Boredom Dividend

Boredom Dividend: mine the shelf

Practices die two deaths, and only one of them is a verdict. Some died wrong: the premise failed, or the practice optimized a bad target. Some died right but tedious, killed by their labor cost while their verdicts stayed good. Bob's rule for reviving the second kind fits in a single sentence: "these guys are fast and they don't care how boring the work is and they will do what I tell them to do" (C8). Tedium priced those practices out for humans; agents reprice them to near zero. That repricing is the dividend. This island is the method for collecting it: inventory the shelf, triage each practice, feasibility-check the survivors, revive the winners as gates.

The proof cases

Both of the pack's founding gates came off this exact shelf (C8):

• CRAP (C6) was shelved ~2000 on labor cost alone: "it took me forever to go through every one of those functions… although it was interesting, I kind of set it aside." It came back as an agent loop: "why don't you run crap over everything you've just done and it would run crap and then it would clean up the code." Full metric ground: crap-metric.md. The revived gate lives at crap-gate.
• Mutation testing (C7) has been right since the 1970s and was impractical at ~2000 scale, where a full run ate a night. Now: "Maybe it took it 30 minutes instead of an overnight run and then it would plug all the holes." The cost curve inverted along exactly the axis that shelved it, on incremental caches and diff-scoped mutants (mutation-testing.md). The revived gate lives at mutant-hunt.

The pattern generalizes past one person's shelf. The roster names exhaustive dependency checks as the next specimen (02-ROSTER-50.md, island 20), and the literature's shelf is far larger: techniques that never left the papers because running them by hand was absurd.

The mining method

1. Inventory the shelf

Enumerate the practices that were abandoned — yours, the team's, the literature's:

• your own set-asides ("interesting, but I stopped")
• team lore ("we used to do X before every release")
• checklist items everyone skips, audits that quietly stopped running
• published techniques that stayed academic because the run cost was hours of human attention

Done when the inventory is a list, not a vibe: each entry names the practice, roughly when it was shelved, and who shelved it.

2. Triage: was it wrong, or was it tedious?

Bin every entry by why it died:

• wrong: the premise failed, it optimized a bad target, or evidence retired it. Stays shelved. Reviving a wrong practice buys a fast gate on a bad target.
• tedious: the verdicts were good, and hand-scoring, overnight runs, or mind-numbing repetition killed it anyway. Candidate.
• unverified: nobody remembers why. Record that honestly and park it. Guessing a bin launders ignorance into a verdict.

3. Feasibility pass

A tedious-but-right candidate revives only if an agent can run it in a loop with a deterministic verdict. Three letters, each pass/fail:

• (a) mechanical execution: a tool runs the practice end to end, with live human judgment fully outside the loop.
• (b) binary verdict: an exit code, a threshold breach, a killed/survived count. A prose impression cannot anchor the fix-until-green loop (C4's shape, owned by the gate islands).
• (c) margin-safe cost: fast enough to run per task or per story. This is the letter that usually flips. Overnight down to 30 minutes was the whole mutation-testing revival (C7). A candidate too slow today goes back on the shelf with a note naming the cost, because cost curves keep moving.

Record the three letters per candidate. A fail on any letter re-shelves the candidate with the failing letter written down, so the next mining pass starts from the record instead of re-litigating.

4. Revive as a gate: content into existing plumbing

A revived practice becomes gate content only: a formula, a threshold, an input contract, verdict semantics. It ships as its own island, and it plugs into plumbing this pack never rebuilds:

• hook and pre-commit plumbing (where the check actually executes) belongs to agent-guardrails
• gate infrastructure (loopback routing, the ledger, band caps) belongs to archipelago
• the captured verdict (stdout + exit code) enters evidence-packet format as one rung of the packet's ladder, never a second evidence format

And before it guards anything, every revived gate must pass known-dirty-fixture acceptance: red on a known-bad fixture, green on a known-good one, both kept beside the gate. A revived practice that has never failed is nostalgia wearing a uniform. When the revival ports a human threshold, retuning the number is threshold-port's seat: experiment, never agent vote.

Enforced vs advisory

• enforced — the acceptance bar every revived gate must clear before guard duty. known-dirty-fixture's prove-gate.sh exits non-zero unless red-on-bad AND green-on-good both hold (known-dirty-fixture).
• enforced — this island's own shape. The pack validator (../../scripts/validate-island.py) checks it mechanically, exit-code gated.
• advisory — the mining itself. Inventory completeness, the wrong-vs-tedious binning, and the three feasibility letters are judgment calls, and no mechanical check covers them today. A later wave can script part of the pass, verifying that a candidate tool exists and exits non-zero on a seeded violation. v0 states these honestly as advisory rather than laundering them into enforced.

Done when

• Shelf inventory enumerated, every entry naming practice, era, and who shelved it
• Every entry binned wrong / tedious / unverified, with the death-reason recorded
• Every tedious candidate carries a written (a)(b)(c) feasibility verdict
• Each revival shipped as gate content wired to the three named plumbing islands, and prove-gate.sh exited 0 on its fixture pair before it guards anything

Verify-fix-reverify: a candidate failing a letter re-shelves with the letter recorded; a gate failing acceptance is fixed and re-proven; the whole mining pass re-runs whenever the shelf grows.

Shelved-for-wrong stays shelved. Shelved-for-boring becomes a gate, because the agent doesn't feel the boredom (C8).