It does not make the decision faster. It removes the reason for making it.
The current story's most expensive panel was a rational choice: check all six places, because six unreconciled sources leave no other route to certainty. With one aggregated view, that stops being the only route — so the panel that cost the most now costs nothing.
The same six beats, redesigned
The diff between the two curves is the redesign. Every value in the current state is negative; every value in the future state is at or above neutral.
| Panel | Current | After | Δ | What moved it |
|---|---|---|---|---|
| 1 · Trigger | −1 | 0 | +1 | The cold open is gone — the view exists before she sits down, so there is nothing to brace for. |
| 2 · Action | −1 | +1 | +2 | One grouped view replaces the first of six manual checks. |
| 3 · Discovery | −2 | +2 | +4 | The half-thread-no-answer moment cannot occur — everything is aggregated before she looks. |
| 4 · Decision | −3 | +2 | +5 | The “must check all six places” commitment disappears; there is nothing left to cross-reference by hand. |
| 5 · Obstacle | −4 | +3 | +7 | The largest movement in the set — the thirty to forty minutes is the exact cost removed. |
| 6 · Ending | −2 | +4 | +6 | She starts design work with the focus the current state spent on archaeology. |
Panel 6 was the hardest to fix without cheapening it. A system that congratulates you for removing its own defect reads as hollow — so nothing is staged. The labour simply never happens, which leaves nothing to celebrate recovering from.
What ships, against what was diagnosed
Each surface traces to a root cause from the audit. Nothing here exists because it was a good idea in isolation.
| Workflow | Root cause | What ships |
|---|---|---|
| W1 · Morning reconstruction | Thirty to forty minutes a day cross-referencing six sources before real work can start. | Signals aggregates every connection's activity by project, read-only — showing silence as prominently as activity, so the Ghosting Sequence and Feast-or-Famine are visible by design rather than filtered away. |
| W2 · Delivery → Approval | Work ships with no interpretive scaffold, so feedback defaults to taste; a verbal “approved!” is the entire record of a financially binding transaction. | Rationale captures intent at the moment of delivery. The Approval Gate produces a timestamped, retrievable sign-off that cannot exist without a rationale already on record. |
| W3 · Stall → Re-entry | No signal tells the user whether a quiet project is alive or dead. | Days since last activity, surfaced per project — observed rather than requested from the client, and never inferred. |
| W4 · Proof of work | The decision record needed for case studies was never captured when capture was cheap, so it is rebuilt from memory or skipped. | Proof of Work auto-drafts from rationale, sign-off and recorded outcomes — records that already exist because W1 and W2 captured them. Reviewed and accepted, never auto-published. |
How the plan changed on contact with building it
Four workflows and their root causes never moved. Only the mechanisms serving them did — each one getting more honest about what it could actually claim.
- A digest became a view
- The feed originally marked items as acknowledged and showed only what had not been cleared. A misclick removed a card with no way back. Rather than patch it with an undo, the screen was redefined: nothing is cleared, silence is shown as loudly as activity, and filters derive from records the product already owns. The rename followed the redefinition — every other destination is named after the entity it lists, and this one was named after a ritual.
- One entity split into three
- “A Slack workspace” and “a Figma file” are not the same kind of thing. Separating them made project attribution a declared mapping instead of a matching process that sometimes failed silently. It also fixed a real limitation: one connection per provider made the hero user unrepresentable, since a freelancer is a guest in several clients' separate workspaces.
- Publication inverted from a redaction list to an allow list
- Loaded live during the audit, a published showcase leaked four things to a real outsider — including another client's confidential reasoning and a named approver at a different company. The fix was not patching what got hidden. The page is now assembled from fields explicitly cleared for release, so the component never sees the full dataset and a private field cannot leak by omission.
- Search that refuses to read messages
- Wiring up an inert search field raised a question already settled once: should it search client message content? It does not — and the palette says so on screen rather than returning a silent empty result.
The central claim survives every revision above: ambiguity is held, never resolved by inference.
The unattributed pile, the gap between a pending gate and a client saying “approved from our side!”, and the blank confidential summary that blocks publication until it is written — these are not left-over rough edges. They are the visible evidence that the rest of the product is telling the truth.
Not resolved, deliberately
Stated rather than quietly closed, because a case study that only reports its wins is not evidence of anything.
- The concept test has not run
- The target behaviour statement is a hypothesis proven internally consistent — not yet validated against real users.
- One success metric needs rewriting
- It counted unacknowledged items, a state that no longer exists after the feed was redefined.
- Client consent is not captured
- Before a client appears in a confidentiality-summarised showcase, the user's own assertion is trusted. Recorded as a real gap for fieldwork to validate, not an oversight.