Parent completion review · 28 September 2026

Completed and verified: 21 / 21

The Excel-first proposal editor is deployed inside both isolated staging portals. The original contract and criterion texts below are preserved. Production has not been released.

Test staging A Test staging B Open all acceptance evidence

What got done

Decisive proof: 56 repository checks, 25 contact test groups, 652 grid/PDF checks with zero failing entries, final deployed screenshots and independent image comparisons. The final app source is e3433f9. All 250 historical PDFs were preserved. Detailed receipts and rollback references remain beside each criterion.

What differed, and verification limits

Optional next directions: copy one prompt

These are alternatives for a future task. Neither has been started by this completion review.

1. Recommended: rehearse one real job with Péter

End goal: Confirm that the verified editor supports Péter's daily proposal workflow with one realistic, sanitized job.

Main deliverables: A filled staging proposal and branded PDF, a ten-minute operator walkthrough, and a short list of concrete usability findings.

  • Success verification: Compare every entered value and the final total against the supplied job source. Reopen the proposal to prove persistence. Capture the full grid, yellow cell-to-PDF feedback and every exported PDF page. Record operator feedback if Péter is available, otherwise clearly label the agent-run rehearsal.
  • Constraints: Recover a supplied job source or ask only for essential missing job data. Use sanitized staging data. Preserve the current Excel grid and agreed prices. Do not invent job facts, add duplicate entry forms, release production or send clients/leads messages. External spend must stay within USD20 total for the new task.
2. If rollout is the priority: promote the verified build

End goal: Put the verified proposal editor into the existing production portal without losing records, historical PDFs or calendar behavior.

Main deliverables: A narrow Cloudflare production release of the verified app, a verified backup and rollback route, deployed version receipts, and a concise release report.

  • Success verification: First verify the actual production account, bindings and current version. Preserve and compare records and historical PDF hashes. Verify portal navigation, shared customer tabs, autosave/reopen, workbook totals, live highlights and exact preview/download PDF agreement on production-safe fixtures. Capture desktop and mobile screenshots of the deployed result. Roll back if a material regression appears.
  • Constraints: This pasted prompt authorizes only the narrow production promotion and its reversible checks. Preserve prices, unrelated changes and existing calendar behavior. Do not copy synthetic staging records into production, add a security layer, change existing access policy, make client commitments or send clients/leads messages. External spend must stay within USD20 total for the new task.
Kertújítók · LLL contract

Verification and Constraints

COMPLETE

Excel-first proposals inside the existing client portal

Revision 1527 September 202621 / 21 implementation criteria verified

21 of 21 original criteria verified. All six deliverables are complete on both isolated Cloudflare staging copies. Evidence, independent visual comparisons and rollback references are linked beside each checked criterion.

Goal and finish line

End goal: Let Péter work in his familiar Excel-style table inside the existing client portal while edits save automatically and update a clear, branded PDF beside it.

Done means: One B-based proposal editor uses the existing portal header and hamburger, gives the spreadsheet most of the screen, keeps all proposal and settlement entry in the grid, saves edits automatically, and shows yellow cell-to-PDF impact highlights. The selected tabbed customer panel works across the existing portal. Every new PDF page uses the same smaller logo and green branding. Verified staging links, preserved data and rollback evidence accompany the result.

21 / 21criteria verified
6deliverables
Completecurrent step

Rules that stay fixed

  • Preserve source workbook/business meaning, current records, archived PDFs and live calendar work.
  • All proposal and settlement data entry stays in the table. No new product hierarchy or workflow.
  • No new access gate. Bare URLs with non-sensitive review content and existing isolated synthetic staging only.

Selected B proposal interface · Chosen tabbed-panel reference

Proof standard: Every GUI criterion requires an actual screenshot after its named test on the changed deployed interface, a fresh independent gpt-6-luna high description receiving the image without the expected result, and the executor’s comparison. Attach all three beside the criterion. GUI images supplement persistence, timing, arithmetic, PDF-byte and provider proof. Missing evidence stays unchecked. Use the required Luna section QA skill for changed deployed app surfaces, with genuine gpt-6-luna xhigh desktop/mobile workers and fresh verification after any repairs.

Keep the function, simplify the interface

The original workbook already contains settlement columns and paid/debt totals. The separate web accordions are unnecessary. Dated payments and requested amounts remain useful, with input moved into the grid.

DecisionElementReason
KeepQuote, Pótmunka and settlement arithmeticThese belong together in the original Excel layout.
Keep, move into the tableRequested amounts and dated paymentsProposed retention based on historical Q10 and settlement acceptance: keep requested/received amounts and dated payment history in table-native cells, retire duplicate accordions. This placement is not a newly confirmed user choice.
KeepLive PDF and yellow affected-region highlightsExplicitly retained by Matt.
ReplaceManual Save and tall dirty-state stripAutomatic server persistence with a compact truthful status.
RemoveDemo pages, test banner, collapse button and duplicate panelsNo separate design playground in the working portal. Close remains.
SimplifySummary, editor header and PDF controlsKeep useful amounts and export choices while giving the table its space back.
Apply globallyTabbed customer panelLatest specific annotation chooses the tabbed layout. The standalone demos are retired.

End goal system

Confirmed requirementProposed definitionPending step

Green describes the agreed target, not completed implementation.

flowchart TD P["Existing client portal
Same header and hamburger"]:::confirmed --> T["Excel-style table
Proposal and settlement input"]:::confirmed P --> C["Tabbed customer panel
Shared across portal views"]:::confirmed T --> S["Automatic saving
Saved after server receipt"]:::confirmed C --> S T --> V["Live PDF preview
Small logo, green branding"]:::confirmed T -. Hover or focus .-> H["Yellow impact highlights
Show changed text and totals"]:::confirmed H --> V V --> X["Explicit PDF download
Keep previous PDF versions"]:::confirmed S -. Network or conflicting edit .-> R["Recover the edit
Keep input, show the issue"]:::confirmed classDef confirmed fill:#E8F5E9,stroke:#2E7D32,color:#173B25 classDef assumption fill:#FCE4EC,stroke:#C2185B,color:#5B1430 classDef open fill:#FFFFFF,stroke:#94A3B8,color:#475569,stroke-dasharray:5 5

Route to the end state

flowchart LR E["Original Excel
Verified source"]:::confirmed --> RV["Contract read
21 binding criteria"]:::open B["B and portal
Working base"]:::confirmed --> RV RV --> L["Execution authorized
Astra Medium"]:::open L --> I["Integrate
Grid and panel"]:::open I --> Q["Verify
Edits and PDFs"]:::open Q -->|Pass| D["Staging ready
Proof and URLs"]:::open Q -->|Mismatch| F["Fix and retest
Repeat the proof"]:::open F --> Q classDef confirmed fill:#E8F5E9,stroke:#2E7D32,color:#173B25 classDef assumption fill:#FCE4EC,stroke:#C2185B,color:#5B1430 classDef open fill:#FFFFFF,stroke:#94A3B8,color:#475569,stroke-dasharray:5 5

Confirmed, proposed and open

Confirmed
  • Variant B is the proposal basis. Existing portal header and hamburger must be reused.
  • The original Excel is the main visual and interaction authority. Quote and settlements stay together.
  • Real-time edits, durable autosave and yellow PDF impact highlighting are required.
  • The latest specific annotation selects the tabbed customer panel globally. Standalone prototypes are unnecessary.
  • Original workbook A1:P37 and SHA256 have been verified directly. Separate web settlement panels are an implementation choice, not source structure.
Proposed definitions
  • Global rollout covers every contact-panel entrypoint in the existing Cloudflare staging portal. Production promotion is a separate later release, preserving the prior boundary.
  • Compact screen geometry intentionally adapts the workbook’s tall print rows and font sizes while preserving its logical grid and familiar styling. The viewport, visibility, readable-text, toolbar-height and latency numbers below are proposed acceptance limits, not verbatim user numbers or proven fit.
  • Keep dated payments, requested-versus-received distinction, draft/final exports and historical PDF integrity. Simplify their UI instead of deleting useful accepted behavior.
Review and discovery
  • No unresolved acceptance items. Physical-phone testing and production promotion were outside this staging task.

Work and proof

Open each deliverable for acceptance tests

Unify the client portal

Make proposals and customer details feel like one familiar portal.

3 / 3
Criteria, constraints and evidence
Required proof

Desktop/mobile route journey and screenshots of portal, menu, list and editor. Compare to the existing shell.

PASS A/B. Existing portal menu, correct Ajánlatok destination, direct saved editor and return navigation pass on desktop and390px for both copies. Independent54desktop and20mobile image observations were compared to the actual route, focus, identity and provider receipts. Final664desktop/150mobile section screenshots found no confirmed remaining defect. Caption/crop limitations are explicitly reconciled in the executor comparison.
Actual test images (24)
PASSED · Existing portal menu, correct Ajánlatok destination, direct saved editor and return navigation pass on desktop and390px for both copies.

Deployed contact, persistence and timing results

54 independent desktop-route observations

20 independent mobile observations

Executor final integration comparison

Final desktop section review

Final mobile review and model-receipt reconciliation

Original complete mobile advisory report

Actual pointer and close-focus checks

Required proof

Six entrypoint journeys, tab screenshots and server readback for each field group. Restore/archive only test-owned records.

PASS A/B. Independent final-route observations cover all six entrypoints and three selected tabs. Final read-only captures sample different synthetic records; same-record equality is established by the separate controlled six-entrypoint test and original screenshots. This distinction and caption title omissions are explicit in the executor comparison.
Actual test images (72)
Final shared-panel route and tab 8Final shared-panel route and tab 8Final shared-panel route and tab 9Final shared-panel route and tab 9Final shared-panel route and tab 10Final shared-panel route and tab 10Final shared-panel route and tab 11Final shared-panel route and tab 11Final shared-panel route and tab 12Final shared-panel route and tab 12Final shared-panel route and tab 13Final shared-panel route and tab 13Final shared-panel route and tab 14Final shared-panel route and tab 14Final shared-panel route and tab 15Final shared-panel route and tab 15Final shared-panel route and tab 16Final shared-panel route and tab 16Final shared-panel route and tab 17Final shared-panel route and tab 17Final shared-panel route and tab 18Final shared-panel route and tab 18Final shared-panel route and tab 19Final shared-panel route and tab 19Final shared-panel route and tab 20Final shared-panel route and tab 20Final shared-panel route and tab 21Final shared-panel route and tab 21Final shared-panel route and tab 22Final shared-panel route and tab 22Final shared-panel route and tab 23Final shared-panel route and tab 23Final shared-panel route and tab 24Final shared-panel route and tab 24Final shared-panel route and tab 25Final shared-panel route and tab 25Final shared-panel route and tab 35Final shared-panel route and tab 35Final shared-panel route and tab 36Final shared-panel route and tab 36Final shared-panel route and tab 37Final shared-panel route and tab 37Final shared-panel route and tab 38Final shared-panel route and tab 38Final shared-panel route and tab 39Final shared-panel route and tab 39Final shared-panel route and tab 40Final shared-panel route and tab 40Final shared-panel route and tab 41Final shared-panel route and tab 41Final shared-panel route and tab 42Final shared-panel route and tab 42Final shared-panel route and tab 43Final shared-panel route and tab 43Final shared-panel route and tab 44Final shared-panel route and tab 44Final shared-panel route and tab 45Final shared-panel route and tab 45Final shared-panel route and tab 46Final shared-panel route and tab 46Final shared-panel route and tab 47Final shared-panel route and tab 47Final shared-panel route and tab 48Final shared-panel route and tab 48Final shared-panel route and tab 49Final shared-panel route and tab 49Final shared-panel route and tab 50Final shared-panel route and tab 50Final shared-panel route and tab 51Final shared-panel route and tab 51Final shared-panel route and tab 52Final shared-panel route and tab 52Controlled same-record entrypoint 1Controlled same-record entrypoint 1Controlled same-record entrypoint 2Controlled same-record entrypoint 2Controlled same-record entrypoint 3Controlled same-record entrypoint 3Controlled same-record entrypoint 4Controlled same-record entrypoint 4Controlled same-record entrypoint 5Controlled same-record entrypoint 5Controlled same-record entrypoint 6Controlled same-record entrypoint 6Controlled same-record entrypoint 7Controlled same-record entrypoint 7Controlled same-record entrypoint 8Controlled same-record entrypoint 8Controlled same-record entrypoint 9Controlled same-record entrypoint 9Controlled same-record entrypoint 10Controlled same-record entrypoint 10Controlled same-record entrypoint 11Controlled same-record entrypoint 11Controlled same-record entrypoint 12Controlled same-record entrypoint 12Controlled same-record entrypoint 13Controlled same-record entrypoint 13Controlled same-record entrypoint 14Controlled same-record entrypoint 14Controlled same-record entrypoint 15Controlled same-record entrypoint 15Controlled same-record entrypoint 16Controlled same-record entrypoint 16Controlled same-record entrypoint 17Controlled same-record entrypoint 17Controlled same-record entrypoint 18Controlled same-record entrypoint 18Controlled same-record entrypoint 56Controlled same-record entrypoint 56Controlled same-record entrypoint 57Controlled same-record entrypoint 57Controlled same-record entrypoint 58Controlled same-record entrypoint 58Controlled same-record entrypoint 59Controlled same-record entrypoint 59Controlled same-record entrypoint 60Controlled same-record entrypoint 60Controlled same-record entrypoint 61Controlled same-record entrypoint 61Controlled same-record entrypoint 62Controlled same-record entrypoint 62Controlled same-record entrypoint 63Controlled same-record entrypoint 63Controlled same-record entrypoint 64Controlled same-record entrypoint 64Controlled same-record entrypoint 65Controlled same-record entrypoint 65Controlled same-record entrypoint 66Controlled same-record entrypoint 66Controlled same-record entrypoint 67Controlled same-record entrypoint 67Controlled same-record entrypoint 68Controlled same-record entrypoint 68Controlled same-record entrypoint 69Controlled same-record entrypoint 69Controlled same-record entrypoint 70Controlled same-record entrypoint 70Controlled same-record entrypoint 71Controlled same-record entrypoint 71Controlled same-record entrypoint 72Controlled same-record entrypoint 72Controlled same-record entrypoint 73Controlled same-record entrypoint 73
PASSED · All six entrypoints expose the shared Kapcsolat/Jegyzet/Dátumok panel. Controlled same-record regression and exact saved-field API readbacks pass on A/B. Fresh final section images independently confirm the shared panel on every route.

Deployed contact, persistence and timing results

54 independent final-route observations

Executor route comparison and same-record proof distinction

Unchanged final contact/sidebar behavior

664-image final desktop section review

150-image final mobile section review

Required proof

Open/close screenshots, keyboard focus and background-interaction checks. The two old demo URLs redirect to the real portal contact/proposal entry instead of a demo.

PASS A/B. Close/reopen restores working space and focus, background pointer/menu access remains usable, and retired demo redirects pass on desktop/mobile for both copies. No collapse/expand control remains. Independent54desktop and20mobile image observations were compared to the actual route, focus, identity and provider receipts. Final664desktop/150mobile section screenshots found no confirmed remaining defect. Caption/crop limitations are explicitly reconciled in the executor comparison.
Actual test images (24)
PASSED · Close/reopen restores working space and focus, background pointer/menu access remains usable, and retired demo redirects pass on desktop/mobile for both copies. No collapse/expand control remains.

Deployed contact, persistence and timing results

54 independent desktop-route observations

20 independent mobile observations

Executor final integration comparison

Final desktop section review

Final mobile review and model-receipt reconciliation

Original complete mobile advisory report

Actual pointer and close-focus checks

Constraints

  • Use the existing portal header, hamburger, logo and navigation behavior. Do not invent a separate application shell or new portal product.
  • Preserve existing CRM/calendar fields, original enquiry, own note, reminders, work-period dates, calendar visibility, archive behavior, source metadata and linked-proposal identity.
  • The tabbed prototype supplies layout only. The deployed panel must use actual shared CRM persistence. Retire demo navigation/routes without deleting saved customer data.

Restore the Excel working surface

Péter enters proposal and settlement data in one familiar grid.

4 / 4
Criteria, constraints and evidence
Required proof

Source workbook hash and cell/style map, original reference render, actual editor screenshot and side-by-side differences. Any intentional extension is labelled in the comparison. Include a compact source-row/column mapping and disposition of every difference. The ambiguous source M5 180,000 Ft stays unmapped, is not treated as a payment and is not copied into new proposals.

PASS A/B. All source rows/columns accounted for, 34 source rows visible, A:P preserved, readable headings and adjacent PDF. Independent image observations and explicit source/extension comparison linked.
Actual test images (4)
PASSED · Final original-workbook comparison and measured desktop geometry verified on both deployed copies.

Original workbook and compact screen comparison

Final-build grid, mapping, PDF and arithmetic results

14 independent final-layout image observations

Executor source/layout comparison

Required proof

Actual viewport screenshots and measured table/workspace rectangles, source-row coverage, font sizes and PDF width, plus source-to-screen comparison for readable labels, values and wrapping. Compact geometry may reduce print whitespace, not omit required content. These proposed near-full-fit limits are not verified feasibility claims. Failure remains unchecked with measured tradeoff; no silent waiver or additional design-approval phase.

PASS A/B. All source rows/columns accounted for, 34 source rows visible, A:P preserved, readable headings and adjacent PDF. Independent image observations and explicit source/extension comparison linked.
Actual test images (4)
PASSED · Final original-workbook comparison and measured desktop geometry verified on both deployed copies.

Original workbook and compact screen comparison

Measured final-build geometry

Final-build grid, mapping, PDF and arithmetic results

Final CSS source and both deployed Worker readback

14 independent final-layout image observations

Executor source/layout comparison

Required proof

Actual table-input journey, screenshots after each meaningful state and model/API checks. Payment dates/amounts and any already-existing notes remain accessible in the grid. Stage4/5 preservation is grounded in historical C14, not new scope.

PASS A/B. Quantity, labor, transport, material, extra-work, stages4/5, requested amount and two dated payments are visibly entered in grid cells. Selected-cell states and resulting grid/PDF amounts correspond to exact server/model assertions; no separate settlement editor is needed. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (28)
PASSED · Quantity, labor, transport, material, extra-work, stages4/5, requested amount and two dated payments are visibly entered in grid cells. Selected-cell states and resulting grid/PDF amounts correspond to exact server/model assertions; no separate settlement editor is needed.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Required proof

Keyboard/paste interaction recording, resulting grid screenshot and exact saved values. Preserve read-only calculated cells and Hungarian numeric input.

PASS A/B. Two-row Hungarian numeric paste and keyboard navigation preserve the intended cells and protected calculations. Images011/111 show the selected pasted range with comma-decimal values; exact focus/value assertions and reopened grid checks pass. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (4)
PASSED · Two-row Hungarian numeric paste and keyboard navigation preserve the intended cells and protected calculations. Images011/111 show the selected pasted range with comma-decimal values; exact focus/value assertions and reopened grid checks pass.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Constraints

  • Use the actual Munka1 A1:P37 workbook and original visual reference, not the current distorted web grid as the authority. Preserve column relationships, grouping, terms, Pótmunka and calculation meaning. Preserve logical order, labels, groupings, formulas and familiar styling, while explicitly adapting print-oriented row heights and font sizes to compact screen geometry. This is not literal pixel parity.
  • Keep all proposal/settlement input in table cells. Remove separate settlement accordions and duplicate forms. Keep requested versus received amounts and dated-payment history through compact table-native rows/cells. Preserve existing payment notes if present, but do not add a new required note field. The retained dated-payment UI is a proposal grounded in historical Q10.
  • Do not hide required columns/rows or reduce text to unreadable size to claim a fit. No Excel/Sheets embed, external spreadsheet dependency, project hierarchy, invoicing or added business workflow.

Save edits as they happen

Typing updates durable data and the PDF without a manual-save routine.

4 / 4
Criteria, constraints and evidence
Required proof

Repeat each edit class three times, record every preview/ack latency on a documented online, unthrottled test connection, and attach exact API readback plus fresh-session screenshots. Use a distinctive internal-note sentinel and assert absence in extracted preview/download PDF text. Timing applies to editable supported fields, not a fabricated edit to read-only metadata; failures remain visible rather than averaged away.

PASS on both staging copies. Independent Luna High observations of contact images 036-039 and 091-094 identify the latest edited totals, profile fields, notes, dates, reopened editor and exported PDF state. Root comparison matches these visible states to the exact fresh-session/API readbacks. All 120 final-build timing observations passed: acknowledgement <=991.3 ms and relevant PDF preview <=600.6 ms. The internal-note sentinel is absent from extracted preview/download text. Screenshots alone do not prove latency or backend persistence; the linked controlled test receipts supply that evidence. No unresolved visual contradiction affects this criterion.
Actual test images (8)
PASSED · Verified on A and B: 120/120 final-build timing observations, fresh-session/API persistence, private-note exclusion, and independent screenshot comparison.

Deployed contact, persistence and timing results

Independent per-image observations compared with saved-state receipts

Required proof

Controlled delayed-response test, server readback for both record IDs and reopened screenshots.

PASS A/B. Fresh independent image-only observations of frames001-002 and010-011 identify the correct first and second record titles, saved status and explicitly pending PDF render labels. Those captures are transitional PDF states, not claimed finished previews. Exact delayed-response server readbacks and reopened latest input values establish that continued edits remain with their correct record and late saves cannot overwrite the second record. Root compared observations with controlled two-record receipt and found no contradiction.
Actual test images (4)
PASSED · PASS A/B. Fresh independent image-only observations of frames001-002 and010-011 identify the correct first and second record titles, saved status and explicitly pending PDF render labels. Those captures are transitional PDF states, not claimed finished previews. Exact delayed-response server readbacks and reopened latest input values establish that continued edits remain with their correct record and late saves cannot overwrite the second record. Root compared observations with controlled two-record receipt and found no contradiction.

Deployed contact, persistence and timing results

Fresh individual image observations of delayed, recovered and invalid states

Independent pending and contact-conflict observations

Required proof

Offline/reload/reconnect test with server readback and screenshots for pending and saved states. No data loss or false Mentve.

PASS A/B. Independent observations of original images044-045/099-100 describe Failed to fetch, retained local drafts and retry state. Fresh independent frames003/012 show recovered OFFLINE-contact notes under the correct records with Mentve, while PDF refresh is separately pending. Exact service-worker offline reload/reconnect tests and fresh server/session readback establish durable recovery. Root rejected prior mismatched image101 descriptions and uses the new filename-grounded observation instead.
Actual test images (6)
PASSED · PASS A/B. Independent observations of original images044-045/099-100 describe Failed to fetch, retained local drafts and retry state. Fresh independent frames003/012 show recovered OFFLINE-contact notes under the correct records with Mentve, while PDF refresh is separately pending. Exact service-worker offline reload/reconnect tests and fresh server/session readback establish durable recovery. Root rejected prior mismatched image101 descriptions and uses the new filename-grounded observation instead.

Deployed contact, persistence and timing results

Fresh individual image observations of delayed, recovered and invalid states

Independent pending and contact-conflict observations

Required proof

Two-session conflict and invalid-input tests, field-level comparison/readback and actual recovery screenshots. Normal use must not require clicking a Save button. Exercise an actual choose/retry/correct path and verify resolved values through a new session and server readback.

PASS A/B. Fresh independent frames004-009 and013-018 describe both conflicting values/choice controls, resolved saved state, invalid retained date/quantity, selected blank with unknown calculations, and explicit zero with known calculations. Accepted original images102-104 show the contact-field conflict and its selected resolution. Exact two-session different-field merge, both choose paths, correction and fresh-session readbacks supply the persistence evidence. Root comparison distinguishes deliberately induced error states from defects; no silent overwrite or blank-to-zero conversion is observed.
Actual test images (15)
PASSED · PASS A/B. Fresh independent frames004-009 and013-018 describe both conflicting values/choice controls, resolved saved state, invalid retained date/quantity, selected blank with unknown calculations, and explicit zero with known calculations. Accepted original images102-104 show the contact-field conflict and its selected resolution. Exact two-session different-field merge, both choose paths, correction and fresh-session readbacks supply the persistence evidence. Root comparison distinguishes deliberately induced error states from defects; no silent overwrite or blank-to-zero conversion is observed.

Deployed contact, persistence and timing results

Fresh individual image observations of delayed, recovered and invalid states

Independent pending and contact-conflict observations

Constraints

  • Server acknowledgement is required before showing Mentve. Local recovery drafts must not be labelled saved. Preserve valid intermediate typing and never silently turn blank/unknown into zero.
  • Keep proposal/contact identities, versions, server-owned baselines and document history. Serialize saves per record, retain newer edits during in-flight saves and prevent cross-contact or cross-proposal writes.
  • Autosave applies to proposal cells and the shared customer panel throughout the portal. Do not expand this into a rewrite of unrelated blog or marketing settings. Autosaving data does not create/export/send a final PDF.

Keep PDF feedback immediate

Retain the useful yellow mapping with a quieter preview toolbar.

4 / 4
Criteria, constraints and evidence
Required proof

Mapping assertions and actual screenshots of input cell plus every affected PDF region. Test standard and landscape with a multipage fixture.

PASS A/B. Both standard and landscape mapping journeys cover quantity, unit rate, transport, customer, terms, settlement and payment inputs, every affected page, hover/focus, pin and Escape clear. Independent observations identify selected inputs and yellow/gold mapped regions. Exact region keys, pages and pin/clear assertions establish correspondence beyond thumbnail-scale readability. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (92)
PDF test image 035PDF test image 035PDF test image 036PDF test image 036PDF test image 037PDF test image 037PDF test image 038PDF test image 038PDF test image 039PDF test image 039PDF test image 040PDF test image 040PDF test image 041PDF test image 041PDF test image 042PDF test image 042PDF test image 043PDF test image 043PDF test image 044PDF test image 044PDF test image 045PDF test image 045PDF test image 046PDF test image 046PDF test image 047PDF test image 047PDF test image 048PDF test image 048PDF test image 049PDF test image 049PDF test image 050PDF test image 050PDF test image 051PDF test image 051PDF test image 052PDF test image 052PDF test image 053PDF test image 053PDF test image 054PDF test image 054PDF test image 055PDF test image 055PDF test image 056PDF test image 056PDF test image 057PDF test image 057PDF test image 059PDF test image 059PDF test image 060PDF test image 060PDF test image 061PDF test image 061PDF test image 062PDF test image 062PDF test image 063PDF test image 063PDF test image 064PDF test image 064PDF test image 065PDF test image 065PDF test image 066PDF test image 066PDF test image 067PDF test image 067PDF test image 068PDF test image 068PDF test image 069PDF test image 069PDF test image 070PDF test image 070PDF test image 071PDF test image 071PDF test image 072PDF test image 072PDF test image 073PDF test image 073PDF test image 074PDF test image 074PDF test image 075PDF test image 075PDF test image 076PDF test image 076PDF test image 077PDF test image 077PDF test image 078PDF test image 078PDF test image 079PDF test image 079PDF test image 080PDF test image 080PDF test image 081PDF test image 081PDF test image 135PDF test image 135PDF test image 136PDF test image 136PDF test image 137PDF test image 137PDF test image 138PDF test image 138PDF test image 139PDF test image 139PDF test image 140PDF test image 140PDF test image 141PDF test image 141PDF test image 142PDF test image 142PDF test image 143PDF test image 143PDF test image 144PDF test image 144PDF test image 145PDF test image 145PDF test image 146PDF test image 146PDF test image 147PDF test image 147PDF test image 148PDF test image 148PDF test image 149PDF test image 149PDF test image 150PDF test image 150PDF test image 151PDF test image 151PDF test image 152PDF test image 152PDF test image 153PDF test image 153PDF test image 154PDF test image 154PDF test image 155PDF test image 155PDF test image 156PDF test image 156PDF test image 157PDF test image 157PDF test image 159PDF test image 159PDF test image 160PDF test image 160PDF test image 161PDF test image 161PDF test image 162PDF test image 162PDF test image 163PDF test image 163PDF test image 164PDF test image 164PDF test image 165PDF test image 165PDF test image 166PDF test image 166PDF test image 167PDF test image 167PDF test image 168PDF test image 168PDF test image 169PDF test image 169PDF test image 170PDF test image 170PDF test image 171PDF test image 171PDF test image 172PDF test image 172PDF test image 173PDF test image 173PDF test image 174PDF test image 174PDF test image 175PDF test image 175PDF test image 176PDF test image 176PDF test image 177PDF test image 177PDF test image 178PDF test image 178PDF test image 179PDF test image 179PDF test image 180PDF test image 180PDF test image 181PDF test image 181
PASSED · Both standard and landscape mapping journeys cover quantity, unit rate, transport, customer, terms, settlement and payment inputs, every affected page, hover/focus, pin and Escape clear. Independent observations identify selected inputs and yellow/gold mapped regions. Exact region keys, pages and pin/clear assertions establish correspondence beyond thumbnail-scale readability.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Required proof

Controlled stale-render test, revision/region comparison and final screenshot.

PASS A/B. Controlled delayed rendering ends at the newest revision with quantity5 and total195200Ft, retaining the selected mapped region. Images083/183 independently show that final state, and the revision/region assertions reject stale replacement. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (2)
PASSED · Controlled delayed rendering ends at the newest revision with quantity5 and total195200Ft, retaining the selected mapped region. Images083/183 independently show that final state, and the revision/region assertions reject stale replacement.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Required proof

Desktop/mobile screenshots and layout measurements: proposed desktop editor identity row at most 60px, summary at most 36px, default PDF toolbar at most 48px. Necessary errors may expand temporarily.

PASS A/B. Independent final-layout frames001/004/008/011 show compact saved/total identity and aligned PDF controls. Mobile frames006/007/013/014 show compact summary and reachable PDF controls below the horizontally scrollable worksheet. Root comparison matches measured rectangles and actual pointer/expanded-preview receipts. Deliberately captured PDF-refresh states are labelled truthfully, not treated as completed renders.
Actual test images (8)
PASSED · Final compact identity44px, desktop summary28px and PDF toolbar38px pass their60/36/48px limits. Mobile summary fits390px and expanded preview is382px wide. No test banner or proposal-design variant label remains.

Measured final-build geometry

Final-build grid, mapping, PDF and arithmetic results

Final CSS source and both deployed Worker readback

Independent final-layout observations

Executor layout comparison

Actual mobile summary and expanded-preview measurements

Required proof

Pre-click/display/download/server SHA256 equality, unknown-price case and archive hash comparison, plus screenshots.

PASS A/B. All four orientation/draft-final combinations have equal pre-click displayed, downloaded and stored SHA256 values. Visible export banners/statuses and unknown-price disabled-final states are independently described. Earlier exports and all250historical files retain exact hashes, and no preview overlay appears in exports. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (16)
PASSED · All four orientation/draft-final combinations have equal pre-click displayed, downloaded and stored SHA256 values. Visible export banners/statuses and unknown-price disabled-final states are independently described. Earlier exports and all250historical files retain exact hashes, and no preview overlay appears in exports.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Constraints

  • Preserve yellow hover/focus impact marks, click-to-pin, Escape-to-clear and affected-page tracking. Preview overlays must never appear in a downloaded PDF.
  • Keep the existing standard/landscape capability and meaningful draft/final choice, but tuck secondary controls into a compact predictable control area. No second manual data editor beside the table.
  • Only offer a current valid download. Keep first-final-export baseline semantics, immutable historical documents and explicit distinction between saving data and exporting a PDF.

Brand every PDF page consistently

Every page looks like the same Kertújítók document.

2 / 2
Criteria, constraints and evidence
Required proof

All-page PDF renders, sampled palette/logo dimensions and fresh independent image descriptions for every page.

PASS A/B. All34exported stress pages (12standard+5landscape on each variant) contain the same compact logo/header, green table/section palette and complete counters. Fresh descriptions exist for every page. Root resolved missed-logo/cropped-footer descriptions using all34literal header/footer crops: logo pixel hashes are identical and every complete counter is visible. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (35)
PASSED · All34exported stress pages (12standard+5landscape on each variant) contain the same compact logo/header, green table/section palette and complete counters. Fresh descriptions exist for every page. Root resolved missed-logo/cropped-footer descriptions using all34literal header/footer crops: logo pixel hashes are identical and every complete counter is visible.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

All34logo pixel hashes and complete header/footer crops

Required proof

All-page renders, PDF text/content/bounds checks and per-page screenshots.

PASS A/B. All34stress pages preserve long descriptions, accented Hungarian text, five settlements, repeated table headings, totals and footers. Independent page-body observations plus exact text/content/bounds assertions show proper continuation without overlap, clipped document text or missing content. Mid-row continuation of the deliberately oversized description is intentional and complete across pages. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (35)
PASSED · All34stress pages preserve long descriptions, accented Hungarian text, five settlements, repeated table headings, totals and footers. Independent page-body observations plus exact text/content/bounds assertions show proper continuation without overlap, clipped document text or missing content. Mid-row continuation of the deliberately oversized description is intentional and complete across pages.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

All34logo pixel hashes and complete header/footer crops

Constraints

  • Use the existing approved logo and its green shades with neutral text/paper. Preserve yellow only as the editor impact overlay. No unrelated blue palette or fabricated contact/commercial details.
  • Apply the same smaller-logo header and coherent typography to every page, including terms and continuation pages. Preserve source content, Hungarian accents, totals and readable pagination.
  • Regenerate only new previews/exports. Never repaint or overwrite previously stored PDFs.

Deliver one verified integrated experience

Ship the selected B-based proposal design inside the existing staging portal, with proof and recovery.

4 / 4
Criteria, constraints and evidence
Required proof

Independent original workbook comparison and targeted model/API/GUI/PDF tests, plus resulting table/PDF screenshots.

PASS A/B. Original-source fixture shows374500Ft =191000labor+25000transport+158500materials in grid and PDF. Quantity-only edits preserve rates and count transport once; excluded, unknown, blank and zero states remain distinct. Requested/received/debt and dated payments reconcile by exact model/API/PDF assertions. Original M5 remains unmapped and is never copied as a payment. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (20)
PASSED · Original-source fixture shows374500Ft =191000labor+25000transport+158500materials in grid and PDF. Quantity-only edits preserve rates and count transport once; excluded, unknown, blank and zero states remain distinct. Requested/received/debt and dated payments reconcile by exact model/API/PDF assertions. Original M5 remains unmapped and is never copied as a payment.

Original workbook and compact screen comparison

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Required proof

Provider source/version/binding readback, anonymous HTTP checks, six-entrypoint smoke tests, A/B data isolation and desktop/mobile postdeployment screenshots. Deploy shared portal changes to both existing staging copies from their respective current sources. Before public staging deployment, inventory exposed routes/API responses, assets, records and PDFs against the exact staging-only bindings and verify all exposed customer data/documents are synthetic, using provider/data readback and anonymous samples. This preserves the historical synthetic-data boundary without adding an access gate. B is the sole proposal design basis, not a new A/B comparison: apply required shared integration to both existing staging copies while retaining each current calendar delta.

PASS A/B. Both final anonymous staging copies pass source/version/binding,64route/assets, synthetic-data and isolation checks. Fresh desktop/mobile images confirm integrated B-based proposals, shared contacts and preserved distinct calendars. Independent54desktop and20mobile image observations were compared to the actual route, focus, identity and provider receipts. Final664desktop/150mobile section screenshots found no confirmed remaining defect. Caption/crop limitations are explicitly reconciled in the executor comparison.
Actual test images (10)
PASSED · Both final anonymous staging copies pass source/version/binding,64route/assets, synthetic-data and isolation checks. Fresh desktop/mobile images confirm integrated B-based proposals, shared contacts and preserved distinct calendars.

Provider versions, isolated bindings and64anonymous checks

Original record and250historical PDF preservation

Executed staging rollback and return rehearsal

Final CSS source and both deployed Worker readback

54 independent desktop-route observations

20 independent mobile observations

Executor final integration comparison

Final desktop section review

Final mobile review and model-receipt reconciliation

Original complete mobile advisory report

Actual pointer and close-focus checks

Required proof

Narrow backups and restore rehearsal, original-record and PDF hashes, changelog, scoped diff and completed evidence board. No completion claim for missing evidence.

PASSED · 26 original tables and all250historical PDFs preserved, with4documented reversible synthetic-identity substitutions. Exact restore and Worker rollback/return rehearsal, final source commit, preserved calendar deltas, deployed URLs and criterion evidence are included in the completed report.

Provider versions, isolated bindings and64anonymous checks

Original record and250historical PDF preservation

Executed staging rollback and return rehearsal

Complete outcome, source, recovery and21-criterion report

Scoped source delta and documentation checkpoint

Verified immediate rollback version availability

Original-contract and complete evidence audit

Required proof

Existing historical C05/C06/C09/C14 regression journey: exact record IDs/counts, model/API arithmetic and payment readback, fresh reload and actual grid screenshots. Use existing supported actions, no new workflow.

PASS A/B. Lost-create-response retry produces exactly one linked contact/proposal, and reopening retains its ID. Regular and extra-work rows can be added/removed with stable IDs and totals; stages4/5 and unrelated dated-payment history survive. Independent create-error/recovered/editor images supplement exact IDs/counts and saved arithmetic checks. Exact652-check A/B receipt passed on unchanged final JavaScript/PDF bytes. Fresh independent per-image observations are linked; executor resolved fine-print/crop caption errors against source renders and machine receipts in the comparison note.
Actual test images (16)
PASSED · Lost-create-response retry produces exactly one linked contact/proposal, and reopening retains its ID. Regular and extra-work rows can be added/removed with stable IDs and totals; stages4/5 and unrelated dated-payment history survive. Independent create-error/recovered/editor images supplement exact IDs/counts and saved arithmetic checks.

Final-build grid, mapping, PDF and arithmetic results

200 fresh individual blind image observations

Executor comparison and explicit caption corrections

Unchanged functional/PDF source in final deployment

Constraints

  • Apply the shared tabbed customer panel globally to portal entrypoints. Preserve each staging variant’s calendar behavior/data, including the newest A quick-entry refinement, by coordinating current source and deployment ownership. Do not deploy an old snapshot over concurrent changes.
  • No production release, GAJA/GHL access, client sending, irreversible deletion, new security layer or commercial commitment. Preserve existing data/terms/amounts and the user’s unsaved browser tab. Bare review/staging URLs only, public artifacts contain no credentials or private source data.
  • Existing account, D1 and R2 bindings remain the authority. Fresh readback and a verified narrow rollback are required. Do not reset databases or regenerate unrelated marketing assets.

Execution authority

Authorized: implement, test and reversibly deploy both existing synthetic staging copies. USD15 aggregate external ceiling. No production release, client sends, new commitments, irreversible deletion or added security layer.

Complete: all21original criteria passed with linked proof. Source, rollback and preservation records are saved. No original-parent message was sent.

Parent ↔ child messages

Meaningful handoffs only
PARENT · Contract preparation27 September 2026

The latest feedback supersedes the standalone prototype/demo and manual-save design. Existing data and useful calculation behavior stay. The header-only draft has not been deployed.

21 of 21 original criteria verified. All six deliverables are complete on both isolated Cloudflare staging copies. Evidence, independent visual comparisons and rollback references are linked beside each checked criterion.

Parent read set and recoveryExecution records

Executor maintains a deliverable list, main-prompts list, session identity and this board. The completion report is saved locally. Matt’s latest instruction prohibits sending a message to the original parent.

Recovery state: exact SQL and asset backups passed restoration. Four identified staging identity rows were reversibly sanitized, with every other field and 250 historical PDFs preserved. 21 of 21 criteria currently verified. See each criterion for its evidence and comparison.