TAHOMA × REVASCENT
00Cover
1 / 12
Weekly briefing · July 14, 2026
Two delivered early.
Two land this week.
Last Tuesday we put four dates on a slide. Today we score them, show the run data behind each of your issues, and put the Power Automate conversion numbers on the table.
Today's order
How we'll run this.
- 01Last week's commitments, scored
- 02Your issues, old number to new number
- 03Remit, straight
- 04The files still in the folders
- 05Power Automate: the conversion utility, with dataTechnical
Every number in this deck comes from production run data, not status updates.
01 · Last week's commitments
The four dates, scored.
Jul 10
Version pinning
Shipped July 8. Zero version-drift failures since.
DoneVM stabilization
Runs start but don't finish. Hands-on VM work is underway now.
Pending deployJul 14
Live view + failure reporting
Shipped July 9, five days early. Step-by-step progress, overdue warnings, next-run times, and retrying kept separate from stuck.
Delivered earlyJul 17
Local screen-reading
Software shipped. Switching it on rides the same VM session as the item on the left. Still on for Friday.
On trackThis wk
Session reuse
Shipped, first enablement pulled back for a safety check. Re-enabling this week with a measured before/after. No throughput claims until the number exists.
In flight
02 · What changed since last Tuesday
Old number, new number.
Failures needing your team's attention
231/wk→62/wk
down 73%
Alert noise on your workflows
267/wk→97/wk
down 64%
Face-sheet searches stuck in queue
1,010→172
system records: 543 confirmed delivered · 296 confirmed not-found · rest retries every 4h · the visible folder cleanup is item 04
A file vanishing from the share
errored forever→retired after 2 tries
~20 files a day are still being moved on the share; we need to know by whom · item 04
03 · Remit
What the new monitoring found.
What it shows
Remit runs are starting but not finishing. The new detection surfaces a hung run in about 30 minutes; before, it sat invisible for 6 and a half hours. Better monitoring made an old problem visible, and we own it.
What we're doing
Hands-on VM work is underway now. For Wednesday the 15th: plan your write-offs on Power Automate this cycle. We'll confirm Tuesday night whether the done-signal is ready, and our target is the August 3 cycle, verified end to end before we ask you to rely on it.
We'd rather hand you this today than have Wednesday morning hand it to you.
04 · The files still in the folders
Ryan is right, and here is exactly why.
For every trip file, the system does two separate jobs: search the portal for the face sheet on days 7 through 10 and record the answer, then move the file out of the working folder once the answer is final.
The searching · fixed and current
1,010 files had no recorded answer. Now they do: 543 confirmed delivered, 296 confirmed not-found, the rest retrying on a 4-hour cycle. The records are done.
The moving · built, not yet live
The step that moves finished files out of the folders is written and reviewed, but the release carrying it never reached production. So nothing has physically moved: 33 files past day 10 still sat in five Inspira folders at the July 11 check.
Our records say done. The folders look untouched. Both are true, and the folders are what Ryan sees. The move step deploys this week, every move shows on the ops report, and this is fixed before next week's call.
05 · Power Automate Technical
We extracted the real bot logic from your exports.
- The actual automation was embedded inside the export files, which is why earlier reads missed it. The Word docs were stale summaries; the script is the ground truth, and we now hold it complete for both flows.
- Across all five scripts, the whole vocabulary is about 40 kinds of action. Small vocabulary means a plain program, no AI in the loop, can read a Power Automate export and write a draft Tahoma workflow.
- Every conversion ships with an honesty ledger: exactly which steps converted mechanically and which need a human look, line by line, with the reason.
PA export→
parser→
draft Tahoma workflow→
fidelity ledger
05 · Power Automate · calibrated on known ground Technical
It earned a score before touching anything new.
We ran the converter on the two flows where a hand-built Tahoma version already runs in production, and diffed the output step by step.
NJShine browser lane
87.8%
705 read · 612 converted · 7 upgraded · 86 flagged for review
Remit terminal lane
70.6%
428 read · 298 converted · 4 upgraded · 126 flagged for review
The remaining 17 flows start from a machine-generated draft plus a checklist, and rediscovery shrinks to short validation passes. To run it: the remaining exports (as offered), the live config location, and one-time terminal session settings.
This week
Four actions, one week.
- →VM session: unblocks Remit, screen-reading, and the done-signal.
- →Session reuse re-enabled, with the measured throughput number.
- →Folder-movement deploy: day-10 files move, verifiably.
- →Remaining Power Automate exports over to us.
Then Friday closes the last of last week's four dates.
For reference · presented July 7, 2026
Last week.
Everything from here on is last Tuesday's briefing, unchanged, so the commitments above can be read against exactly what was said.
Weekly briefing · July 7, 2026
Two workflows in.
Sixteen to go.
Where the Revascent automation stands today, what the last few weeks actually bought, and the path to scaling it across the division and beyond.
Today's order
How we'll run this.
- 01What we've heard
- 02The Remit reality: working, and why it looked broken
- 03What the first two workflows bought us
- 04Under the hood: the architecture, and how we define doneTechnical
- 05The plan, and the two live workflows
- 06NJShine is retiring: where we point next
- 07What gates scale, and where this goes
Questions welcome throughout. We hold the floor at the end.
01 · What we've heard
Everything you've raised, in one place.
- 1Remit looks broken. The live view goes dark, so a run that finished looks like it failed.
- 2NJShine can't keep pace with the daily volume, so charges stack up (88 behind at last count).
- 3Face sheets land in the wrong folders, holding up cash flow, and your Cape staff fix it by hand.
- 4The logic came as a recording, not on paper, so it feels like we would be guessing.
- 5Failures every day, code changes every day, the same treadmill as the developers before us. The process is not effective.
- 6NJShine is retiring anyway. Is this worth it over reverting to Power Automate?
All of it is fair. We take each one head-on, and close with a straight answer to every one.
02 · The Remit reality
It ran every time. The window into it went dark.
What the team saw
waiting for execution to start streaming…
Three mornings last week, Remit was flagged as down. The live view was black.
What actually happened
✓
Remit completed
unposted moved to the folder
Three times, already complete. The remits posted. The work was done.
The automation did its job. The live view is the bug, and it is the single biggest reason this feels broken. It is also a fast fix.
03 · What the first two workflows bought
The first two were the expensive part.
- We learned an environment nobody had documented, logic that lived only in people's heads.
- We built the architecture that makes the rest cheap: local, cached, self-correcting.
- We mapped exactly what gates scale, so the surprises that cost us the last month do not repeat.
The next sixteen are faster because these two were hard.
04 · Under the hood Technical
We don't ask the AI to re-solve your workflow every time.
- Most computer-use automation keeps a model in the loop for every click: observe, reason, act, repeat. Slow, costly, and it drifts.
- Ours compiles a known workflow into a typed map of states and steps. It reads the screen locally, matches a known state, runs a verified step. The AI is called only when something genuinely new appears.
observe→
match state→
run verified step→
check result→
escalate only on novelty
This is a moat for exactly your kind of work: high-volume, known, repeated. It gets faster and cheaper over time, where general-purpose computer use stays expensive and brittle.
04 · Under the hood Technical
"It ran" is not our bar.
- Completed with the AI barely involved, low escalation.
- No duplicate actions, ever.
- Every step leaves an auditable record. You can see exactly what happened and why.
- Recoveries are rare and safe, never a guess on a live portal.
When we call a workflow green, this is what we mean. It is a stricter bar than most vendors hold, and it is the bar that lets you trust the next sixteen.
The fair question
Is this just constant firefighting?
- The setup before us fixed every failure by hand, forever, because it had no way to learn. Firefighting was the model, with no end to it.
- Ours is built to get off that treadmill. Every run teaches the system, so the failure surface shrinks instead of repeating. The changes happening now lay that foundation, they do not patch the same leak.
- The daily fixes are a transition with a finish line, not how this runs. The dates they end are on the next slide.
That is the difference that matters. The old way fought fires forever. This one is built to stop needing to.
05 · The plan · in our control
Owners and dates.
Jul 10
Version pinning + VM stabilization
One targeted fix, no full redeploy. Fewer knock-on errors, steadier uptime.
Jul 14
Live view + failure reporting
You see runs happening in real time, and when something breaks we see it first.
Fixes the Remit false alarmsJul 17
Local screen-reading
Replaces the vision model. Faster, more consistent runs, lower cost.
Next wk
Session reuse
About 30% more throughput from your current login, no action needed from you.
NJShine + Remit · honest status
The two live workflows, honestly.
Remit
Completing every cycle. The live view is the fix, landing Jul 14, and that closes the "looks broken" gap. This one goes green.
NJShine
The intake stalling that caused the backlog is already solved. Two things remain, in order. Throughput: keeping pace with the daily volume, which a second login lifts. The folder logic: the ALS rules for where face sheets land, which come right once written down. It retires Jul 25 regardless, so throughput is really the question for whatever comes next.
One goes green. One is winding down on the health system's clock, not ours.
06 · The question going forward
NJShine is retiring July 25. So where do we point next?
- It is sunsetting on the health system's timeline. We will not spend three weeks propping up a system that is going away, and if reverting it for the interim is simplest for your team, that is fine.
- The real decision is where the durable automation goes next, and that is a call to make together.
Option A · carry it forward
Take the logic into whatever replaces NJShine, if that is where you want us.
Option B · skip to the next
Move straight to the next workflow in the backlog, and prove the model on one with a future.
Our steer: put the energy on a workflow that lasts. Which one is the highest-value next win for you? That is where we point.
07 · What unblocks scale
Exactly what stands between here and the rest.
- ◑The two live workflows hold the definition of done. On track for the dates shown.Us · in flight
- ☐The workflow logic written down and approved by you before we build. We capture it whatever way is easiest, a recording, a diagram, or a working session, and turn it into a spec you sign off. No guessing.Together
- ☐A second login provisioned, so we can run at full volume.Revascent
The middle one is the fix for the folder errors and the gaps: logic on paper, approved, before a line is automated.
07 · Where this goes
The path from here.
Live now
2
workflows, this division
Next
1
the workflow we pick together
Then
16
the rest of the backlog
On our side
An expanded team behind it: senior engineering leadership and added build capacity, so scale is resourced from the start.
From Revascent
1. Provision a second login. 2. Send the written logic spec.
Recap · your concerns, our answers
Every concern, with a straight answer.
1Remit looks broken
It completed every time. The live view is the bug, fixed Jul 14.
2Charges stacking up
The intake stalling is already solved. The real limit now is throughput, and a second login lifts it.
3Wrong folders, staff corrections
A separate problem: folder promotion runs on the ALS logic. Written down and approved, those errors stop.
4Logic felt like guesswork
We write it down and you sign off before we build. No guessing.
5Firefighting, like before
The old setup had no exit. Ours shrinks failures over time, and the daily fixes end on a date. Not the same treadmill.
6Worth it vs Power Automate
NJShine is retiring anyway. Point the energy at a workflow with a future, run as a service.
Where this goes
We're driving this.
- ✓Remit works; the visibility fix lands Jul 14.
- ✓NJShine's intake stalling is solved; throughput is the open item, and it retires Jul 25 regardless.
- ✓Platform fixes have owners and dates: 7/10, 7/14, 7/17.
- →Together we pick the next durable workflow, and the logic gets written down first.
The old one is winding down. Let's decide where the next win is.