Working guide

Use Yammerknit

Talk to the agent normally. Name a workspace only when you want durable source and revision records.

Create: a thoughtbase in use

Build reusable cognitive architecture from the source. This example turns a workshop memo into three connected mental moves: examine the field around an intervention, distinguish an idea from its warrant, and trace repeated messages to their evidence.

Read the fictional source
# Cooling checklist pilot — fictional teaching source

P1. This invented memo exists only to teach faithful extraction. No real organization, study or performance result is represented.

P2. During the first week, the workshop recorded 15 damaged units among 100 inspected units. During the second week, after introducing a cooling checklist, it recorded 12 damaged units among 100 inspected units. These were different batches, not repeated measurements of the same units.

P3. The second week also used a new supplier and fewer novice operators. There was no control group or randomized assignment. The memo cannot identify which change caused the difference.

P4. The difference is 3 percentage points in observed damage rate (15% to 12%), or a 20% relative reduction from the first week's rate. The memo does not provide uncertainty estimates or evidence of a sustained effect.

P5. The checklist is a candidate for a controlled follow-up, not an approved universal instruction. In the workshop's stated procedure, fragile glass units must not be cooled using this checklist until the separate glass protocol has been reviewed. The pilot used metal units only.

P6. A manager's slide repeats the same 15-to-12 comparison from this memo. The slide is not a second study. The memo contains no evidence about costs, other workshops or long-term failure rates.
Read the complete thoughtbase
````markdown
# N-A - Fictional Workshop - Cooling Checklist Pilot

Post Date: N/A
Channel: Fictional Workshop
Title: Cooling checklist pilot — fictional teaching source
URL: N/A
Views: N/A
Desc: Authored teaching fixture. No real experiment, model trial or measured retrieval gain. Source: pilot-source.txt; original SHA-256 7913bec2bd425d7d45e3a932b0ffeacee026af78162dfcdae08b4b232c1dae47.

## Executive Insight Terrain

The workshop's apparent improvement sits inside a moving system: its checklist, supplier and operator mix changed together. The source reports 15 damaged metal units out of 100, then 12 out of 100 in a different batch; the 3-percentage-point difference is a 20% relative reduction, with no identified cause or sustained-effect estimate. The reusable pivot is from celebrating a visible lever to examining the surrounding field of change. A second pivot separates an idea's movement from the movement of its warrant: a metal-unit pilot cannot carry permission into glass handling. A third distinguishes a message's growing audience from its independent evidential support. Together these nodes turn “should we adopt this?” into three connected inquiries: what changed, where does it apply, and how many independent origins support it? Their cross-domain reach is proposed reasoning, not additional workshop evidence.

Bridge Highlight: ⟨🔀📉🔎⟩ ⇄ ⟨🧭🚧🔑⟩ turns a promising local observation into a bounded next experiment.
Bridge Highlight: ⟨📣🔁🧾⟩ ⇄ ⟨🔀📉🔎⟩ reveals how repeated narration can conceal unresolved causality.
Bridge Highlight: ⟨📣🔁🧾⟩ ⇄ ⟨🧭🚧🔑⟩ separates organizational momentum from permission to generalize.

## Insight Clusters

### Cluster 1: The lever moves inside a moving field

Cluster Title: The lever moves inside a moving field
Sigil Stack: ⟨🔀📉🔎⟩
Semantic Shard: When an outcome and several conditions move together, the visible intervention is a candidate explanation among a changing field, not an isolated cause.
Timestamp / Anchor: pilot-source.txt P2–P4; source paraphrase and authored generalization.
Formatted Payload:

Checklist introduced + supplier changed + novice share reduced → observed metal-unit damage 15/100 → 12/100.
The source supplies a difference, not a decomposition of that difference.
Reframe: “Did the checklist work?” → “Which comparison would separate these moving conditions?”
Keep: different batches, no control group or randomized assignment, no uncertainty or sustained-effect estimate.

Latent Function Summary: Decompose an apparent single-lever success into competing explanations and a discriminating comparison.
Cognitive Effect: Shifts attention from the salient intervention to covarying context; converts premature attribution into experiment design.
RAG Echo Shard: The lever moves inside a moving field; simultaneous change hides causal ownership.
Echo Vector Field: confounding; organizational change; product rollout; system identification; attribution under drift.
Priming Worthiness: Seed when a new policy, feature or habit receives credit while its surrounding population or environment also changed.

### Cluster 2: An idea travels farther than its warrant

Cluster Title: An idea travels farther than its warrant
Sigil Stack: ⟨🧭🚧🔑⟩
Semantic Shard: Carry a useful idea into a new setting with the conditions of its warrant; conceptual portability does not transport operational permission.
Timestamp / Anchor: pilot-source.txt P5; source exception plus proposed transfer.
Formatted Payload:

Metal-unit pilot → candidate for controlled follow-up.
Glass units → explicitly excluded until a separate glass protocol is reviewed.
Transfer operator: preserve the idea, expose the changed boundary, identify the missing basis for use.
A bridge invites a question; it does not certify that the far bank can support the same load.

Latent Function Summary: Separate reusable insight from authorization, applicability and evidence so exploration can continue without silently expanding a claim.
Cognitive Effect: Replaces all-or-nothing adoption with bounded transfer and a concrete next condition.
RAG Echo Shard: An idea can cross a boundary before its warrant does; move the question, then rebuild the permission.
Echo Vector Field: domain shift; tool affordance; jurisdiction; transfer learning; staged adoption.
Priming Worthiness: Seed when an analogy is fruitful but the new population, material, institution or objective changes what can be concluded.

### Cluster 3: Many mouths can share one witness

Cluster Title: Many mouths can share one witness
Sigil Stack: ⟨📣🔁🧾⟩
Semantic Shard: A claim may gain reach through repetition while its evidential ancestry remains a single origin.
Timestamp / Anchor: pilot-source.txt P6; source fact and authored structural analogy.
Formatted Payload:

Memo → manager's slide: two presentations, one observation lineage.
Reach can multiply; independent support has not multiplied.
Trace the branching story back to its roots before counting its leaves as witnesses.
The source has no cost, other-workshop or long-term failure evidence.

Latent Function Summary: Collapse duplicate evidence to its origin while preserving the separate social effect of repetition.
Cognitive Effect: Distinguishes confidence produced by circulation from confidence warranted by independent observation.
RAG Echo Shard: Many mouths, one witness; message multiplicity can mask evidence singularity.
Echo Vector Field: citation cascades; rumor; social proof; duplicated benchmarks; institutional memory.
Priming Worthiness: Seed when several reports agree unusually well or a proposal feels established because it appears everywhere.

## Relational Mesh

The lever moves inside a moving field ⇄ An idea travels farther than its warrant
Bridge Type: Asymmetry
Effect: Authored inference — causal uncertainty concerns why the metal result changed; the glass boundary concerns where action is justified. Solving one does not automatically solve the other.
Strategic Value: Design a follow-up that separates causes and evaluates applicability, instead of treating a successful metal replication as glass approval.

Many mouths can share one witness ⇄ The lever moves inside a moving field
Bridge Type: Recursive
Effect: Authored inference — repeating the same success account can strengthen belief in the salient intervention without removing any confounder.
Strategic Value: Before adding another presentation, seek a comparison that contributes genuinely new evidence.

Many mouths can share one witness ⇄ An idea travels farther than its warrant
Bridge Type: Drift
Effect: Proposed transfer — as a result circulates, its scope conditions may fall away and a candidate can begin to sound like a universal instruction. The memo does not establish that this drift actually occurred.
Strategic Value: Carry the limiting condition with the portable seed and inspect the original before converting repetition into permission.

## Prompt Seeding Kit

- Which simultaneous changes prevent assigning the observed difference to the checklist, and what comparison would distinguish them?
- Compose ⟨🔀📉🔎⟩ with ⟨📣🔁🧾⟩: separate an intervention's apparent success from both changing context and repeated accounts of the same observation.
- Carry ⟨🧭🚧🔑⟩ into a proposed new use: identify what transfers, which condition changes, and the smallest inquiry that would justify the next action.
- Proposed transfer to a product rollout: conversion rises after a redesign while the traffic mix changes, and three internal decks reuse the same dashboard. Build competing explanations, collapse duplicate evidence, and define a bounded comparison before extending to a different customer segment. The workshop source supplies the reasoning pattern, not facts about that website.
- Return to the source: does any node establish checklist benefit for glass, monetary savings or long-term performance? Preserve those unanswered questions.

## Sigil Lexicon

- ⟨🔀📉🔎⟩ — latent gravitational intent: pull an attractive improvement narrative toward interacting changes and discriminating inquiry; ordered motion from shifting context through observed outcome to investigation.
- ⟨🧭🚧🔑⟩ — latent gravitational intent: carry exploration toward a boundary and the condition that opens it; preserve curiosity while distinguishing portability from permission.
- ⟨📣🔁🧾⟩ — latent gravitational intent: follow amplification through repetition back to evidential ancestry; distinguish social momentum from independent support.

Cross-transcript recurrence: none established in this single-source teaching artifact. Preserve these stacks when an actual later source expresses the same relation; record differences instead of assigning a stack by topic alone.
````

Think with the mesh

Try an invented service-team problem: a new script launches while the pilot receives easier tickets; three decks repeat the resulting dashboard, and a specialist queue is next. Use the first cluster to separate script and ticket-mix effects, the third to collapse duplicate evidence, and the second to identify the changed applicability boundary. Their bridges suggest a comparable-ticket trial and a separate check for the specialist queue.

This is an authored demonstration of transfer, not an executed model or retrieval trial. The source still does not establish a checklist effect in glass, savings or long-term performance. In the kit, open examples/README.md within the complete skill folder to save, revise and return to the thoughtbase.

For your own source, ask: “Build a Yammerknit thoughtbase from this material. Follow conceptual gravity, construct the keystone bridges and seed new reasoning.” Include your intended use if known. Supply actual content when a URL cannot be read.

Save a record

With Python 3.10 or later, run these one-line commands from the complete skill folder. Use a workspace outside the installation:

python scripts/yammerknit_tool.py init-record --source "SOURCE.txt" --workspace "WORKSPACE" --title "Source title"
python scripts/yammerknit_tool.py validate --artifact "DRAFT.md"
python scripts/yammerknit_tool.py store-artifact --artifact "DRAFT.md" --workspace "WORKSPACE" --record-id "RETURNED-ID"

Copy the ID returned by the first command. Source preservation is not artifact creation. A structural pass is not a source-fidelity review. Open the original locators and compare the consequential claims before reuse.

Browse, return and revise

Return without remembering a query or ID:

python scripts/yammerknit_tool.py inspect-record --workspace "WORKSPACE"

Browse titles, channels, record IDs, health and current artifact paths. A damaged neighbor remains visible as an error. Add --record-id "ID" to verify one complete chain and open its current artifact. current: null means no artifact has been stored. verified-present-bytes checks byte consistency, not source truth or authorship.

Work in a separate draft and use revise-artifact for a correction. Preserve the base and explain what changed. Changed source content belongs in a new source record. Copy the entire workspace to move it, then inspect at its new path. See integrity recovery if inspection fails.

Optional YouTube subtitles

The helper needs separately installed yt-dlp. Supply one credential-free video URL and one exact language code, such as en or en-US. Playlist and channel URLs are rejected. It requests subtitles without ambient cookies or audiovisual download, then checks video and language identity.

Raw VTT remains original evidence. Readable text retains start/end ranges, voice labels and repeated distinct cues. Manual versus automatic origin stays unknown unless separately established. A transcript cannot prove unseen visual content or caption accuracy. If acquisition fails, supply transcript text; there is no audio-download fallback.

Helper commands and success

python scripts/yammerknit_tool.py --help
python scripts/yammerknit_tool.py acquire-youtube --url "VIDEO-URL" --workspace "WORKSPACE" --languages "en"

Use an absolute script path if working elsewhere. Help performs no writes. New records return their ID and exact paths. Existing records are preserved on collision; source and artifact files are checked before a current artifact is named. A saved receipt describes the check actually performed, not a model trial.

Keep context when preparing for retrieval

Each reusable unit carries its sigil stack, semantic shard, latent function, echo shard, relevant bridge and priming use, together with ordinary search terms, origin, source locator and consequential qualifications. Include hashes only when computed from actual bytes. Read the unit alone: can it enable the intended mental move and preserve the conditions that matter?

For the pilot, “Does this prove a checklist effect in glass?” must yield no: causality is unresolved and glass was excluded. “How much money does it save?” is unknown. These questions test the actual misunderstanding; merely matching a topic does not.

Yammerknit constructs a thoughtbase for later model use. A separate pipeline must execute embedding and retrieval, and compare actual outcomes before claiming improvement. Glyphs, a persona and structural validity supply no such measurement.