What Fiction Skips

lore.md explains what the watch model took from fiction. This document is about the opposite: the problems fiction never had to solve, because a story can cut away from them.

That's where original product lives. A mechanic copied from a story is a mechanic some other product can copy too. A mechanic invented because real operators hit a problem no story has is, by construction, not a reskin.

Not scope. Current scope is the sign-on → Query → answer loop and nothing else (CLAUDE.md). This is research feeding later design. Nothing here is committed to, and several entries collide with rules that would win.

The test

A candidate passes if it could not have been copied from another product, because nothing else has the problem. Two failure modes to watch:

• Solved elsewhere — if Discord, PagerDuty or a safety app already does it, we're reskinning
• Solved by removal — if the honest answer is "don't build anything," that's a real result and it belongs here too

1. Coming home and being counted

Fiction cuts away. The hero wins and the scene ends. Nobody watches them get home.

Real operators end every single patrol this way, and the ending is the part that carries weight. Stood down already exists as a signal, and the watch confirms it — but the docs treat it as bookkeeping (an entry clears from the board) when it is arguably the most emotionally load-bearing moment in the product.

→ Candidate: stand-down is a closing exchange, not a status update. The watch confirms by name, and the operator's record shows the night closed cleanly. Cost is close to zero; the difference between "entry removed" and "Raven confirmed you're home" is entirely in the wording.

⚠️ Collides with nothing. This may be the highest value-per-byte item in the document.

2. Burnout, and coming back

Fiction has no vocabulary for this. Heroes retire dramatically or die. No story shows someone going hard for six weeks, disappearing for three months, and quietly returning.

That is the single most common real pattern, and the Heart exists to represent it. C5 already bans streaks, badges and absence commentary, which handles the harm. Nothing handles the return.

→ Candidate: re-entry is silent and complete. No "welcome back," no summary of what was missed, no implication that time passed. The board on your first night back looks exactly like the board on any other night.

⚠️ The temptation will be a gentle re-onboarding. That is a nudge, and nudges are deletion events for this population.

3. Declining the call

Fiction's heroes always answer. Refusing is a character arc, and it's always a failure to be redeemed.

Real volunteers are tired, sick, working tomorrow, or already out. Assist currently has one honest failure — it goes unanswered and that's reported. There is no way to decline, which means the only ways to not help are silence or guilt.

→ Candidate: an explicit "not tonight" response to Assist, visible to the watch as a resolved non-answer rather than a gap. Declining is a legitimate operational state and saying so out loud is cheaper than letting silence carry the meaning.

⚠️ Must never accumulate. A count of declines is a compliance metric, and C5 kills it.

4. The ninety-five percent where nothing happens

Fiction is made of the five percent. Every scene is the night something happened.

Most patrols are uneventful, and most console shifts are quiet. console.md already says the board "is mostly calm" and that when nothing is happening "it says so quietly" — which is the right instinct and is currently one sentence.

The real design question is unanswered: what makes an empty night feel like service rather than failure? Every conventional answer is banned here — no activity feed, no streaks, no "you covered 4 hours!" Those are exactly the mechanics that would make it worse.

→ Candidate: the quiet night is the deliverable, and the interface should say so without congratulating anyone. "Nothing happened. Three operators got home." That's a report, not a celebration, and it's true.

⚠️ This is the hardest one in the document and the most likely to be solved badly. It is also where an app that respects its users separates from one that doesn't.

5. Being wrong administratively

Fiction's heroes are wrong dramatically — the wrong suspect, the trap, the betrayal. Never wrong on paperwork.

Real watch failure is administrative: an overdue that passed without contact, a query answered from a stale record, a handover that dropped someone. The accountability log already records actions and inaction, which is unusually good.

→ Candidate: nothing new to build — but the framing should be explicit somewhere operator-facing. The watch will get things wrong, the log will show it, and that is the system working rather than the system failing.

⚠️ Do not turn this into performance review. The log exists so operators can check the watch, not so the watch can be scored.

6. Who watches the watcher

Fiction never asks. Oracle is trustworthy because she's Oracle.

This is already the best-developed answer in the repo — the Hostile Watch, can take watch as an endorsement scope, reviewable actions, and the right to decline to sign on under a specific watch silently. Listed here because it is the clearest existing example of the pattern this document is hunting: fiction assumed it away, and taking it seriously produced original design.

7. Someone is actually hurt

Fiction's aftermath is a hospital scene and a speech.

Real aftermath is a person who was on station when it happened. The escalation ladder is fully specified up to ACKNOWLEDGED — and then the documents stop. Failure mode 3 in escalation.spec.md already admits this: "acknowledged then nothing happens → ladder stopped; known limitation, documented."

→ Candidate: the honest minimum is that the system knows an incident closed and stops pretending it's still operational. Anything beyond that is human work.

⚠️ Hard boundary. Any actual guidance here is playbook content, which agents must not generate — the Medic's kill trigger is confident wrong guidance, and this is the highest-stakes possible place for it. Whatever ships here is written by people with real expertise or it doesn't ship.

8. Money, and who is carrying it

Fiction's heroes are billionaires or the funding is invisible.

Real operators buy supplies out of pocket, and the person doing the most is often the one who can least afford it. funding.md solves pseudonymous receipt, which is the hard technical half. The unsolved half is that the app deliberately shows no totals, so nobody can see who is carrying the cost — which is correct (C16) and leaves a real problem unaddressed.

→ Candidate: the team supply fund already noted in funding.md is probably the whole answer — crew-level rather than individual sidesteps the incentive problem entirely. Worth promoting from a footnote.

9. The first night

Fiction's origin story is dramatic and instantaneous.

A real first patrol is frightening and mostly consists of not knowing what you're doing. propagation.md has a verification task for the first contribution; nothing addresses the first shift.

→ Candidate: the honest answer may be that this is not a software problem. The Trainer exists in the twelve, trained with me is already an endorsement scope, and a person is a better answer than an onboarding flow.

⚠️ Anything engaging here is banned outright. Onboarding must not be made fun.

10. When the person who knew everything leaves

Fiction's mentors die and their knowledge passes by inheritance or flashback.

Real networks lose the operator who knew every intake rule, and the knowledge simply goes. The two-axis standing model means their contribution is credited and their knowledge is already in the directory — which is a genuinely good structural answer that nobody designed on purpose.

→ Candidate: make it deliberate. The directory is the succession mechanism, and that is an argument for contribution that no engagement mechanic could ever make: what you write down outlives your involvement.

Which of these are actually worth building

Ranked by value against cost, and honest about which are not software problems.

The pattern underneath

Seven of the ten resolve to saying a true thing plainly, at the right moment — not to a feature. Stand-down confirmed by name. Declining as a legitimate answer. A quiet night reported rather than celebrated. Re-entry with nothing said at all.

That is consistent with a product whose strongest rules are all prohibitions, and it suggests the not-a-reskin quality lives in register as much as in mechanics. Every comparable product would solve these with engagement surface. The differentiated move is to solve them with one accurate sentence and no interface at all.