Intelligence · the layer that watches with youFour AI models, one job: catch what a principal would.
You can't sit in every client channel or audit every timesheet. TrackerPlus reads them for you — rating client frustration on each message as it lands and reading the whole channel overnight, ranking the projects drifting toward risk, and scanning the activity log (audit trail) for quiet integrity issues. Four leading AI models cross-check every read, so if the models disagree, that's your signal to look.
AI Channel Monitoring Report
Each message scored by four models — ranked by severity.
- ClientRiverside House · 9 Jul · 4:12 pmWe were expecting the DA set on Friday. Can you confirm where it’s at?GPT-5.5 Thinkingcritical
A committed date was missed; the client is now chasing.
Claude Opus 4.8criticalMissed deadline + a direct chase — the relationship is heating.
Gemini 3.1 ProconcernFollow-up on an overdue deliverable; tone still measured.
Gemini 3.5 FlashconcernChasing a late set — worth a same-day reply.
- ClientDocklands Fitout · 9 Jul · 11:38 amThat’s the third time we’ve gone back and forth on this same detail…GPT-5.5 Thinkingconcern
Repetition and friction — the client feels unheard.
Claude Opus 4.8concernRepeated rework on one detail; a cue to close it out.
Gemini 3.1 ProwatchMild frustration at back-and-forth, not yet escalated.
Gemini 3.5 FlashwatchSome impatience; keep an eye on the thread.
- ClientCoastal Pavilion · 8 Jul · 5:03 pmNo worries at all, whenever you get a chance — no rush from us.GPT-5.5 Thinkingwatch
Second soft deferral this week — polite, but cooling.
Claude Opus 4.8concernReads as disengagement rather than patience.
Gemini 3.1 ProsafeTaken at face value — genuinely relaxed.
Gemini 3.5 FlashsafePositive, no action needed.
Whole-channel holistic rating (0–100), four models compared.
- critical · 86/100Riverside House #24-018#riverside-house · 214 messagesClaude Opus 4.886primaryGPT-5.5 Thinking82Gemini 3.1 Pro74Gemini 3.5 Flash71
Client chased the DA set twice this week and the last two replies ran terse. A committed date was missed — the relationship is cooling.
- high · 58/100Docklands Fitout #24-031#docklands-fitout · 168 messagesClaude Opus 4.858primaryGPT-5.5 Thinking54Gemini 3.1 Pro41Gemini 3.5 Flash38
Repeated back-and-forth on a single detail is wearing thin. No missed dates yet, but the thread is trending warmer.
- medium · 44/100Coastal Pavilion #24-009#coastal-pavilion · 96 messagesClaude Opus 4.844primaryGPT-5.5 Thinking31Gemini 3.1 Pro22Gemini 3.5 Flash18
A third soft deferral this week. Opus reads disengagement where Flash reads patience — the split is itself the signal.
No single model gets the final word.
Every read is graded by four leading AI models in parallel. You pick which ones compare and which one is primary. When they agree, you can trust the call. When one reads safe and another reads critical, that disagreement is itself worth a second look.
GPT-5.5 Thinking
Deep reasoning over a whole transcript.
Claude Opus 4.8
The default primary — careful reasoning, clear verdicts.
Gemini 3.1 Pro
A strong, independent second opinion.
Gemini 3.5 Flash
Fast, cheap cross-check on every message.
Feel a client cooling before it costs you the job.
It reads each message the instant it lands and fires an owner alert on anything hot. Overnight, it reads the whole channel as one document and scores every project and team leader for frustration — a missed date, a terse reply, a third round on the same detail.
- Every channel read as each message lands — and in full each night
- A 0–100 score and a plain-English summary per project
- Four models cross-check the same channel — if they disagree, that is your signal to look
Owners are reached by Slack and email always — and by SMS the moment a channel turns critical.
Client chased the DA set twice this week and the last two replies ran terse. A committed date was missed and the relationship is cooling — worth a principal picking up the phone.
A committed DA date slipped and the client is now chasing directly. Two terse replies in a row read as a cooling relationship.
- J. Fielding71/100led 62% of the chatLed the block where the DA date was committed and then missed — most of the friction sits here.
- A. Nguyen48/100led 38% of the chatHandled early coordination cleanly; tone only warmed once the deadline was raised.
Agrees the missed date is the driver; flags the direct chase as an escalation the studio should answer same-day.
Reads the thread as high but recoverable — a prompt, dated reply likely resets the tone.
Lands a tier lower than Opus; the disagreement itself earns the channel a second look.
Projects At Risk
3 projects ≥ 60
The worst project, at the top, every morning.
A nightly pass rates every active project and lands the results on the dashboard, ranked worst-first. The owner opens the studio and reads the hottest channel at a glance — no digging, no waiting for a status meeting to surface the problem.
- The nightly sweep ranks every active project by risk score
- Worst first — the hottest channel reads at a glance
- A freshness stamp shows exactly when the AI last checked
Every anomaly weighed — not just flagged.
A daily scan of the audit trail runs five checks for time-tracking manipulation. Each hit is handed to Claude, which weighs the innocent explanation against the hard-to-explain pattern and returns a verdict — suspicious, benign or inconclusive — in plain English, with the audit trail one click away.
- Five checks scan the audit trail every day
- Each hit adjudicated by Claude into a plain-English verdict
- Owners get the suspicious ones by Slack
Integrity Findings
Suspicious time-tracking behaviour flagged by the daily scan and assessed by Claude — estimates retro-fitted to logged time, edits to invoiced entries, and similar patterns.
Concept estimate raised 40h → 96h after 62% was already logged
The estimate was increased well after most of the hours were booked against it — the budget appears to have been refitted to the time already spent rather than set up front.
A non-round 37h estimate equals 37h logged exactly
An unusual exact match between a non-round estimate and the hours logged — worth a look, but a plausible coincidence on a small, well-scoped task.
Task created two hours after its own earliest time entry
The task was back-dated a couple of hours — consistent with a mid-session rename rather than manipulation.
See the risk your team can't — in time to fix it.
Turn on frustration detection across multiple AI models, the nightly risk radar and integrity monitoring. Start free — no card required.