How-To / How-to
How to Add Reliable Source Citations to a RAG App
Build traceable RAG citations by preserving source metadata, linking claims to passages, and testing every reference.
Decision intelligence for AI operators
Read the change. Trace the consequences. Carry out a decision.
Context Wire separates evidence from inference, follows impact through three orders, and preserves its countercase and rollback.
Background branches
The reporting behind the Signals
How-To / How-to
Agent Session Handoffs: A Portable Artifact Guide
Preserve interrupted AI agent work with a vendor-neutral, schema-validated handoff that records diffs, tests, failures, approvals, and safe recovery.
How-To / How-to
How to Review AI-Added Dependencies Before You Merge
A dependency review method for AI-written pull requests covering lockfiles, vulnerabilities, licenses, provenance, maintenance, and removal.
Interactive signal map
Follow the wire from evidence to action
Select a node—or tab through the buttons—to inspect what changes at each stage.
- Verify what changed: Start with checked evidence, freshness, source status, and the strongest countercase.
- Trace what follows: Follow first-, second-, and third-order effects without hiding uncertainty or causal gaps.
- Choose a reversible move: Match the action to role, horizon, exposure, capacity, and a named rollback trigger.
Layer 01 · Signals
Evidence before interpretation
Each Signal carries source status, freshness, checked dates, a countercase, and an explicit release state.
3 active fixtures
Model tiers turn routing discipline into a cost control
A three-tier model family makes per-task routing more consequential than choosing one vendor-wide default.
- Checked
- Review by
- Evidence
- 2 entries
A launch-price window can distort the business case for agents
Temporary model pricing makes prototype cost data unsafe as a steady-state budget unless teams model the post-window rate.
- Checked
- Review by
- Evidence
- 3 entries
A reported review window could add a second clock to frontier releases
A reported US pre-release review framework could make frontier-model launch timing depend on a government process as well as vendor competition.
- Checked
- Review by
- Evidence
- 3 entries
Layer 02 · Implications
A consequence is a chain, not a set of cards
The wire makes the mechanism visible, keeps confidence attached to each step, and gives the strongest countercase a branch of its own.
Impact wire
Model tiers turn routing discipline into a cost control
mixed
Follow the mechanism, not only the headline. The countercase remains attached to the same decision path.
Signal
A three-tier model family makes per-task routing more consequential than choosing one vendor-wide default.
Checked
01 · First order
Unit cost separates by task difficulty
Input and output rates differ materially across the family.
A single-model default now embeds a measurable overpayment or quality trade-off.
02 · Second order
Evaluation moves from model level to route level
Promotion and demotion rules require task-specific failure and retry evidence.
Teams need small repeatable test sets for each workload, not one leaderboard score.
03 · Third order
Portable evaluation becomes bargaining leverage
A workload-level scorecard can compare replacement routes without rewriting the product.
Vendor switching becomes less of a migration event and more of a controlled routing change.
Decision gate
Preserve the safe state
Return the affected workload to the last measured default route.
Layer 03 · Decision
A brief shaped by role, horizon, exposure, and capacity
Choose only operating constraints. The same fixture and choices always produce the same ranked brief and reasons. No personal or company-confidential input is requested.
Open the local Brief builderInitial roles
- Founder / Operator
- Engineering leader
- Procurement / Strategy
Save or delete on this device. Export and import validated JSON. Nothing is uploaded.