SShortSingh.
Back to feed

Solo founder warns: poor SR&ED documentation turned a claim into three weeks of archaeology

0
·20 views

A solo founder of CloudWise, an AWS cost optimization tool, spent roughly three weeks reconstructing documentation for a Canadian SR&ED tax credit claim after filing it himself without a consultant. Canada Revenue Agency requires claimants to document technological uncertainty — what was unknown, what was attempted, and why initial approaches failed — but the founder had recorded only code commits, not the reasoning behind failed experiments. He successfully received a CRA refund following a T2 reassessment notice, but warns that rebuilding a 14-month narrative from git history and AI coding sessions was far costlier than anticipated. The core issue is that commit logs capture what changed, not why earlier approaches were abandoned — information that fades quickly from memory. He now advocates a simple habit: writing two lines per dead end on the day it occurs, which he estimates would have reduced the reconstruction effort to a few days.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

Why DDR3 and newer DRAM interfaces need tuning at boot: a look at the PHY

A documentation page from the LiteDRAM project on GitHub explains the DRAM interface and the physical layer (PHY) used to connect memory to a controller. It focuses on why DDR3 and later memory standards must be calibrated during system startup. The material is part of the project's manual and is hosted in its documentation directory. The article text provided is only a link and metadata, so the specific technical details are not included in this summary. The post drew a modest response on Hacker News, with 4 points and no comments at the time it was captured.

0
ProgrammingDEV Community ·

Why Multi-Agent Coding Systems Fail in Production and a Proposed Isolation Fix

A technical report argues that many enterprise multi-agent coding projects stall because the same model both writes and judges its own code, letting blind spots compound. The author describes a containment design that partitions state by role, so an orchestrator, a contracts owner, and separate backend and frontend workers each control only their own files. Each phase must pass two independent checks: a clean-room tester that runs real builds and audits, and a read-only auditor that cannot edit what it reviews. After three failed reviews, automated retries stop and the issue is escalated with evidence. The report frames reliability as a state-isolation and control problem rather than a prompting one.

0
ProgrammingDEV Community ·

Keep Unowned Stale Vendor TXT Records Until Dependencies Are Confirmed During Mail Cutovers

The piece argues that during an edtech company's mail provider change, unknown vendor TXT verification records should be kept rather than deleted. Its reasoning is that an unowned record may still support a third-party verification dependency, and deleting it is the main risk even though retained records mostly add clutter. It recommends recording ownership when records are created, reviewing the zone periodically, and using stable record names so verifications do not accumulate. The author proposes a review process that keeps approved mail intent separate from the observed DNS state and labels unresolved tokens as needing ownership review. The hypothetical example is framed as illustrative planning rather than measured data.

0
ProgrammingDEV Community ·

Seven Java library changes that compile and pass tests but break existing users

A DEV Community article lists seven library changes in Java that compile and pass the library's own tests yet still break its users. The author distinguishes source compatibility, whether callers' code still compiles, from binary compatibility, whether previously compiled code still runs against the new jar. Because applications often update a library without recompiling its dependents, binary compatibility is especially important. Each example is checked on Java 25 and shows the real error messages that result. The article suggests safer alternatives, such as adding overloads, keeping return types, or giving new interface methods default implementations.