rd-capture

star 0

Capture a contemporaneous R&D record at the close of engineering work that resolved genuine technical uncertainty — novel architecture, an experimental integration, or an approach whose outcome was not knowable in advance. Appends to the FY26 R&D activity register for the 43.5% R&D Tax Incentive. Do NOT use for routine work — CRUD, config, copy, dependency bumps, or bug fixes with a known solution are not R&D.

Acurioustractor By Acurioustractor schedule Updated 5/30/2026

name: rd-capture description: Capture a contemporaneous R&D record at the close of engineering work that resolved genuine technical uncertainty — novel architecture, an experimental integration, or an approach whose outcome was not knowable in advance. Appends to the FY26 R&D activity register for the 43.5% R&D Tax Incentive. Do NOT use for routine work — CRUD, config, copy, dependency bumps, or bug fixes with a known solution are not R&D.

R&D Capture — contemporaneous record for the R&D Tax Incentive

ACT claims the 43.5% refundable R&D Tax Incentive for FY25-26 via A Curious Tractor Pty Ltd (Path C; lodgement Jul 2026 – 30 Apr 2027). The refund legally requires contemporaneous records — written at the time the work happened, not reconstructed at lodgement. This skill captures that record now, while the work is fresh and the commits are right there. The claim is material (realistically $180–220K) and the ATO can audit — so the discipline cuts both ways: capture real R&D faithfully, and never inflate routine work into an R&D claim.

Eligibility gate FIRST (most engineering is NOT R&D)

Before capturing anything, apply the core R&D activity test. It qualifies as a core R&D activity only if ALL three are true:

  1. Outcome could not be known in advance — a competent professional in the field could not have predicted the result without doing the experimental work.
  2. It resolved a technical/scientific uncertainty — a genuine technical unknown, not a business or design preference.
  3. It followed a systematic progression — hypothesis → experiment → observation → logical conclusion.

If any is false, this is not a core R&D activity. Say so plainly and stop — routine CRUD, config, glue code, UI copy, dependency bumps, and bug fixes with a known solution do not qualify. (Some may be supporting R&D activities if undertaken for the dominant purpose of supporting a core activity — note that separately; don't pad the core.)

Match the existing format (pull, don't guess)

R&D activities live in per-project registers in thoughts/shared/rd-pack-fy26/:

  • act-cg-rd-activity-register.md (CivicGraph) · act-el-rd-activity-register.md (Empathy Ledger) · act-gd-rd-activity-register.md (Goods) · act-jh-rd-activity-register.md (JusticeHub).

Read the relevant register and one existing entry before writing, and match its structure exactly. Pick the register for the project this work belongs to; if it spans projects, capture under the dominant one and cross-reference. Check audit-trail.md in the same dir for how entries are logged.

Capture the record (one pass, while it's fresh)

Append a dated entry covering:

  • Date — today (do not backdate; contemporaneous means now).
  • Activity title + whether core or supporting.
  • The technical uncertainty — what was unknown, and why existing knowledge / a competent professional couldn't predict the outcome. This is the load-bearing field; a vague uncertainty sinks the claim.
  • Hypothesis — what you expected, and why.
  • Experiments — what you actually tried, in sequence. Include approaches that failed — failed experiments are among the strongest R&D evidence.
  • Observations / results — what happened.
  • Conclusion — what you now know that you didn't before.
  • Evidence links — commit SHAs, PR numbers, test runs, the files touched. These are the receipts.
  • Effort — rough hours / who, for cost apportionment.

Never fabricate an uncertainty, experiment, or result to strengthen an entry. If a field is genuinely unknown, mark it and ask — an honest gap is recoverable; a fabricated record is fraud and risks the whole claim. (Same discipline as verification.md and the provenance rule.)

Close the loop

  • Create/update the entry's .provenance.md sidecar (per the provenance rule — this is durable evidence, not disposable scaffolding, per the doc-lifecycle rule in CLAUDE.md).
  • Offer to run node scripts/grade-pack.mjs — it grades the pack and is the feedback loop on evidence quality. A low grade means the uncertainty/experiment write-up is too thin to survive scrutiny; tighten it now, while you still remember the work, not at lodgement.
  • Strategy + live plan: thoughts/shared/plans/rd-tax-incentive-fy2526-path-c.md.
Install via CLI
npx skills add https://github.com/Acurioustractor/act-global-infrastructure --skill rd-capture
Repository Details
star Stars 0
call_split Forks 0
navigation Branch main
article Path SKILL.md
More from Creator
Acurioustractor
Acurioustractor Explore all skills →