# Source and state register (REV-01)

Orchestrator-verified on `lifeos-devbox` 2026-08-30. Every line here was re-derived on the box, not
copied from the mission handoff. Where the handoff's launch-time appendix differs, the difference is
recorded explicitly.

## Review identity

| Field | Value |
| --- | --- |
| Review goal ID | `fusion-ppc-system-audit-20260830` v1 |
| Review goal SHA-256 | `287532968913b5b5fda888cdf38f02496a523ed2314f4da91e4d4ba3b49690e8` |
| Review directory | `outputs/reviews/fusion-ppc-system-audit-20260830/` |
| Orchestrator model / effort | `claude-opus-5` / `xhigh` — readback from `~/.claude/.box-local.json` and `~/.claude/settings.json`, both `{"model":"claude-opus-5","effortLevel":"xhigh"}` |
| Host / session | `lifeos-devbox`, tmux session `fusion` |
| Working branch | `round-13-takeover` @ `8351b59` |
| Spend | USD 0.00 (cap USD 12) |

## The reviewed system: actual state

### Build-run status — the run stopped after T1

`tmux ls` on the box lists no architecture or child session. The running sessions are unrelated
(`ccp-analit`, `ccp-fusion`, `ccp-kertujitok`, `ccp-lovasember`, `client-control-panel`,
`coldemail-hu-smb-1000`, `proposal-studio`, and this review's `fusion`).

| Task | Codex task ID | Model / effort | Branch | Commits ahead of `638e7524` | Child artifacts | True state |
| --- | --- | --- | --- | ---: | --- | --- |
| T1 evidence corpus | `01a05246-2df4-7673-9c36-c0a48c844f2d` | Sol / medium | `codex/fusion-ppc-t1-evidence` | 7 | **YES** | `PARENT_ACCEPTED`, integrated as `b1013882`, goal closed |
| T2 measurement audit | `01a05256-0aec-71b1-870d-501b58d9a53b` | Sol / medium | `codex/fusion-ppc-t2-audit` | 12 | **313 lines, untracked on disk** | ran 18 min, died mid-turn, never committed — see correction below |
| T3 decision dashboard | `01a05256-3afc-7a01-b94e-0493d927bef3` | Sol / medium | `codex/fusion-ppc-t3-dashboard` | 8 | none | `WAITING_DEPENDENCY` |
| T4A single-LP strategy | `01a05256-70b6-7e62-bea7-7433bb2ff1af` | Sol / xhigh | `codex/fusion-ppc-t4a-strategy` | 8 | none | `WAITING_DEPENDENCY` |
| T4B full-suite strategy | `01a05256-b30e-7c22-afd6-ec04f8d45977` | Sol / xhigh | `codex/fusion-ppc-t4b-strategy` | 8 | none | `WAITING_DEPENDENCY` |
| T5 client report | `01a05256-e860-7613-b02f-c9ee9aef1399` | Sol / medium | `codex/fusion-ppc-t5-report` | 8 | none | `WAITING_DEPENDENCY` |
| T6 owner action review | `01a05257-1f06-7572-9905-3d953ff1c739` | Sol / medium | `codex/fusion-ppc-t6-review` | 8 | none | `WAITING_DEPENDENCY` |
| T7 targeting pack | `01a05257-5260-7583-b547-0ba0b75eebb5` | Sol / medium | `codex/fusion-ppc-t7-targeting` | 8 | none | `WAITING_DEPENDENCY` |

The 8 commits on T3-T7 and 12 on T2 are orchestrator control-file commits, not child work. The
`9f92003` / `a286b77` heads are orchestrator commits fast-forwarded into the child worktrees.

Architecture parent acceptance ledger: ARCH-01, ARCH-02, ARCH-03 `PASS`; **ARCH-04 through ARCH-10
`PENDING`**. Architecture `run-state.md` still declares `Status: ACTIVE` and
`Last update: 2026-08-30T12:14:00+00:00` — that is a **stale flag**, not current truth, and nothing
in the system contradicts it.

### ORCH-CORR-1 — the fixed point is not on the reviewed branch

| Check | Result |
| --- | --- |
| `git merge-base --is-ancestor 638e7524 round-13-takeover` | **false** |
| `git branch -a --contains 638e7524` | only `origin/codex/fusion-ppc-architecture-orchestrator` |
| merge-base of the two lines | `08f52cd docs(r18): record hidden review strip` |

Two commits carry the identical subject `plan: freeze Fusion PPC architecture delegation`:

| Commit | Parent | Branch |
| --- | --- | --- |
| `638e752430a7fbce735e46b9381f04e7b40356f8` | `310758d` | `origin/codex/fusion-ppc-architecture-orchestrator` |
| `ce93cef7a8f9ac1d372e232a88e7009a73bc5584` | `f427516` | `round-13-takeover` (reviewed line) |

The frozen goal is byte-identical on both — blob `2421c9e`, SHA-256
`547b20a3f969f6815afec9ebe52f90383b9908fbcda6f41558df95117e9a5f2f`, matching the handoff's claim and
matching the current working copy. The **goal** was re-frozen faithfully; the **branch** it was
re-frozen onto is missing the PPC work.

41 files exist on the orchestrator branch and never existed on `round-13-takeover`, including the
entire PPC evidence base. `git diff --stat 638e7524..round-13-takeover` reports
`131 files changed, 5703 insertions(+), 70003 deletions(-)`; no commit in that range deletes
anything — the 70,003 lines are divergence.

All eight child worktrees were based on `round-13-takeover`.

### Artifact register: what exists, where, and its acceptance state

| Artifact | Location | On reviewed branch? | Acceptance state |
| --- | --- | --- | --- |
| T1 evidence corpus (716 lines MD) | `outputs/loop-runs/20260830-fusion-ppc-architecture-orchestration/children/t1-evidence-corpus/artifacts/` | yes | **accepted and integrated** |
| Orchestration apparatus (`launch_child.py` 291 lines + 5 parent ledgers + 8x5 child control files) | `outputs/loop-runs/20260830-fusion-ppc-architecture-orchestration/` | yes | operational, run halted |
| 30-day PPC/SEO/CRO report package (18 files incl. `report.md`, `report_data.json`, `recommendations.json`, `.pptx`, 6 receipts) | `outputs/loop-runs/2026-08-28-fusiontables-report/` | **NO — only `ledger.json`** | pre-existing, delivered 2026-08-29 |
| KPC keyword-performance-control system (`config.yaml`, 4 advice bundles, 4 advice outputs, 2 evidence ledgers ~21.7k lines each, stage-3 template) | `ads-control/` | **NO** | pre-existing |
| KPC report-recovery run (`index.html`, `build_advice.py`, `browser_qa.js`, `validate_final.py`, 2 gate screenshots) | `outputs/loop-runs/20260824T085117Z-fusion-kpc-report-recovery/` | **NO** | pre-existing |
| Report closeout page | `outputs/loop-runs/20260828T232516Z-fusiontables-report-closeout/` | **NO** | pre-existing |
| Meta creative-control dashboard | `deliverables/fusion-meta-ads-review-2026-08-24/index.html` + `test_dashboard.py` | **NO** | pre-existing |
| PPC audit deliverable 2026-07-30 (executive brief, internal audit, deck, prototypes) | `deliverables/fusion-ppc-2026-07-30/` | yes | pre-existing |
| PPC council deck 2026-07-31 | `deliverables/fusion-ppc-council-2026-07-31/` | yes | pre-existing |
| **Decision dashboard (T3)** | — | — | **UNAVAILABLE — never produced** |
| **Strategy A, single landing page (T4A)** | — | — | **UNAVAILABLE — never produced** |
| **Strategy B, full suite under 1.5x ceiling (T4B)** | — | — | **UNAVAILABLE — never produced** |
| **Hungarian client report from this run (T5)** | — | — | **UNAVAILABLE — never produced** (the 2026-08-28 package is a different, earlier deliverable) |
| **Owner action + voice review (T6)** | — | — | **UNAVAILABLE — never produced** |
| **Targeting pack (T7)** | — | — | **UNAVAILABLE — never produced** |
| **T2 measurement audit** | — | — | **UNAVAILABLE — never produced** |

Everything marked "NO" on the reviewed branch was materialised read-only for the review lanes at
`/tmp/.../scratchpad/orchbranch/` via `git archive origin/codex/fusion-ppc-architecture-orchestrator`.
Nothing was committed to or merged into `round-13-takeover`.

## Account and data identity (from the accepted T1 corpus, re-read on the box)

| Item | Value |
| --- | --- |
| Google Ads account | `293-822-3921` |
| Meta account | `1783210591808804` "Fusion Tables Hungary" |
| Report windows | 2026-07-07..2026-08-05 and 2026-07-25..2026-08-23 |
| Lead-register windows | differ by one day at each edge; kept separate |
| KPC window | 2026-06-25..2026-08-23 |
| GA4 | **UNAVAILABLE** — no property for this client |
| Search Console | **UNAVAILABLE** — no property for this client |
| Qualified leads, accepted quotes, purchases, sales, revenue | **UNAVAILABLE** |
| Search-campaign delivery in the KPC window | source-verified numeric **zero** (a real zero, not missing) |

## Verified figures from the delivered 30-day report (`report.md`, 2026-07-25..2026-08-23)

| Metric | Current | Previous | Source note |
| --- | ---: | ---: | --- |
| Google Ads spend | 92,096 Ft | 29,101 Ft | read-only query |
| Google Ads clicks | 4,410 | 452 | all traffic from PMax; **search campaigns delivered nothing** |
| Google Ads platform conversions | 49 | 1 | the report states plainly this is a platform signal, not a verified enquiry and not revenue |
| Meta spend | 89,778 Ft | 30,865 Ft | authenticated Ads Manager export |
| Meta clicks | 825 | 322 | result column uses a **form-start** event, which does not prove submission |
| Recorded enquiries (client's own register) | **13** | 7 | 7 QA test rows excluded; register has no source or campaign column |

The 49-versus-13 gap between platform conversions and recorded enquiries is the report's own headline
finding and drives its P1 recommendations. Attribution of the 13 enquiries to traffic source is
**impossible** with the current data: the lead register has no source or campaign field.

## Unavailable at review time, with reason

| Item | Reason | Effect on review |
| --- | --- | --- |
| Mac-local paths named in the handoff (`/Users/agency/...`, `/private/tmp/fusion-review-handoff-20260830/chat-gists/`) | devbox-only run; not mirrored | requirements sourced instead from the handoff's verbatim quotes and the accepted T1 session gists, which cover the same sessions |
| `ads-control/reports/stage-3-20260824T111713-2aa95a23.html` (rendered KPC report) | not committed on any branch; only the template and the advice bundles are | the stage-3 **template** was reviewed instead, which doubles as the first-run/empty-state test |
| Live Google Ads / Meta API readback | review authority is read-only inspection; no credential use was required or attempted | platform figures taken from the already-verified 2026-08-28 export receipts, not re-pulled |

## Reproduction

```
cd /home/matt/mirror/Fusion && git checkout round-13-takeover   # @ 8351b59
git merge-base --is-ancestor 638e7524 round-13-takeover; echo $?   # 1 = not an ancestor
git log --format='%H %P %s' -1 638e7524; git log --format='%H %P %s' -1 ce93cef
comm -23 <(git ls-tree -r --name-only origin/codex/fusion-ppc-architecture-orchestrator | sort) \
         <(git ls-tree -r --name-only round-13-takeover | sort) | wc -l   # 41
sha256sum outputs/loop-runs/20260830-fusion-ppc-architecture-orchestration/goal-v1.md  # 547b20a3f9...
```


---

## Correction to this register, made by the orchestrator after Lane 3's evidence

**T2 did start, ran for about 18 minutes, and produced four artifacts. They were never committed.**

My initial ground-truth pass read `children/t2-measurement-audit/run-state.md` (`WAITING_DEPENDENCY`,
`Started: PENDING`) and the empty git diff, and concluded T2 never ran. That was true of git and
false of the filesystem. Verified directly:

```
git -C /home/matt/mirror/Fusion-ppc-children/t2 status --porcelain -uall
 M …/t2-measurement-audit/access-ledger.md
 M …/t2-measurement-audit/run-state.md
?? …/t2-measurement-audit/artifacts/current-state-funnel-and-data-flow.md    82 lines
?? …/t2-measurement-audit/artifacts/event-conversion-taxonomy.md            149 lines
?? …/t2-measurement-audit/artifacts/fresh-evidence-register.md               60 lines
?? …/t2-measurement-audit/artifacts/edit-manifest.md                         22 lines
                                                                     total 313 lines
```

This makes the trigger design's failure worse, not better: T2 verified every hash, ran its access
scouts, wrote real audit artifacts, and then died mid-turn without committing. Its own
`run-state.md` — the first file a recovering agent would read — was never updated and still says it
had not started. The work is recoverable and should be salvaged before any relaunch decision.

Also corrected: `ppc.totals.conversion_value` is **not revenue**. It reads
`{"current": 49.0, "previous": 1.0, "unit": "huf"}` — numerically identical to
`ppc.totals.conversions` `{"current": 49.0, "previous": 1.0, "unit": "count"}`. Every conversion
carries a default value of 1, relabelled as forint. The pipeline then derives
`roas = 0.000532056…` and `cpa = 1879.50 HUF` from that field. There is no revenue in this account's
data at all.

And the five source windows do not align:

| Source | Current window | Previous window |
| --- | --- | --- |
| `google_ads` | 2026-07-25..2026-08-23 | 2026-06-25..2026-07-24 |
| `meta_ads` | 2026-07-25..2026-08-23 | 2026-06-25..2026-07-24 |
| `bigquery` | 2026-07-**24**..2026-08-**22** | 2026-06-24..2026-07-23 |
| `crm_sheet` | 2026-07-**26**..2026-08-**24** | 2026-06-26..2026-07-25 |
| `work_evidence` | 2026-06-23..2026-08-23 (62 days) | `null` |

The enquiry count that forms the last funnel step is measured over a different 30 days than the
impressions and clicks above it.

Finally, `meta_ads.lead_generation.comparability` reports `comparable: true` while the previous
window delivered on **11 of 30 days** (`ratio 0.3667`), because `min_delivery_day_ratio` is
`0.3333`. Meta's period-over-period growth is therefore partly an artifact of more delivery days.