Contextual
Doc 0.4.2Rev 14Compiled 2026.05.02Hash 0xA42F…91CLat 37.7749 / Lon −122.4194
I.
Contextual

Find where a conversation went off track.

Contextual compares two private reflections with the shared conversation record to show where meaning drifted, what each person likely heard, and how to restart the conversation without deciding who is right.

Run a mediationSee how it works
Private reflections stay private
Shared evidence stays cited
Public context is used only when external facts matter
Product boundaries
Contextual does not decide who is right.
Private reflections are used as context, not shared content.
Shared conversation is treated as evidence.
Public context is only used when external facts matter.
Build
0.5.0 — Alpha
Mode
Evidence-aware mediation
Inputs
2 private reads · 1 shared record
Output
Gap analysis · repair starters
Boundary
No verdicts
§ 01
How it works

See the product flow in one pass.

Contextual is easiest to understand when you follow the same path the agent follows: private reflections, shared evidence, comparison, then a repair readout grounded in citations.

01.01 / How it works4 steps
  1. Step 1

    Each person writes a private reflection

  2. Step 2

    Paste the shared conversation

  3. Step 3

    Contextual compares reflection vs evidence

  4. Step 4

    Review the gap analysis and repair starters

01.02 / Example readcompact case
What A thought

B stepping in meant B did not trust A to finish the work.

What B meant

B thought they were helping and had not realized A saw the work as settled.

Shared evidence

The thread shows a direct ownership/help mismatch, followed by a defensive reply.

What Contextual found

The conversation drifted from process into perceived intent before either person checked the other meaning.

§ 02
Inputs

Bring in the two inputs Contextual compares.

Start with two private reflections and one shared conversation record. The reflections stay private. The shared conversation is treated as evidence.

02.02 / Shared evidence — paste conversation
Empty0 lines
paste · cite · compare
Paste the shared record directly, or let the MCP mock fetch it for the demo. Contextual will assign line numbers and cite them in the output.
§ 03
Run Contextual

Run the mediation read against the current inputs.

This step compares the reflections against the shared evidence, decides whether external facts matter, and returns a traceable readout.

02.01 / Private reflections — enter freely
used as context only
Reflection A

Person A private reflection

Used as private context only. Not quoted back as shared evidence.

Reflection B

Person B private reflection

Used as private context only. Not quoted back as shared evidence.

↑ boundary · enforced at write ↓
private by defaultused as context · not evidence
03.01 / Run ContextualIdle
Trace steps · 0/5
    Waiting for trace...
Output
Ready to run

Add both private reflections and shared evidence above before you run the agent.

contextual.read · v0.5.0idle · awaiting inputs
§ 04
Repair output

Review the repair output before going deeper.

Read the concise repair summary first: what the evidence supports, where the conversation drifted, and the draft repair starters for each side.

04.01 / Repair outputawaiting inputs

After you run Contextual, this section will summarize what the shared record supports, where meaning likely drifted, and two neutral repair starters to review.

§ 05
Interpretation gap

Open the interpretation gap only after the summary makes sense.

This is the deeper read. It separates what was said from what was likely heard, then shows the bridge language that could restart the exchange.

05.01 / Interpretation gap
locked until run

This deeper read stays empty until the agent finishes. After the trace completes, cited interpretation gaps from the API response render here.

§ 07
Privacy and consent

Keep the privacy boundary legible.

Contextual is useful only if the boundary between private context and shared evidence remains clear. This section makes that boundary explicit.

07.01 / Privacy and consentboundary · explicit
Stays private
  • Each private reflection is context for the agent, not shared content.
  • Working notes and discarded drafts stay on the private side of the boundary.
  • The agent should not quote a private reflection back as if it were evidence.
Can be reviewed together
  • The shared conversation is treated as evidence and cited by line.
  • Repair starters are drafts to review, not verdicts to enforce.
  • Nothing should leave as a final repair message without clear participant consent.
§ 08
Public context

Use public context only for external facts.

Public context is not a general enrichment layer. It should only enter the run when the disagreement depends on a fact that can be checked outside the conversation itself.

08.01 / Public contextchecked only if needed
Use it when

The disagreement depends on policies, pricing, availability, delays, reviews, or another external fact.

Skip it when

The conflict is mainly about tone, implication, wording, or what each person inferred from the same exchange.

Current status

Public context is off by default until you explicitly enable it for a run.

§ 09
Session log

Keep the log, but move it to the end.

The session log is still part of the system language, but it should follow the main task flow rather than compete with it.

09.01 / Session logportable summary
Current session shape
Artifacts
1 transcript · 2 reflections · 1 repair readout
Output
Gap analysis · repair starters · trace
Export
session-01.read.signed
Review posture
  • Shared evidence stays cited and line-addressable.
  • Private reflections stay on the private side unless a participant chooses otherwise.
  • Repair language remains reviewable and does not decide who was right.