields not surfaced / named

Data exists in files, not in the DB or under a "quality" label
▶
State
craft_result + uat_result on 19/20 rendering files · not in any MCP tool
Decision
Promote craft_result/uat_result into a named quality layer + surface through the MCP, and decide what else quality should capture (confidence? freshness?). Close to free, the data is already there.
Decision

Naming: class_code + ID collisions

V2 Naming — partial, and an ID clash to resolve
▶
State
applied: campaigns (CAM), Tier B id_patterns · not applied: renderings (330_…), atoms (acu_…) collision: 530 means 3 things (Cold Call CS1 / Agent Quick Ref / Email ColdE1A); 400 = content piece AND tech spec
Decision
Opt-in for new only (doc's stated default) vs backfill class_code on the legacy estate. Either way, run an ID-collision pass before any re-import or the routing table points at the wrong asset.
Verify with Werner

Cannot be inferred from data

Is the parser reading surface + moment in production?

Code-state, not data
▶
Ask Werner
Fields exist in the DB; whether the live parser routes on them or still falls back to channel is a code question. Partially reversible once switched.

Q1–Q7 gate status

V2 cannot lock until all 7 answered
▶
Ask Werner
Which of the seven open questions from his 29 May review are now answered, and which still block the next migration phase?

Surface Werner's routing back into the vault?

So Tom can sanity-check it against real usage
▶
Ask Werner
Can the DB's surface/moment classification be shown read-only in the vault, so Tom can check it against how each piece is actually used and feed corrections back?

One sequenced plan covering both levels. Phases run in order; steps inside a phase can run together. Owner + source tags on each.

Update — 5 June: categorisation settled, files formatted

The 32-file re-import package is now reformatted to the brain's canonical frontmatter and routed: 20 → 02_renderings (with resolves_signal tags restored), 544/545/546/547 → a new 13_belief_signals folder (545 canonical), 701 → compliance, 520 → personas, 960 → strategy, 655 → Tier-B GLO in Resources, and 530/548/549/551 split out to the OS vault as sales SOPs (not brain). Package staged in _review/ as PENDING. This pre-completes step 5 below and resolves the step-3 folder question.

Phase 0

Decide the model

gates everything below
1

Decide belief-model authority Tom + Werner

Reinstate 544's signal map (C1–C4 / G·L·P / S1–S6 + states + gates) as the brain's belief layer; retire or demote BS1–BS5 to a coarse overlay. Log a one-line decision to Intelligence/decisions/. Blocks steps 2–7.

2

Confirm the handoff model + Q1–Q7 status Tom + Werner

Lock that Tom curates the vault and Werner implements the DB (no reconciliation needed), and clear which of the V2 Q1–Q7 still gate work.

Phase 1

Build the schema

Werner, from doc 546
3

Stand up the belief-signal schema Werner 546

Build belief_signals (code, definition, state enum Green/Amber/Grey/Red/NA, persona-cluster map) + a gates rule for C4 and Pack 1. This is the bridge: 546 is its spec. Source docs now placed in 13_belief_signals/ (545 canonical, 547 the question instrument) ready to model.

4

Add the low-risk schema fields Werner mostly free

Surface existing craft_result/uat_result as a named quality layer; add lifecycle_stage (5 states); decide + begin backed_by population from existing citations.

Phase 2

Restore the routing

Tom, low effort
5

Re-tag the renderings with resolves_signal Tom done in package

Done — the 20 reformatted renderings in the staged package carry resolves_signal (140=C4, 180=G1, 181=P2, 182=L2, etc.), rebuilding 544's one-belief→one-asset routing. Confirmed on import.

6

ID-collision pass before any re-import Werner do before Phase 3

Resolve the clashing IDs (530 ×3, 400 ×2) so the routing table can't point at the wrong asset.

Phase 3

Fill the content holes

Tom, authoring
7

Bring in the conversion layer Tom authoring

140 Access Explainer first (gate-critical), then 120 Pack 1, 130 IIM, 150 Secret Weapon, 160 Case Studies, 170 IHT Planning, 180/181/182. Each created with ACU dependencies + a resolves_signal tag. Fills the empty BS4/BS5 end of the funnel.

8

Backfill behavioural pieces + reconcile + gap register Tom

Add 450 Social Proof (S1), 540 Clicked-Not-Booked, 470 VM1. Resolve the boundary docs (701 / 520 / 655). Import 960 Content Gap Build Plan so the brain carries its own backlog.

The critical path in one line

Decide the model (1) → Werner builds the signal schema from 546 (3) → Tom re-tags existing renderings to restore routing (5) → author 140 + the packs to fill the closing end (7). Everything else is parallel or follow-on. The C4 gate (step 3) is the compliance-critical item, do not ship conversion content without it.

The 32-file re-import package (unlock-brain-reimport.zip) audited against the brain's locked ACUs + prohibited list. Headline: structurally clean, and on verification the tax-claim flags resolved in the files' favour — the brain is the stale party, not the package. Deterministic issues are fixed; company-specific financials remain the only genuine human-gate.

32/32
Files found + coded
2
Fixed for you
3
Verified → brain to update
1
Genuine human-gate
Fixed — done

Deterministic, no judgement needed

Done

U-code → canonical 544 codes (4 behavioural files)

440 / 450 / 470 / 540 normalised, legacy trail kept
▶
Applied
440 belief_target: U1 + P3 → C1 + P3 450 belief_target: U4 → S1 470 belief_target: U1 → C1 540 belief_target: U1 → U3 → C1 → C3 each keeps belief_target_legacy for the origin trail
Result
All four now carry canonical 544 codes. Corrected package re-zipped (unlock-brain-reimport-corrected.zip).
Confirmed clean

Prohibited-phrase sweep passed

No banned figures in deployable copy
▶
Checks
"7.8×" banned EIS multiple: 0 hits anywhere ✓ "22p" in 530: it's the agent card TEACHING the ban (❌ "22p" ✅ "~27.5p") ✓ correct use EIS 30% + 38.5p loss: consistent with locked ACUs across all files ✓
Verified → brain to update

I researched these; the files are right, the brain is stale

Drafted corrections below are RVL-ready — approving each is a yes/no, not work. I did not auto-overwrite locked ACUs (that's your approval gate).

Brain stale

BPR £2.5M cap — VERIFIED CORRECT

Files right · brain's acu_finance_act_2026 incomplete
▶
Verified (primary + adviser sources)
From 6 April 2026: 100% BPR up to a £2.5M per-person allowance (raised from the originally-announced £1M on 23 Dec 2025), 50% relief above, AIM shares 50% in all cases, ~£5M combined for a couple. The 12 files asserting this are accurate.
The brain is wrong here
locked acu_finance_act_2026 says only: "BPR continues to apply to qualifying EIS shares held 2 years, taking qualifying assets outside the taxable estate" — no cap, implies unconditional 100%
Drafted correction (approve to lock)
"The Finance Act 2026 reforms Business Property Relief from 6 April 2026. Qualifying business property, including EIS shares held for at least two years and held at death, receives 100% relief up to a £2.5M per-person allowance, with 50% relief on the excess. AIM-listed shares receive 50% relief in all cases. The allowance is not transferable on death but unused 100% relief passes to a surviving spouse, allowing a couple to shelter up to approximately £5M. Tax treatment depends on individual circumstances and may change." (replaces the uncapped wording; needs RVL sign-off)
Brain overstated

Pension IHT framing — files' "subject to final legislation" is accurate

Brain campaign says "enacted" — overstated
▶
Verified
Draft legislation published 21 July 2025; applies to deaths on/after 6 April 2027; still subject to final regulations. So "from April 2027, subject to final legislation" (the files) is correct; "enacted under the Finance Act 2026" (CAMP-001 compliance_override) overstates it.
Drafted correction (approve to apply)
CAMP-001 pension_iht_framing: change to "from 6 April 2027, under draft legislation published July 2025, subject to final regulations." Apply the same phrasing wherever the brain treats the pension change as enacted.
New ACU

SEIS 27.5p loss — arithmetically verified, lock it

Used in 12 files; brain has no SEIS-loss ACU
▶
Verified by calculation
SEIS: 50% income tax relief → 50p at risk per £1 loss relief on the 50p at additional rate (45%) = 22.5p recovered effective loss = 50p − 22.5p = 27.5p ✓ (the banned "22p" was the wrong figure)
Drafted ACU (approve to lock)
acu_seis_loss_relief: "For an additional rate taxpayer, the maximum effective loss on a failed SEIS investment, after 50% income tax relief and loss relief, is approximately 27.5 pence per pound invested. Tax treatment depends on individual circumstances and may change." Sibling to the locked EIS 38.5p ACU.
Also fix
The brain's acu_prohibited_22p note mislabels the correction as "38.5p (EIS)" — it should point to this new 27.5p SEIS figure. Drafted that correction too.
Genuine human-gate

I cannot verify this — it's yours

RVL — internal model

The conversion-layer financial model

Company-specific, not publicly checkable
▶
Unlocked claims (in 100/110/120/130/197/198)
~11× illustrative return · exit £50M–£80M · Y5 revenue £8.78M 5-yr total £16.3M · Y5 EBITDA ~£5.7M · £6.5M pre-money · £40K minimum
Why it's yours
These are management projections about Cloudworkz/Unlock itself — no external source can verify them, and they sit outside the brain's locked ACUs. They need RVL against your internal model + consistent "illustrative, not a forecast or guarantee" caveating before they become brain-canonical. This is the one item I'm genuinely handing back, because guessing here would be wrong. (Confirm the Tony Vine-Lott credential attribution in the same pass.)

Import verdict

The belief-signal docs (544/545/546/547) and the four behavioural files are claim-light and clear to import now. The tax claims that looked like blockers are verified correct, so the conversion layer is cleared on tax, gated only on the internal financial model RVL. All 32 files are now reformatted to brain schema and staged in _review/ (PENDING). Net unblocking action: approve the three drafted brain corrections (BPR cap, pension framing, SEIS ACU) + the internal financial-model RVL, then seed from the staged package.

Sources for the verified facts: GOV.UK — APR/BPR changes · Withers — 6 April 2026 BPR changes · Linklaters — pensions IHT proposed changes

Werner, this is your action list. Everything you need is staged in the vault at Projects/Unlock-Content-Brain/_review/ — the formatted package (unlock-brain-import-formatted.zip) and the full handoff note (2026-06-05-WERNER-reimport-handoff.md). Start there.

The one thing to understand first

The brain stored a generic 5-stage funnel (BS1–BS5) where Tom's real model is the 544 signal map (QT/QL, C1–C4, the G·L·P clusters, S1–S6, with states + gates). Restoring that signal model is the highest-leverage move and is what re-enables transcript scoring and call routing. The source docs for it are now in the new 13_belief_signals/ folder.

Werner's task list

In dependency order — tick each off

Status saves in this browser. 1 and 4 gate what follows.

Gates 2,3,5

1 · Model the belief-signal schema

from 545 (canonical) + 546 (spec), in 13_belief_signals/
▶
Build a belief_signals table (code, definition, state enum Green/Amber/Grey/Red/NA, persona-cluster map) plus a gates rule for the C4 compliance gate and the Pack 1 gate. 547 is the 25-question capture instrument that feeds it.
Low-risk

2 · Add the low-risk schema fields

quality layer · lifecycle_stage · backed_by
▶
Surface the existing craft_result/uat_result as a named quality layer; add lifecycle_stage (5 states); begin backed_by population from the citations already in atom sources.
Needs Tom's approval

3 · Apply the 3 ACU corrections

BPR £2.5M cap · SEIS 27.5p · pension framing
▶
Update acu_finance_act_2026 to the £2.5M-cap wording; add acu_seis_loss_relief (27.5p); fix CAMP-001 pension framing to "subject to final regulations." Full text in the handoff note + Re-import QA tab. Blocked until Tom approves.
Do before seeding

4 · Run the ID-collision pass

530 ×3 · 400 ×2
▶
Namespace the clashing IDs before seeding any renderings, or the routing table points at the wrong asset.
Sequenced seed

5 · Seed: belief-signal + behavioural first, renderings on Tom's RVL clear

renderings stay PENDING until the financial-model RVL
▶
The 13_belief_signals and behavioural files are claim-light and can seed now. Hold the 20 02_renderings until Tom clears the internal financial-model RVL.

Where everything lives

This page (read-side, this URL) · the handoff note + formatted package in _review/ (vault, synced) · the routing + status detail in the Re-import QA and Roadmap tabs here. Open questions for you sit in the Schema Gaps tab (parser status, Q1–Q7).

Merged audit + V2 schema gaps + re-import QA + Werner handoff · live read of Projects/Unlock-Content-Brain, /brain MCP, doc 544, 138-file matrix, V2 proposal, reimport package · tax facts verified against gov.uk + adviser sources June 2026 · working document, decisions not yet ratified.