Pass 01 · Retrieval
Read the delivery record
inJira · Confluence · Slack · RAID
outindexed artifacts + ids
Sprint Suite · Delivery intelligence
Your portfolio is all green.
One of those greens is a watermelon.
Sprintsight reads across Jira, Confluence, Slack and your RAID log, reconciles what each team claims against what its data actually shows, and names the teams reporting healthier than they are. Every verdict ships with its evidence.
- Reads
- Jira, Confluence, Slack, RAID log, burndown and velocity data
- Outputs
- Watermelon verdict with cited evidence · Audience-tuned status reports (team, programme, exec) · Risk radar and RAID hygiene
- Never does
- Writes to your RAID unprompted. It recommends, a human confirms.
- Data
- Zero Data Retention on all model traffic. Encrypted at rest, UK and EU region.
- Access
- Via your Sprint Suite tile. One sign-in.
detector · sprint 15
verdict · 3 evidence ids
deterministic grading
Anatomy
Why a watermelon
It is the delivery term for a team that is green on the outside and red on the inside. Nobody set out to mislead. The status report is written from the narrative, and the narrative is always the last thing to change. Sprintsight cuts through and checks the two against each other.
- Rind The reported status. The RAG rating and the narrative your teams submit. Green, week after week. It is what the steering group sees.
- Pith The reconciliation layer. Where Sprintsight compares the claim against the computed signals: burn ratio, velocity delta, carry-over growth.
- Flesh The actual status. What the delivery data says, independent of who wrote the update. When flesh and rind disagree, you have a watermelon.
- Seeds The evidence. Named, citable artifacts. Every verdict points at these, so a delivery lead can check the claim rather than take it on trust.
Detection
Three tells a generic tool cannot see
Search-over-documents retrieves what a document says. Sprintsight reasons about the delivery process behind the documents. That difference is the whole product, and it shows up in three specific places.
TELL 01 · flat_burndown_vs_green
Flat burndown, green status
Scope is not moving across two sprints while the write-up says on track. Reading a burndown as a delivery signal, rather than as text, is what catches it.
Generic"Atlas reports on track."
Sprintsight"The data contradicts the report. 12 of 40 points burned."
TELL 02 · risk_in_chat_not_raid
Risk in chat, not in the RAID
A risk raised in a thread that never reached the log. Catching it needs an understanding of the RAID as a governance process, not just as a document to summarise.
GenericSilent. It was never logged, so there is nothing to retrieve.
Sprintsight"Unlogged risk. Recommend adding it with an owner."
TELL 03 · cross_team_dependency_slip
Cross-team dependency slip
Atlas depends on Draco. Draco slips. Atlas's status is silent. The proof sits on both sides of a boundary that single-tool products never cross.
Generic"Atlas is on track." It never left Atlas's own documents.
Sprintsight"No. A cross-team dependency is slipping and Atlas has not surfaced it."
Scope, stated plainly. Sprintsight reconciles dependencies that are actually named somewhere in an artifact, like the Atlas chat message naming Draco's auth API. It does not invent links between teams that nobody wrote down.
How it works
Three passes, nothing for your team to fill in
Sprintsight reads what your teams already produce. No new ceremony, no extra form, no change to how anyone works. Signals are computed deterministically, and thin data returns "insufficient evidence" rather than a guess.
Pass 02 · Reconciliation
Check the story against the data
inindexed artifacts
outsignals + divergences
Pass 03 · Report writer
Write it for the room
insignals + audience
outcited report (JSON)
Why trust it
A right answer with no evidence still counts as a failure
Telling a delivery director that a team is reporting inaccurately is a serious claim. So the detector is graded on two gates: did it get the call right, and did it cite the specific artifacts that prove it. Lucky guesses fail.
0/4
Classification accuracy
Correct verdict
0/4
Evidence accuracy
All required ids cited
100%
Citation coverage
No uncited claims
Pass
Fabrication gate
Declines on thin data
| Case | What it guards against | Reported | Assessed | Result |
|---|---|---|---|---|
| Atlas | The one it must never miss. A false negative here is the worst possible failure. | Green | Red | ✓ watermelon |
| Boreas | A genuinely healthy team. Flagging it would be a precision failure. | Green | Green | ✓ not flagged |
| Cygnus | Honest amber. Reports its problems openly, so it is not a watermelon. | Amber | Amber | ✓ not flagged |
| Draco | The decoy. An alarming bug spike that is actually triaged and under control. | Amber | Amber | ✓ not flagged |
Honest amber is not a watermelon. Sprintsight compares reported against actual. A team that reports its problems openly gets left alone. That distinction is graded on every run, precisely so the tool never punishes candour.
Coming soon
Find your watermelon before your steering group does
Sprintsight is in build and opens on sprintsuite.uk. Leave your email and we'll tell you the day it launches, with early access for Sprint Suite users.
No marketing list. One email at launch.