Metadata Plan for the Move
Tab & Shift Engineering Ltd. · Clean, organize and restructure in one pass
| Prepared for | Prepared by | Date | Status |
|---|---|---|---|
| Carl Shift Co-founder and Principal Engineer | Floor 16 Technologies | Oct 6, 2026 | v1.1 · For sign-off |
Summary
Every file is tagged as it moves, from what its folder, its name and its content already say. Each tag carries a confidence score, and owners see only the files the rules are unsure about, grouped so one click confirms a whole folder. Nobody tags 410,000 files by hand.
The figures come from a trial run on a copy of the three pilot projects, 14,000 files, from September 28 to October 2. The thresholds in this plan were set from that run, not guessed.
One pass, three jobs
Cleaning, tagging and restructuring usually happen as three projects, months apart, and the second and third rarely happen at all. Here they happen to each file in the same pass, on the night it moves.
| Job | What happens to each file | Files |
|---|---|---|
| Clean | Copies and old revisions found. The newest becomes the current file, earlier ones become its version history or are marked Superseded. FINAL, v3 and Copy of come off names, and dates in names move to a column. | 38,000 |
| Organize | Six tags applied from the folder path, the file name, the content and the project register, each with a confidence score. | 410,000 |
| Restructure | The file lands in the project site and library its tags point to, at most two folders deep, instead of where the old folder happened to put it. | 410,000 |
The six tags
Each tag answers one question. Signals are listed strongest first; when the strongest is missing, the next one is used and the confidence drops.
| Tag | Answers | Values | Read from, strongest first |
|---|---|---|---|
| Project | Which project is this? | 142 live projects in the register | Project number in the path or name, matched to the register. Then a contract or purchase order number in the content |
| Client | Who is it for? | 38 clients in the register | Always from the project's register entry. Never guessed from the file alone |
| Discipline | Which discipline owns it? | CV, ST, ME, EL, PR, GN | Code in the name, the title block, CAD layer prefixes, then the folder name |
| Document type | What kind of thing is it? | 24 types, from Drawing to Site Photo | First-page heading or title block, keywords in the name, the extension, then the folder |
| Status | Is it current? | Draft, Final, Issued for Construction, As-Built, Superseded | A stamp or watermark, a signature on a PDF, a newer revision beside it, then the word draft in the name |
| Year | When is it from? | 2009 to 2026 | The date in the content, then a date in the name, then the date last modified as a last resort |
A blank is better than a wrong tag. If Status or Year cannot be read with confidence, it is left empty and the file appears in a Needs status view, rather than being guessed. Project, Document type and Discipline decide where a file lands, so those three are never left empty: a file unsure of any of them goes to its owner.
How a tag is decided
- Agreement raises it. A project number in the path that matches the register, in a drawing whose title block names the same project, scores 0.98.
- Disagreement lowers it. If the path says P-2214 and the title block says P-2231, the file scores below 0.70 and goes to review with both shown.
- Weak signals are capped. A Year taken only from the date last modified never scores above 0.75, because copying a file changes that date.
- Client inherits. Client is read from the register entry for the project, so its confidence is the project's confidence.
Examples from the trial run
| File as found | Tags derived | Confidence | Why |
|---|---|---|---|
| Projects\2214 Hwy 9\Dwgs\Civil\C-101 rev C.pdfBecomes P-2214-CV-DWG-101, Rev C | 0.97Applied | Project number in the path matches the register. The title block reads Issued for Construction, Rev C, June 2025. | |
| Projects\2214 Hwy 9\Dwgs\OLD\C-101 rev A.pdfBecomes an earlier version of P-2214-CV-DWG-101 | 0.96Applied | Rev C of the same drawing exists. Kept in version history and hidden from current views. | |
| Projects\2214 Hwy 9\Specs\03 30 00 Cast-in-place Concrete.docxBecomes P-2214-ST-SPC-033 | 0.94Applied | The section number format marks a specification. A DRAFT watermark is in the header. | |
| Projects\2231 Pump Stn\Calcs\pump sizing FINAL v3.xlsxBecomes P-2231-ME-CAL-007, with v1 and v2 as versions | 0.93Applied | Calcs folder, and the sheet's Checked by field is filled in. Two older copies sit beside it. | |
| Templates\Copy of Copy of transmittal.dotxBecomes Transmittal.dotx in Tab & Shift Hub › Templates | 0.91Applied | Identical to 14 other copies across the server. One kept, 14 removed. | |
| Proposals\2026\Autosave Hwy 2 proposal draft.docxGoes to Business Development › Proposals | 0.88Sampled | Client name in the file name matches the client list. Tracked changes still open. No project exists yet, so it is not filed under one. | |
| Admin\Scans\SCAN_0041.pdfBecomes 2025-03-03 Progress Meeting 07 in P-2240 › 04 Meetings | 0.81Sampled | The name says nothing. The text reads Minutes of Meeting No. 7, Northgate Warehouse, March 3, 2025, and the project name matches the register. | |
| Users\mhall\Desktop backup\New folder\IMG_4471.jpgGoes to Sam Okafor's review queue | 0.58Review | The photo's date and location place it at the pump station site, but nothing in the path or name says so. | |
| Projects\1488 Riverbend Bridge\Final Report.pdfGoes to the archive, not the move | 0.95Archived | The project closed in the register in 2019. Tagged the same way so the archive is searchable too. |
Confidence across the move
Projected from the trial run onto all 410,000 files. A file's confidence is the lowest of its three placement tags: Project, Document type and Discipline.
When the rules are unsure
The 7% go to an owner review queue. It is built so a division lead clears it in an afternoon, not a month.
| Rule | How it works |
|---|---|
| Who reviews | The person who edited the source folder most, from the Scan. If they have left, the division lead. |
| Grouped by folder | Files are grouped by the folder they came from, with suggested tags filled in. Confirm a group, change one tag for the whole group, or pass it on with Not mine. |
| When | The queue opens ten working days before the division's wave and should be under 2% by two days before. That is pre-check 5 in the wave plan. |
| If nobody answers | The file still moves on its night, with only the tags that were certain, and shows in a Needs review view on its site. Review never blocks a wave and nothing is left behind. |
| It learns | Every correction becomes a rule for the next wave, so later divisions see fewer files in their queue. |
| Effort | About 1,180 groups across four divisions. Most groups take under a minute, so three to four hours per division lead, spread over two weeks. |
| Source folder | Files | Suggested | |
|---|---|---|---|
| Field\Photos\Pump stn misc | 214 | Site Photo · P-2231 · 2024 | Confirm 214ChangeNot mine |
| Users\mhall\Desktop backup | 88 | Site Photo · P-2231? · 2024 | Confirm 88ChangeNot mine |
| Field\Daily Reports\2023 misc | 41 | Daily Report · P-2198 or P-2205 | Choose projectNot mine |
Before and after: one library
The Design files of P-2214 Highway 9 Interchange, as found on the server and as they land in SharePoint.
Before · \\cesrv01\Projects\2214 Hwy 9
Dwgs ├─ Civil │ ├─ C-101 rev C.pdf │ └─ OLD │ ├─ C-101 rev A.pdf │ └─ C-101 rev B.pdf ├─ Civil - current copy of Civil ├─ FINAL DRAWINGS │ └─ FINAL DRAWINGS (2) again └─ Struct └─ Mark's working a person └─ New folder └─ 2025 └─ June … 11 levels Calcs Specs Misc anything
After · P-2214 › Design
| View: Current issued | Disc. | Rev | Issued for |
|---|---|---|---|
| P-2214-CV-DWG-101 | CV | C | Construction |
| P-2214-CV-DWG-102 | CV | B | Construction |
| P-2214-ST-DWG-201 | ST | 1 | Construction |
| P-2214-ST-CAL-014 | ST | 0 | Review |
| P-2214-ST-SPC-033 | ST | A | Review |
Folders: Civil · Structural · Mechanical · Electrical · Process. Other views: My drafts · Superseded (hidden).
| Measure | Before | After |
|---|---|---|
| Files a person sees in the library | 4,812 | 3,522 |
| Old copies and revisions | 1,290 loose | 1,290 in history |
| Deepest folder | 11 levels | 2 levels |
| Places a current drawing could be | 6 | 1 |
| Files tagged | 0 | 100% |
| Tagged by a person | 6%, in 25 min |
Where it runs and what it reads
Inside Tab & Shift's own tenant. The reader runs on Tab & Shift's migration server and in Tab & Shift's own Microsoft 365 and Azure subscription. It reads title blocks, first pages and file properties to suggest tags, keeps no copy, and sends no file content anywhere else. Folders marked Restricted in the folder map, such as HR and payroll, are tagged from their path and name only and are never read.
Decisions needed
| # | Decision | We recommend | Owner | By |
|---|---|---|---|---|
| 1 | Confidence thresholds | Apply at 0.90 and over, review under 0.70, as set on the trial run | Marcus Chen | Oct 9 |
| 2 | Files nobody reviews in time | They move with a Needs review flag rather than wait | Marcus Chen | Oct 9 |
| 3 | Reviewer for files from people who have left | The division lead | Division leads | Oct 9 |