AI for Biometrics
Find Hours, Budget Reconciliation (effort only)
A visual, runnable guide, from the biostatistics lens
Run / share / skim / file / build, pick your mode. This page shows the reconciliation (diagrams + a live roll-up computed from the ledger), the runnable R Shiny app lets you run one interactively, the dashboard is a shareable snapshot, the one-pager is the skim, the Excel workbook is what you file, and the build-it walkthrough shows how the app is made. Operations-only, synthetic, hours / effort only (no dollars, rates, or headcount). We say participant, not subject.
The whole idea, in one line
When a new out-of-scope ask lands, you do not immediately ask for more. First you find the hours you already have: honest under-run slack across the tasks that are tracking under plan, plus the management reserve. Only the residual becomes a change order.
\[\textbf{Net findable} \;=\; \big(\textstyle\sum \text{Planned} \;+\; \text{Contingency}\big) \;-\; \textstyle\sum \text{EAC}\]
Under-run slack is the part of your planned effort you are forecast not to spend (planned minus estimate-at-completion, floored at zero). Contingency is the management reserve. Add them and you have the effort you can redirect before touching the contract.
What to do with a new ask, the decision tree
The tool automates the first two branches: it sources hours from under-run slack first, then contingency, and reports whatever is left as the change-order residual. Deferring / re-sequencing is a human decision you make before you accept a residual.
Where every number comes from, the source map
The ledger you see below is a picture built from synthetic data. On a real study each column is fed from a system of record, and the tool never invents a number.
| Column | Source of record |
|---|---|
| Planned | SOW / budget grid in CTMS |
| Actual to date | Timesheet system |
| % complete, EAC | Your estimate + the progress tracker |
| Change order | Contract process: lead biostatistician + PM review, sponsor sign-off, SOW amended |
The tool does not connect to CTMS or timesheets. You transcribe current figures into the ledger, on purpose, so the reconciliation is attributable and stands on its own. Every number reported here comes from the validated ledger and the shared logic core, never from an assistant.
The live roll-up, computed from the ledger
This is the same logic the Shiny app and the Excel workbook use, rendered here from the effort ledger, project_hours_ledger.csv (in the templates/ folder). Edit that ledger (or run the app) and these move.
850
Planned (h)
513
Spent to date (h)
831
Forecast EAC (h)
40
Contingency (h)
59
Net findable (h)
Net findable here is 19 h of under-run slack (planned 850 − EAC 831) plus 40 h of contingency = 59 h you can redirect before a change order.
Tasks where the blue (planned) bar sits above the gold (forecast) bar are the ones giving up slack. Where gold exceeds blue, that task is over-running and is being watched, not raided.
| Deliverable | Planned | Actual | % done | EAC | Under-run slack |
|---|---|---|---|---|---|
| SAP + randomization schedule | 80 | 78 | 100 | 78 | 2 |
| ADaM specifications | 60 | 55 | 100 | 55 | 5 |
| ADaM programming | 120 | 90 | 70 | 128 | 0 |
| ADaM QC / validation | 80 | 40 | 60 | 70 | 10 |
| TLF programming | 200 | 120 | 60 | 210 | 0 |
| TLF QC (double-programming) | 120 | 50 | 45 | 110 | 10 |
| PK / NCA (Phoenix) | 60 | 30 | 55 | 55 | 5 |
| Interim / DSMB outputs | 40 | 40 | 100 | 40 | 0 |
| CSR tables + narrative support | 90 | 10 | 10 | 85 | 5 |
Under-run slack is per-task max(0, planned − EAC). It sums to the portfolio net slack of 19 h.
Testing an ask, two worked examples
Ask: 45 h → Absorbable internally
19 h from under-run slack
26 h from contingency reserve
0 h residual (change order)
Contingency left after this ask: 14 h.
A 45 h ask is fully covered: 19 h from under-run slack and 26 h from contingency, no residual. It is absorbable internally with 14 h of reserve still held back.
Ask: 90 h → Change order needed: 31 h
19 h from under-run slack
40 h from contingency reserve
31 h residual (change order)
Contingency left after this ask: 0 h.
A 90 h ask exhausts the 19 h of slack and the full 40 h of contingency, leaving a 31 h residual. That residual is genuine new scope: it goes to the change-order process (lead biostat + PM review, sponsor sign-off, SOW amended). The tool drafts the hours-only note; pricing is handled downstream by the PM.
Run it, and build it
The interactive tool is a real R Shiny app you own. From the package root:
shiny::runApp("budget_reconciliation/shiny")
# needs: install.packages(c("shiny","bslib","DT","ggplot2"))Edit a task’s EAC, change the ask, and the net-findable roll-up and verdict recompute live. Download the reconciled ledger to CSV to drop into the Excel workbook.
You want to build these yourself. The whole app is small: shiny/R/logic.R holds the reconciliation math (the functions this page also called), and shiny/app.R is the UI + reactivity. The build-it walkthrough takes you through it piece by piece. Change the ledger CSV to your study and it is your reconciliation.
The other formats
| Mode | Artifact |
|---|---|
| Run | Run & build the Shiny app, edit EAC, test an ask, download |
| Share (no R) | Dashboard, the roll-up as a self-contained page |
| Present | Slides |
| Skim | One-page cheat-sheet (PDF) |
| File | Excel workbook, the record you keep |
| Build | Build-it walkthrough |
Synthetic data throughout. Hours / effort only. A change order is a named-human, contracted decision; pricing is handled downstream by the PM / project finance.