🐴 LOVASEMBER · ASTRA HIGH EXECUTION

Verification and Constraints

● PARTIAL · REVISION 3 · CYCLE 2

Run the optimized Meta campaign with reliable Webflow booking-intent tracking and independently prove that leads reach the correct Google Sheet.

Astra High · pinned orchestrator15 / 18 criteria verified2–3 creatives · 2–3 copy variants4,000 Ft/day configured budget

Goal and finish line

Selected ads serve at 4,000 Ft/day using two or three creatives and two or three fixed copy variants. The closest reliable Webflow action is measured consistently through GTM and Meta Pixel. A separate source-to-row receipt proves the actual lead integration reaches the correct Sheet. A click is never reported as a completed booking.

Authority and autonomy

Matt reviewed v1 and explicitly instructed this extension and launch. Astra chooses the simplest implementation that satisfies the deliverables and records its evidence.

Parent steers once now, stays idle for 20 minutes, then checks once. Further checks happen after 60-minute idle intervals until every criterion is proved. One consolidated steering prompt per check, only when needed.

15 of 18 checks are verified. Exact GTM rule execution and Meta Processed event receipt are now proved. Fresh Sheet intake/retry await the notified-test decision or a fresh genuine source. Meta deduplication is still parsing. Four selected ads, budget and offer remain unchanged.

End goal system

Selected Meta ads

2–3 creative and 2–3 copy variants.

Webflow booking intent

Consistent paid offer and Calendar handoff.

Reliable measurement

GTM and Pixel prove the owned intent action.

Separate Sheet proof

A genuine source record reaches the right tab.

Route to the end state

Reconcile history

Find blockers and rank strongest evidence.

Optimize and repair

Choose assets, audience and minimal fixes.

Test both paths

Webflow event receipts and source → Sheet proof.

Launch and prove

Read live settings and fresh delivery.

Evidence and open questions

CONFIRMED STARTING POINT

26 Sep snapshot: active lead campaign at 4,000 Ft/day, two active ad sets and 16 ads. The current optimization event is an appointment CTA click. The named Google Sheet and its consultation/form tab headers are readable.

AUTHORIZED CHOICE

Use the existing campaign or a replacement. Spend only on the selected 2–3 creatives and 2–3 fixed copy variants. Use submission evidence for ranking and audience narrowing where available.

TO PROVE

Fresh Sheet intake/retry and Meta browser/server deduplication remain unproved. Exact GTM execution, a processed Meta event with matching ID, eight browser interaction cases, historical source match and selected-ad delivery are verified. Historical rankings remain provisional.

The former five-body/five-headline asset feed returned no individual copy outcomes. It has been replaced in the four selected ads by two fixed copies. The choices are provisional and individually identifiable for future measurement.

Selected fixed copy

Two visual concepts and two complete copies. Copy choices remain provisional because historic individual-copy outcomes are unavailable.

A_v9

Ló-asszisztált coaching és egyéni elvonulás Szálkán. Lovak, természet és személyes figyelem egy helyen. A programot online konzultáció előzi meg, amelynek díja 5 000 Ft. Ismerd meg a lehetőségeket és foglalj időpontot.

Ló-asszisztált coaching Szálkán
Online konzultáció: 5 000 Ft

CTA: LEARN_MORE · https://www.lovasember.hu/

B_v1

Önismeret a természetben, egyéni figyelemmel. Ismerd meg a szálkai ló-asszisztált coaching és elvonulás lehetőségeit. Az első lépés egy online konzultáció, ahol átbeszéljük a folytatást. A konzultáció díja 5 000 Ft. Részletek és időpontfoglalás a weboldalon.

Önismeret a természetben
Egyéni program Szálkán

CTA: LEARN_MORE · https://www.lovasember.hu/

Existing creative referencesSelected images · outcome ranking remains provisional

**v9 · Historical traffic reference**
Two high-volume ads shared this Sanyi/horse image. Submitted-lead ranking cannot be joined to the historical records.

**v1 · Recent delivery concentration**
More recent delivery favored this swan creative. Delivery alone does not prove more received forms.

Execution proof checklist

15 of 18 verified

Diagnose the delivery and lead gap

Explain the lead gap from exact account and matching business evidence.

2 / 2

Criteria, constraints and proof

PASS · When exact-account status, billing/restriction, settings history and matched before/after performance are read, then a dated diagnosis names proved blockers, proved changes and unresolved causes in the private evidence ledger and a concise board summary.
Fresh exact-account API read: Active, disable_reason 0, HUF. History confirms payment-needed → disabled → payment-needed → active transitions, with Active restored 21 Sep. 25 Sep budget change 2,000 → 4,000 is confirmed. Billing-related interruption is evidenced; lead quality remains unjoined. Private receipts: account.json, activities.json, campaigns.json, historical/recent insights.
- Redacted live configuration receipt
PASS · When Meta rows are reconciled with historical form/Sheet and booking records for matching periods, then creative, copy and audience performance is ranked using submitted/qualified outcomes where available, with counts, spend, source confidence and missing joins visible.
Matched windows reconciled: Jul 1–Aug 25 has 23 raw Calendar records / 23 distinct event IDs. Sep 21–26 has zero recorded rows, not verified zero bookings. No ad/UTM join or payment/qualification status exists. Therefore no submitted-lead creative/copy/audience ranking can be proved. Selection uses explicitly labelled traffic proxies. One old sync error was found, current sync health remains unknown.
- Redacted tracking and historical proof

Preserve

Confirm the exact account before interpreting payment warnings.
Preserve the distinction between clicks, submissions, bookings and qualified business outcomes.

How to verify

Provider exports/readbacks, dated settings diff and reproducible reconciliation script or ledger. Do not publish customer records.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Select the strongest ads and audience

Concentrate spend while making evidence and uncertainty visible.

2 / 2

Criteria, constraints and proof

PASS · When the selection ledger is opened, then it contains precisely 2–3 chosen creatives and 2–3 complete copy variants, previews/exact wording, historic evidence or explicit provisional rationale, and a keep/pause decision for the rest.
Exactly two image hashes selected: v9 Sanyi/horse and v1 swans. Two complete fixed copy variants saved with exact text and price in the private selection ledger. Individual copy performance is unavailable in the asset breakdown export, so both are labelled provisional. Keep four existing ads across two current audiences; pause the other twelve.
- Redacted live configuration receipt
PASS · When the proposed targeting is compared with historical submission evidence, then every material narrowing or exclusion has a documented reason and scope, or the board states why a broader audience was retained.
No new demographic narrowing or exclusion was introduced because the historical records lack an attribution join. Existing HU geography, age control and two Advantage+ audiences were retained. Interests, lookalikes and suggested age range are expansion suggestions. The inherited 180-day Lovas_coaching_lead exclusion is preserved but its business meaning is unverified.
- Redacted live configuration proof

Preserve

Do not declare an underdelivered variant a loser or a mixed asset-feed sentence a proven winner.
Use factual claims and rights-cleared assets, preserving offer terms.

How to verify

Private selection table, selected asset hashes/IDs and text, aggregate historical outcomes, Ads Manager preview/screenshots as needed.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Make the conversion path work

Make the paid offer and the owned Webflow-to-Calendar handoff coherent and reliable.

2 / 2

Criteria, constraints and proof

PASS · When each chosen ad destination and primary CTA is followed on the actual desktop and mobile site, then the intended form or paid booking route opens with consistent service, price and next-step wording.
Published paid FAQ wording/link and SEO description agree with the live 30-minute 5,000 Ft Calendar route. All selected ads share the verified homepage. Desktop/mobile routes and actual postpublication section screenshots reviewed. Mobile is emulated, not a physical handset.
- Postpublication interaction matrix
PASS · When the owned Webflow booking CTA is activated on desktop and mobile, including its text, icon, padding and keyboard activation, then it opens the intended Calendar route once. Invalid or abandoned inputs do not appear as completed bookings in our tracking.
All eight text/icon/padding/Enter cases pass at 1280x900 and390x844. Each opens the intended Calendar once. Page load, unrelated anchors and abandoned Calendar emit no false completion. Mobile uses viewport emulation, not physical handset.
- Postpublication interaction matrix

Preserve

Preserve the 5,000 HUF self-knowledge offer and unrelated pages.
Use the existing booking route. A paid test is optional, within the shared USD 20 extra-spend cap, and cannot justify replacing the funnel or contacting clients.

How to verify

Postpublication desktop/mobile screenshots, actual handoff URL and controlled interaction results. Confirm consistent 5,000 HUF offer wording and no false booking-completion signal.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Prove the correct Google Sheet receives leads

Prove a genuine source record reaches the correct Sheet while preserving the client’s existing working data.

1 / 3

Criteria, constraints and proof

PASS · When the accessible live integration and a genuine source receipt are traced, then the exact destination spreadsheet, raw tab and field mapping are identified and matched to the intended consultation record.
Independent Stripe payment and paid checkout match the production Calendar schedule and one raw consultation row. Exact destination, raw-tab mapping and composite identity match verified. This is an August 24 accepted record freshly queried today, not new September intake.
- Independent source-to-row receipt, redacted
PENDING · When a uniquely identified accepted lead or booking passes through the actual integration, then exactly one matching raw record appears in the intended Sheet tab with the expected service, timestamp and source identifier. Use a controlled QA receipt or a fresh genuine record with an independently matching source receipt.
Historical accepted source is independently proved, but fresh intake remains unobserved. Parent asked Matt whether one standard automatic Sanyi notification may accompany a 5,000 Ft test booking. No answer forwarded. No notified purchase or Sheet mutation performed.
- Independent source-to-row receipt, redacted
PENDING · When the proved source record is read again after a normal sync or a safe retry, then its unique identifier still has one raw row. Invalid or abandoned test input adds no accepted record, and derived CRM copies are distinguished from raw duplicates.
Raw source identifier occurs once and the matched row is unchanged on reread. Derived CRM copies are reconciled separately. Current source-sync or safe production replay is unobserved, so retry idempotency is not claimed.
- Independent source-to-row receipt, redacted

Preserve

Preserve all real rows, formulas, formatting, tab names and working notifications. Do not clean up or reorganize the client’s Sheet.
Read first. Any necessary test must use conspicuous task-owned data through the real intake, with a minimal reversible footprint. Never send email/messages to clients or leads or trigger known customer-facing notifier tests.

How to verify

Bounded before/after Sheet reads plus an independently matched accepted source receipt or handler trace. Match by source ID/test marker, not counts. Redact all personal data from the public board.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Prove Google Tag Manager behavior

Prove the chosen Webflow intent event and the exact GTM trigger work consistently.

2 / 2

Criteria, constraints and proof

PASS · When a marked Webflow booking-intent action and negative controls are tested, then the exact published GTM container/resource, relevant tag identifiers, trigger inputs, consent behavior and executed destinations are proved. Use either Tag Assistant’s full container event detail or a reproducible trace that connects the loaded published container’s version/fingerprint and rule to captured live event variables and actual tag requests from the same controlled action. Static source alone, a tag merely being installed, or aggregate analytics totals do not pass.
Cycle 2 equivalent proof: exact loaded GTM-PWHBSLGB resource 15, fingerprint, tags 41/42 and click/element-ID rule match live gtm.click 24 and two intended destinations once. Page load and negative clicks emit no intent. Consent commands/default absence and request parameters recorded unchanged. Reproducible verifier passes. Admin panel/version metadata not inspected.
- Exact loaded resource and controlled execution proof
PASS · When the CTA text, icon, padding and keyboard action are tested on desktop and mobile, then each intended activation sends the chosen intent event once to the correct measurement destinations. Page load, unrelated clicks and invalid or abandoned input do not send false intent or completed-booking events.
Each accepted action sends one appointment_cta_click to GA4 G-71ZMEVMRZX (204) and Pixel 755189613954638 (200). Negative controls send no false intent or completion. Clean mobile retest excludes the earlier overlapping debug-session confound.
- Postpublication interaction matrix

Preserve

Preserve legitimate consent behavior and existing unrelated tags.
Do not count Tag Assistant Summary, tag presence or GA4 aggregate Realtime as controlled event proof.

How to verify

Tag Assistant event-detail screenshot plus controlled-session network receipt/provider debug detail and published GTM version readback if changed. If admin access fails, exhaust existing routes and prove the layers still accessible; leave only the unavailable layer pending.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Prove Meta Pixel conversion behavior

Prove Meta receives the selected Webflow intent event and the campaign uses that same signal.

1 / 2

Criteria, constraints and proof

PASS · When the marked Webflow booking-intent action is performed, then Pixel/dataset 755189613954638 receives the exact selected event, time and identifier in Events Manager Test Events or an equivalent exact provider receipt, and the campaign’s selected optimization event matches.
Recovered Test Events in existing agency business context. Pixel 755189613954638 shows appointment_cta_click Processed Browser at 22:32:49 UTC with exact event ID matching the controlled Chrome request. Both selected adsets use this same Pixel/event. Browser receipt, paid attribution and booking completion remain separate.
- Exact processed Meta event receipt
PENDING · When normal and negative actions are exercised, then there is one browser event per intended activation, no false completed-booking event, and no duplicate conversion. If browser plus server events remain enabled, matched event IDs and provider ingestion/deduplication are proved, or a verified single-source route is used with the tradeoff documented.
One browser intent and negatives pass. Exact Browser receipt now proved. Production provider lists Meta-hosted CAPI counts, but event deduplication says Still Parsing Your Data. Matching bridge IDs remain transport evidence only. No verified narrow single-source control found. No integration change made.
- Exact receipt and deduplication limit

Preserve

Do not fabricate a Calendar completion signal or claim test-event receipt proves paid attribution.
Keep personal data out of public evidence and send only permitted, consented event data.

How to verify

Exact Pixel/dataset and custom-event receipts, controlled timestamps/IDs, Meta Test Events or equivalent provider readback, negative tests, and independently read campaign optimization settings. Technical event receipt is distinct from paid attribution and completed bookings.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Launch the optimized campaign configuration

Apply the optimized setup and verify every active setting and budget.

3 / 3

Criteria, constraints and proof

PASS · When the selected configuration is saved and read back from Meta, then the live existing or replacement campaign uses only the chosen 2–3 creatives and 2–3 copy variants, the evidence-based audience, correct destination and selected verified optimization event.
Fresh exact live configuration has only two selected images/two copies/four ads and retained audience. Both active adsets select Pixel 755189613954638 / appointment_cta_click / OFFSITE_CONVERSIONS, matching the event proved by the eight-case browser matrix. Exact Meta ingestion/dedup remains separately pending under D6.
- Postpublication interaction matrix
PASS · When all budget owners and active objects are inventoried, then configured active daily budgets total at most 4,000 HUF, unselected variants cannot receive new delivery, and the previously paused unrelated sales campaign remains paused.
Fresh account-wide campaign and lead-campaign ad-set readback: active budget sum is exactly 4,000 HUF/day. Exactly four selected ads remain enabled under the two active sets. All twelve unselected eligible ads are paused. The unrelated sales campaign remains paused.
- Redacted live configuration receipt
PASS · When ad previews and object diagnostics are checked, then chosen ads show correct asset/copy/CTA/destination combinations and no unresolved rejection or account delivery blocker is hidden behind an active toggle.
21:01 UTC fresh API readback: all four selected ads are ACTIVE without issues_info or ad_review_feedback. Exact saved copies, images, CTA and canonical URL match the independent blind preview review. No current account disable reason.
- Redacted configuration receipt

Preserve

Preserve a before/after manifest and rollback; pause rather than delete unselected assets.
Do not spend on additional unselected variants, silently increase the budget, or promise higher performance from configuration alone.

How to verify

Exact object IDs/settings, provider readback after writes, selected previews, budget-owner sum and active/paused inventory. Configured daily budget is not a hard daily debit guarantee.

Blind Luna High description completed for both provider previews. Parent comparison confirms the selected image subjects, fixed copies, 5,000 Ft price, Learn more CTA and canonical domain. Delivery/review status is verified separately.

Verify delivery and hand over honestly

Prove new delivery and hand over the verified state with honest limits.

2 / 2

Criteria, constraints and proof

PASS · When fresh insights after activation are read, then new impressions or spend prove intended ads are delivering, or the exact provider delay/blocker remains pending with its next check. No prior-period spend is presented as post-change delivery.
Fresh read 23:01 Budapest: 32 hourly rows from 13:00 onward after all creative edits/review. All four selected ads appear, total 1,803 impressions and 1,808 HUF. Fresh account/ad diagnostics remain clear, configured budget 4,000 HUF/day.
- Post-change delivery receipt
PASS · When the final board/report and Sanyi draft are opened, then each accepted claim links to its evidence, remaining access/attribution/business-outcome limits are explicit, the optimized settings and rollback are recorded, and the Hungarian email describes only verified work while remaining unsent.
Cycle 2 report, current board and unsent Hungarian draft include 15/18 verified checks, exact GTM execution and Meta Processed receipt, remaining Sheet and provider-dedup limits, rollback and USD 0 extra spend. Parent retains monitoring and final acceptance.
- Revision 3 evidence and limits

Preserve

Do not claim lead uplift or a qualified customer from QA submissions.
Do not wait weeks in the foreground or invent completed future reviews; preserve a clear pending outcome review owner/date.

How to verify

Timestamped post-change insights, redacted evidence index, tested scripts/readbacks, current board and concise terminal report. Technical completion and long-term performance remain distinct.

GUI evidence, blind Luna High description and comparison: pending where applicable.

Meaningful constraints

Stay within the existing 4,000 Ft/day total configured budget. Preserve rollback and unrelated paused campaigns.
Preserve the 5,000 Ft self-knowledge offer. Keep the separate future free leader offer distinct.
Preserve customer rows, formulas, existing notifications and legitimate consent behavior. Use marked QA data.
No new security layer. Keep this URL public and customer data/private evidence off it.
Do not send emails or messages to clients/leads or promise a lead uplift. Optional tests and extra tools share a USD 20 total limit. A click is not a paid booking.

Live progress and parent feedback

Parent: Matt explicitly authorized the extended scope and autonomous Astra High launch. One scheduled parent check will occur 39 minutes after launch. The prior mandatory seven-day wait and 14-day image test are replaced by outcome-based deliverables.

12:20 Budapest · Execution ownership established. Campaign selection and configuration are handled separately from the form/Sheet/GTM/Pixel evidence work.

12:23 Budapest · Exact-account diagnosis passed. Existing lead campaign is active at 4,000 Ft/day. The unrelated sales campaign remains paused. Asset-level body/title/image breakdown requests returned no rows, so individual copy winners cannot be proved from this API export.

12:29 Budapest · New-creative API was blocked by app development mode, so existing ads are being edited through authenticated Ads Manager. First fixed copy is published and independently read back. Twelve unselected ads are paused with receipts. No budget increase.

12:32 Budapest · Both fixed copies are now saved in Meta. Exact inventory readback confirms four eligible ads, two image hashes, two creative objects and no mixed asset feeds. Tracking investigation found the CTA event misses text/icon clicks but fires on button padding. A narrow repair is in progress.

12:35 Budapest · Blind preview review confirms both saved ad combinations. Tracking tests distinguish normal text clicks from button-padding clicks. The client draft now states only verified work and explicitly leaves booking, Sheet receipt and repair open. It is unsent.

12:35 Budapest · All four selected ads have cleared Meta review and read back as ACTIVE with no ad diagnostics. Independent preview comparison passed. Active status is distinct from new delivery proof.

12:38 Budapest · Tracking recovery finished with explicit access blockers. Exact padding-click requests reach GA4 and Pixel, but text/icon clicks miss the event. No accepted booking or QA Sheet row was produced. Website changes were not deployed. All evidence is separated from campaign configuration success.

12:57 Budapest · Single authorized parent adjustment received. Distinct recovery checks cover a suitable existing enquiry form, actual production serving ownership and current sync access. Production DNS points to Webflow CDN with DotRoll DNS, not the agency review host. A one-time delivery check is scheduled in this task for 13:07 Budapest. All 18 criteria and existing constraints remain.

13:03 Budapest · Distinct recovery completed: no equivalent live enquiry form exists on the four sitemap routes, couples CTAs are unrelated packages, and historical /jelentkezes is 404. The agency notifier has recent completed executions, while the client-owned Calendar sync remains inaccessible. Exact asset/account handoff and unapplied minimal repairs are prepared. Waiting only for the scheduled clean-window delivery read.

13:07 Budapest · Both clean-window reads returned zero selected-ad hour-13 rows. Configuration checks still pass. D8.1 remains pending, with a recommended 14:00 read, not scheduled. The one-time continuation is completed and paused. Revision 2 retains all 18 criteria and 8 passes.

Revision 3 · Matt clarified that Webflow booking intent is the measurable optimization outcome, authorized available Webflow/GTM access and up to USD 20 total extra spending, and required minimal-touch genuine Sheet receipt proof. Parent verified the correct Comet Webflow designer. The same Astra High task resumes with a 20-minute first review and 60-minute later reviews. No check is newly passed by changing its wording.

23:05 Budapest · Production HTTP readback confirms the scoped CTA descendant-click CSS, paid FAQ sentence, booking-section link and corrected meta description. Two independent Luna xhigh visual checks launched. Clean post-change Meta buckets contain 1,803 impressions and 1,808 Ft across all four selected ads.

23:48 Budapest · All eight text/icon/padding/Enter cases pass after publication, with one outgoing GA4 and Pixel intent per activation and no false completion. Fresh access recovery separates available GA4/Pixel transport from unavailable GTM container and exact Meta ingestion/dedup details. Historical Stripe-to-row match is proved, fresh intake and sync remain pending. Extra spend USD 0. Client draft remains unsent.

00:37 Budapest, 27 Sep · Cycle 2 closes exact GTM proof through a loaded-resource/live-event trace and recovers Meta Test Events using the existing agency business context. Exact Processed appointment_cta_click matches the 22:32:49 UTC browser event ID. Deduplication still says Still Parsing Your Data. Fresh Sheet test awaits Matt’s notified-booking exception. 15/18, no new production change or spend.

Draft email to SanyiVerified facts only · unsent

This draft records the current verified state and remaining gaps. It remains unsent.

Tárgy: A Lovasember hirdetéseinek és időpontfoglalásának ellenőrzése Szia Sanyi! A kampányt két meglévő képre és két rögzített szövegváltozatra szűkítettük. Ezek a két jelenlegi célközönségnél összesen négy hirdetésben szerepelnek. A többi tizenkét hirdetést szüneteltettük, a napi beállított keret 4 000 Ft maradt. Az új szövegekben feltüntettük az 5 000 Ft-os online konzultációt. A képek kiválasztásához a korábbi forgalmi adatokat használtuk. A korábbi jelentkezéseket nem lehetett egyértelműen az egyes hirdetésekhez vagy szövegekhez kapcsolni, ezért ezekből nem állítunk fel jelentkezési vagy ügyfélminőségi rangsort. Javítottuk és élesítettük az időpontfoglaló gomb mérését érintő kattintási hibát. Az oldal alsó GYIK-részében és keresőleírásában az ingyenes konzultációra utaló szöveget egységesítettük az 5 000 Ft-os online ajánlattal. A GYIK jelentkezési gombja most a foglalási részhez vezet. A friss Tag Assistant ellenőrzésben a foglalógomb kattintása megjelenik a megfelelő Google Analytics-mérésben. Egy korábbi, ténylegesen kifizetett konzultáció Stripe-bizonylatát összevetettük a jelentkezési táblázattal. Az adatok egyetlen eredeti konzultációs sorhoz illeszkednek. Ez egy augusztusi foglalás ellenőrzése, ezért önmagában nem igazolja a jelenlegi teljes adatátvitel működését. Új fizetős próbafoglalást nem végeztünk, a táblázatot és az értesítéseket nem módosítottuk. Mind a négy kiválasztott hirdetés ténylegesen megjelent a módosítások után. A szeptember 26-i, 13 óra utáni elkülönített adatokban összesen 1 803 megjelenés és 1 808 Ft költés szerepel. Ezek kiszolgálási adatok, nem új ügyfelek vagy fizetett foglalások. A Meta tesztesemény-felületén a foglalógomb kattintását feldolgozott eseményként is ellenőriztük, a böngészőből küldött esemény pontos azonosítójával. A böngészős és szerveres események duplikációkezeléséről a Meta még nem mutat kész eredményt. Egy új foglalás teljes táblázatba érkezésének ellenőrzése szintén nyitott. A kampány az időpontfoglaló gombra kattintásra optimalizál. Ezt külön kezeljük a befejezett jelentkezéstől és a fizetett foglalástól. Üdv, Mátyás

Evidence and recoveryPrivate receipts, public redacted proof

The orchestrator maintains a private evidence ledger, deliverables list, main prompts, current session identity and rollback manifest. Public criteria link only to redacted evidence. A narrow access gap leaves that criterion pending while other authorized work proceeds.