J-Jones™ / Temporal knowledge map

What value does a private intelligent knowledge base give me?

J0. Pick up a project after switching AI tools, trace a decision to its source, and choose which proposed changes become part of your working record.

Tap a branch to explore its questions, or open the release timeline.

Illustrative release review

What did we know then?

Tap a dated record to see what the team knew.

Monday meeting note

Friday is the agreed launch date.

Tuesday test report

Sign-in failed. The agreed launch date remains Friday.

Wednesday morning proposal

Moving launch to Monday is proposed. Friday still governs.

Wednesday afternoon decision

The owner approves Monday. Readiness remains unknown because no passing retest is recorded.

J1 / Audience

Who benefits?

J1.1 · Returning to client work?

Recover the agreed scope and why earlier options were rejected before drafting the next deliverable.

J1.2 · Investigating a configuration?

Read the original approval and evidence before reversing a production safeguard.

J1.3 · Reviewing a release?

Compare the latest test report with the scope of the earlier sign-off.

J1.4 · Is file search enough?

A known keyword can locate a file. A disputed decision also requires dates, sources and approval history.

J1.5 · What must a pilot confirm?

Verify the relevant release, supported operating systems and import formats against the records you actually use.

↑ Back to map
J2 / Product direction

Reuse records

J2.1 · Where does a note start?

Keep dated notes and reports under local control. Retrieve the relevant passage when the project resumes.

J2.2 · What changes as evidence becomes knowledge?

A test log records an event. An extract quotes it. A model interprets it. A human approves any change to the maintained decision.

J2.3 · What if I forget the original chat?

Ask about the decision across the records available in your workspace. Check the source before acting.

J2.4 · How do I verify coverage?

Try representative files against the documented release and inspect what was retrieved.

J2.5 · Which formats are supported?

The supplied outline leaves the V2 format list, export schemas and indexing contract open.

↑ Back to map
J3 / Provenance

Trace evidence

J3.1 · What anchors a claim?

Inspect the exact passage, author, date and whether the words support the conclusion.

J3.2 · What if sources conflict?

Keep both visible. Compare scope, timing and approval authority before resolving the disagreement.

J3.3 · What does a citation prove?

It identifies the source. You still judge whether it applies today.

J3.4 · How do I check the interpretation?

Read the paragraphs around the quoted passage. Look for exceptions and conditions.

J3.5 · Which verification controls exist?

Hash-backed links, line references and the conflict interface are open implementation questions in this outline.

↑ Back to map
J4 / Governance

Approve changes

J4.1 · Who approves an update?

AI drafts a proposal. You review the evidence and decide whether it becomes a maintained record.

J4.2 · How do permissions differ?

Asking reads context. Proposing drafts a change. Approval authorizes persistence. This boundary applies across Strict Clone, General AI and Full AI.

J4.3 · What governs while a proposal waits?

In the example, the earlier approval stays active. A dated decision record establishes when the replacement took effect.

J4.4 · Which roles are planned?

The supplied roadmap describes human authority and places team administration in future work.

J4.5 · What remains unspecified?

The exact proposal states, review queue and audit schema require release-specific verification.

↑ Back to map
J5 / Historical reasoning

Reconstruct history

J5.1 · Which times matter?

Distinguish when an event occurred, when someone recorded it, when the change was approved and when it took effect.

J5.2 · Can a superseded decision still be accurate?

Yes. Friday was the agreed launch date on Tuesday, even though the owner changed it on Wednesday.

J5.3 · Why did the decision change?

Follow the approval back to its proposal and cited test report. Mark gaps as unknown.

J5.4 · How do I test historical recall?

Ask the same question at two cutoffs. Check whether the earlier answer uses only evidence available then.

J5.5 · Are all four times stored automatically?

The supplied material leaves the metadata fields and point-in-time query interface open.

↑ Back to map
J6 / Owner-described configuration

Run local AI

J6.1 · What would I hold?

In the owner-described setup, you hold Qwen3 1.7B weights and a task adapter, with Ollama running inference on your machine.

J6.2 · What does the adapter learn?

A repeatable review task, such as citing release evidence, identifying blockers and asking for the next verification.

J6.3 · Where does training happen?

A dedicated or rented GPU trains the adapter. Your own machine runs everyday inference.

J6.4 · Where do changing facts live?

Keep today's test result and launch date in the knowledge base. Updating those records does not require retraining the adapter.

J6.5 · Can I complete the task locally?

The proposed local path supplies retrieved passages to Ollama. Installation requires acquiring the model and adapter first.

J6.6 · How do I test specialization?

Give the base and adapted models identical unseen cases. Score citations, missed blockers, unsupported claims and justified abstentions.

J6.7 · What if the model gets it wrong?

Inspect the source, correct or reject the draft, and retain authority over the maintained record.

J6.8 · How do I verify availability?

Check release-specific packaging, model compatibility and license terms before adopting this owner-described configuration.

J6.9 · What is still unmeasured?

The supplied material gives no blind evaluation scores, hardware minimums or measured inference latency.

↑ Back to map
J7 / Privacy boundaries

Control data flow

J7.1 · Where does a local question go?

In the owner-described path, retrieval and inference run on controlled hardware. Verify the network behavior on the target installation.

J7.2 · What changes with an external provider?

Selected prompts and retrieved passages cross the local boundary when sent to the provider.

J7.3 · What about remote access and backups?

The concepts include Tailscale access to a VPS and self-hosted Git. Check backup, export and retention settings separately.

J7.4 · How do I verify local operation?

Observe outbound traffic during a representative review. Repeat with external network access blocked and check task completion.

J7.5 · Which boundaries remain open?

The outline leaves encryption at rest, local logs, telemetry and retention behavior for release-specific verification.

↑ Back to map
J8 / Illustrative case

Review a release

J8.1 · What date was agreed on Tuesday?

Friday. Tuesday's failed test introduced a blocker but did not change the schedule.

J8.2 · Why did the owner move launch?

Wednesday's proposal cited Tuesday's sign-in failure. The owner approved moving the target to the following Monday.

J8.3 · How could local AI help?

Draft a cited assessment identifying the failed test and requesting a passing retest.

J8.4 · Who changed the plan?

The release owner approved it Wednesday afternoon. The morning proposal had no authority to change the baseline.

J8.5 · Are we ready to ship?

Readiness is undetermined. The supplied records establish a new target date and contain no passing sign-in retest.

J8.6 · How do I check for missing evidence?

Inspect the meeting note, test report, proposal and decision record. No retest appears in this four-record packet.

J8.7 · What should the interface clarify?

How it will distinguish schedule approval from a missing readiness check remains an open design question.

↑ Back to map
J9 / Planned evaluation

Measure value

J9.1 · How quickly can I recover a decision?

Time repeated searches for an authoritative decision and its evidence using J-Jones and your existing workflow.

J9.2 · Are citations valid?

Score the share of answers whose passages exist and directly support the claim.

J9.3 · Are historical answers accurate?

Test dated questions against known answers and count cases where later facts leak into an earlier answer.

J9.4 · Can the private task finish offline?

Run a complete review with external networking blocked. Check output quality and attempted connections.

J9.5 · What remains to define?

These are proposed tests. Dataset size, hardware, baseline workflow and acceptance thresholds still need to be specified.

↑ Back to map