Invite-only research environmentFour hundred million repositories.
Four hundred million repositories.
A shortlist you can defend.
DealRadar scores public software repositories on eight deterministic signals, routes every candidate through a human review gate, and turns the survivors into structured research you can hand to a partner.
It reduces the search space and organises the diligence. It does not predict acquisitions, and it does not know whether anything is for sale.
The demonstration runs on illustrative data and needs no account. Every published figure has a command behind it.
Research disclaimer. Public-signal research lead. Availability, ownership, willingness to sell, and transferability are not verified unless explicitly marked as manually verified.
What DealRadar does
A research instrument, not an oracle
Sourcing small software assets is a search problem before it is a diligence problem. The public signal is abundant but unstructured, and the expensive step is deciding which twenty repositories out of millions deserve a human afternoon.
DealRadar does that narrowing with arithmetic you can inspect, then gets out of the way. There is no model, no inference and no learned parameter anywhere in the system. Every score can be recomputed from its inputs and explained line by line.
Explainable by construction
Eight named factors, each with a visible value and weight. No hidden features, no unexplained ranking.
Human review is mandatory
No candidate reaches a buyer without an operator decision. The gate is enforced in the API, not the interface.
Four facts kept apart
Signal score, evidence quality, review state and verified availability never merge into one number.
Reproducible outputs
Every report carries a SHA-256 checksum, a source-capture date and the calibration figures behind it.
Research workflow
Scan, score, review, research, report
The same five steps every time, with a human decision in the middle that cannot be skipped.
- 01
Bounded scan
An operator runs a scan against a niche with an explicit candidate cap. Rate limits and caps are enforced server-side — there is no unbounded crawl.
- 02
Deterministic score
Each candidate is scored on eight public factors. The same inputs always produce the same score, and every factor is shown with its value.
- 03
Human review gate
Nothing reaches a buyer automatically. An operator approves, rejects or flags each candidate, and records availability separately from the score.
- 04
Buyer research
Buyers see approved leads only. They triage, keep private notes, build a watchlist and request custom scans.
- 05
Report and deliver
Approved candidates become a structured report with methodology and disclaimers, delivered over TLS to a named recipient with a full audit trail.
How signals work
Eight factors, all of them visible
Each factor scores 0–100 from public GitHub metadata. Weights adjust within bounded ranges according to data completeness, so a candidate with thin evidence cannot score highly on the strength of one lucky signal.
What a high score does and does not mean
It means
Public evidence suggests this codebase is worth a human research session: it looks maintainable, adoption is real, and the ownership structure does not obviously preclude a transfer.
It does not mean
That the project is for sale, that its owner would entertain an offer, that a transaction is achievable, or that any revenue exists. None of those facts are public, so DealRadar does not claim them.
Sample lead intelligence
What a research record looks like
An illustrative record showing the structure of the interface. The repository below is fictional and the values are synthetic — DealRadar does not publish real candidate data outside an authenticated workspace.
example-org/sample-toolkit
Illustrative exampleA self-hosted workflow automation toolkit. Synthetic record for layout purposes.
4,120 stars318 forksTypeScriptMIT
Signal
82
Signal factors
- Technical quality
- 88
- Business signal
- 74
- Abandonment
- 81
- Monetisation surface
- 62
- Acquisition fit
- 70
- Deal execution risk
- 45
Diligence status
Evidence qualityHigh evidence
AvailabilityNot verified
Review stateApproved
Signals captured12 Aug 2026
Identified risks
- Single maintainer — continuity and IP diligence required.
- Commercial keywords present, but no billing surface was confirmed.
Note the availability row. It reads Not verified because nobody has checked — and it will keep saying so until an operator manually confirms it and signs the record.
Evidence & methodology
The measured numbers, including the weak one
On the committed 70-repository dataset, 4 of the top 5 and 6 of the top 10 were labelled research-worthy by one reviewer. The full-set 70% Precision@10 target was not met.
Labels include records needing more research and do not establish that repositories can be acquired. Precision@10 of 60% is below the 70% gate the programme originally set for itself. That shortfall is disclosed rather than buried, because a sourcing tool that overstates its accuracy is worse than no tool at all.
Ranking performance
Calibration set dealradar-calibration-v1 · 70 repositories · 24 labelled positives
- Precision@5
- 80%
- Precision@10 (full set)
- 60%
- Precision@10 (holdout)
- 70%
- Precision@20
- 60%
- Capture@20
- 50% (12 of 24)
Read this honestly. At 60% Precision@10, roughly four of every ten candidates in a top-ten list will not survive diligence. The value is in narrowing millions to a reviewable shortlist — not in being usually right about any single candidate.
Security & privacy
Built for a controlled environment
Invite-only identity
No public registration exists and none is planned. Accounts are created by an operator; sessions are server-side, HttpOnly and revocable.
Role separation enforced server-side
Buyers can only ever read approved leads. The restriction is applied in the API, not in the interface, and is covered by negative tests.
Contact-data redaction
Personal emails, phone numbers and guessed social profiles are never shown to buyers. Only public repository and profile URLs are surfaced.
Full audit trail
Sign-ins, review decisions, report generation, previews, downloads and every delivery attempt are recorded with actor and timestamp.
Who it is for
And who it is not for
A good fit
- Micro-acquirers and independent operators sourcing small software assets
- Search funds and solo buyers who need a repeatable, defensible sourcing process
- Technical diligence teams who want factor-level evidence, not a black-box score
- Anyone who has to justify a shortlist to a partner or an investment committee
Not a fit
- Buyers who want a list of businesses confirmed to be for sale — DealRadar does not verify availability
- Anyone expecting revenue, customer or financial data — none of it is public, so none of it is here
- Automated outreach at scale — DealRadar deliberately contains no outreach capability