The R&D-Pal method
The full methodology
How the evidence is captured, how it is tested against Revenue's framework, how the score works, and how each person's R&D percentage is derived without timesheets.
01 · Capture
The Evidence Engine
A guided tool your team runs locally pulls a year of engineering activity: commits, pull requests, code reviews and tickets, from your own GitHub, Linear and Jira. Your work becomes the substrate of the claim, not something reconstructed in interviews.
10-15 min
Active time to kick off, then it runs on its own in the background.
4 questions
Per R&D project, answered at your own pace.
Under 1 hour
Designed total active engineering time, start to finish.
Sources converging on the record
Project 1
Project 2
Project 3
Project 4
What Revenue suggests you keep, project by project: Revenue's R&D tax credit file layout →
Safe to run on your machines
Local, transparent, and under your control.
No code, tickets or team activity transits us during extraction. A built-in exclusion list filters secrets, API keys, PII and source code by default, and you can extend it before you run anything.
Local
Runs on your own laptop. No third-party service in the middle.
Your AI
Uses your own AI account and your own GitHub access.
Transparent
Every prompt is plain text in the evidence kit. Nothing hidden.
You decide
You review the evidence folder before anything leaves your machine.
02 · Classify & structure
Classified and structured against Frascati and Revenue
Qualifying work must seek an advancement, resolve a real technical uncertainty and proceed through systematic investigation. Those three tests decide what is in scope before any narrative is written.
Qualifying R&D, the three Frascati tests
Technical uncertainty a competent professional could not readily resolve.
A genuine advancement sought in the field.
Systematic investigation: experiment, not routine delivery.
Organised into the sections Revenue expects
Narrative is cross-referenced against evidence at the section level, not just the project level, which surfaces gaps where the material does not yet support the framing.
The same record, now classified
03 · Structure & test
The scoring model
Our internal read on how a project's narrative and evidence would hold up if Revenue looked closely. We use it to decide what still needs work before we are willing to file.
Score 1
Narrative quality
How clearly and credibly the claim is framed in qualifying R&D terms, legible to a Revenue reviewer.
Score 2
Evidence quality
How well the available evidence grounds the claim: commits, tickets, design docs, test outputs. Weighted highest.
Score 3
Defensibility
Derived from both and weighted toward evidence, because a well-evidenced claim survives scrutiny that a well-written one does not.
Trust checks
Fabrication risk, missing primary evidence and grounding gaps act as a suppression cap, not a peer score. Trust failure is a reliability condition: it constrains the result rather than nudging it down.
Five evaluation lenses
Five evaluation lenses, applied per project
Lenses 1 to 3 feed narrative quality. Lenses 4 and 5 feed evidence quality. A project that fails a lens is not filed.
1Frascati alignment
Uncertainty, advancement and systematic investigation, met substantively rather than rhetorically.
2Revenue qualification
Reads as qualifying Irish R&D, not routine delivery dressed up. Is the project boundary clean?
3In-house attribution
Who did the work, where the in-house boundary lies, and whether connected-party positions hold.
4Evidence grounding
Do claims trace back to source records? Is the support direct, or only inferential?
5Fabrication check
Ratio of unverified claims to total citations. The target is zero.
Outcome
Filed
Filed
Filed
Excluded
Four defensibility bands, anchored to action
Strong narrative, strong evidence, no trust risks. Ready to file with confidence.
Defensible if queried, though some edges could be stronger. File, and tighten the weakest sub-score.
Likely survives a query, but not audit-safe. Fix the specific weakness before filing.
Material weakness, or trust suppression in effect. We do not file it as-is.
04 · Allocate
Per-employee R&D %, without timesheets
Self-reported time is approximate and challengeable. Instead we classify each working day from the engineering signals it leaves behind: commits, pull requests, reviews and ticket activity.
R&D day
Multiple R&D signals on a qualifying project.
R&D — FULL
Blended day
A single R&D-attributable signal on a qualifying project.
BLENDED
Excluded day
Activity only on non-qualifying projects.
NOT R&D
Silent day
No observable signal: leave, planning, meetings, off-system work.
EXTRAPOLATED
Sample engineer · working year, day by day
11 R&D · 4 blended · 3 excluded · 2 silent, of 20 working days
Floor
55%
Provably-R&D minimum
R&D days divided by working days.
Seed
65%
Best middle estimate
R&D days plus half of blended days, divided by working days.
Ceiling
83%
Maximum defensible
Floor plus blended days plus silent days extrapolated at the active R&D ratio.
Every figure is reproducible from your engineering-system exports with nothing more than a spreadsheet and this method. A Revenue auditor with the same exports can rebuild every calculation independently.
What Revenue's manual asks of a staff time split: Staff time allocation for the R&D tax credit →
05 · Submit
How Revenue reviews a claim
The pack is built for the point where a question moves from the story on the page to the source underneath it.
Stage 1
Outside view
Before reading anything, Revenue judges whether your profile looks like one that does R&D. Case selection increasingly runs through risk analytics, comparing claims across a sector.
Stage 2 · where we score
Risk review
What used to be called an aspect query, advisers now read as a Level 2 Risk Review: a focused look at one risk, or a few, on a return, usually by correspondence. It is also the stage most claims were never built to survive. The pack is read for credibility, and this is where our defensibility read applies.
What a Risk Review letter means →Stage 3
Full audit
Rarer. Revenue examines source artefacts, and the question shifts from whether the story is well told to whether it is true. Our evidence traces back to source.
Before and after a claim: The pre-filing notification → · Revenue's section 766(7) expert → · R&D tax credit clawback →
Claim it. Keep it.
Book a call to arrange a demo and an initial claim assessment.