Product lead
Possible job: prepare an evidence packet before prioritization and revisit why a decision was made later.
Learn: what source trail would actually change the roadmap conversation?
Pre-launch · local synthetic demo
Evidence is a chain,
not a sentence.
QuoteWeft explores a better way for small software teams to turn customer research into decisions they can inspect later.
The proposed workspace connects each draft theme to its source passages and keeps disagreement, uncertainty, and human review in the record. The separate local evidence bench uses invented records only; it makes no Claude/API call and is not a live product.
The case below is invented for explanation. No customer, deployment, partnership, or product usage is implied.
Working model / not a product screenshotFIG. 01
show the sourceAn idea for the workflow—not evidence of a built interface.
ProjectQuoteWeftindependent
StartedFebruary 2023project start
NowLocal synthetic demono live product claimed
01 / Problem map
Research gets collected. The reasoning gets lost.
Customer input arrives in different forms and at different times. Teams summarize it under deadline; later, a short statement can travel farther than the evidence that supported it.
Observed frictionCommon shortcutWhat the project would preserve
A / FRAGMENTATIONFeedback sits in separate tools and files.
A single slide compresses input before its source, date, segment, and context are easy to inspect.
B / COMPRESSIONA fluent summary sounds more certain than the notes.
The project would keep excerpts, tentative wording, and explicit uncertainty beside the candidate theme.
C / DISAGREEMENTContradictions disappear when teams look for a clean story.
Counter-evidence should be findable and reviewable—not edited out to make the narrative neat.
Design question
Can a team move faster without making its customer evidence harder to challenge?
02 / Annotated case
A plausible theme is still only a hypothesis.
This miniature case shows the evidence trail the project intends to make possible. Every person, quote, source ID, and conclusion below is synthetic.
Evidence ledger
Illustration onlySRC-014
INT · ¶03
INT · ¶03
“I paste the same request into the ticket, the roadmap, and our Friday notes.”
SRC-021
SUP · L08
SUP · L08
“When support is busy, I lose track of which requests made it to planning.”
SRC-033
INT · ¶06
INT · ¶06
“The separate Friday sheet is useful; it keeps our weekly review focused.”
Question before acceptance → Is the pain duplicate entry, unclear ownership, or simply an intentional separation of work? The source excerpts cannot answer that without more context.
Synthetic content · not customer research
03 / Proposed method
Separate model suggestions from evidence checks.
The main product question is not whether a model can write a summary. It is whether a researcher can trace, question, edit, and reject the output without losing the original record.
01
Choose a permitted source
A team selects an interview note, support export, or survey file and confirms what material may be processed.
Person selects02
Retain source identity
The application records origin and metadata, then creates stable excerpt references from the stored text.
Application03
Suggest a theme
A language model proposes related passages, cautious wording, uncertainty, and counter-evidence to look for.
Model · proposed04
Resolve each reference
A deterministic check looks up every ID in stored records and shows the exact excerpt—or flags the citation as invalid.
Application · planned05
Review, then decide
A researcher checks context and dissent, edits or rejects the candidate, and owns any downstream product decision.
Researcher decidesModel boundary
A future Claude integration is one possibility, not a current feature. The model must never mint source IDs or make the product decision.
04 / Specification
A small first release should prove the link, not promise a platform.
The scope below is a product hypothesis for validation. It is not a description of software that has already been built.
| Object | What it contains | Why it matters | State |
|---|---|---|---|
| 01 / SOURCE | Origin, date, type, language, consent status, and project access rules. | Keep the material's context and permission boundaries. | Proposed |
| 02 / EXCERPT | Exact span of text with a stable reference into a stored source. | Let a reviewer open the original evidence, not trust a paraphrase. | Proposed |
| 03 / THEME | Editable statement, scope, confidence note, and unanswered question. | Keep interpretation visibly separate from source text. | Model draft |
| 04 / EVIDENCE EDGE | Relationship: supports, qualifies, or contradicts a theme. | Make disagreement and counterexamples first-class records. | Proposed |
| 05 / REVIEW | Reviewer identity, edits, decision state, and time of review. | Preserve accountability when a draft becomes a team conclusion. | Proposed |
First version · narrow by design
- Approved plain-text notes and simple CSV exports.
- One workspace, editable candidate themes, source-linked excerpts.
- Visible dissent and a human review state.
- Basic export only after a reviewer accepts the draft.
Not in scope · unless later validated
- Autonomous roadmap changes or customer communication.
- Live CRM, Slack, or helpdesk integrations.
- An automatic sentiment score treated as importance.
- Claims that one quote represents a whole customer segment.
05 / User hypotheses
Three jobs worth testing before writing more code.
These are target-user hypotheses, not interviews completed, customers won, or commitments made.
Founder at a small B2B team
Possible job: compare recurring requests without hiring a full-time research operations function.
Learn: how often does feedback synthesis happen, and what do they use now?
Research practitioner
Possible job: audit whether a synthesized theme is supported, contradicted, or too broad.
Learn: what level of traceability is useful without adding review burden?
Business model hypothesis
A team subscription could be tested after repeated use is demonstrated. Pricing, buyer, packaging, and willingness to pay are unknown.
06 / Evaluation plan
Test where a polished summary is most likely to mislead.
No model benchmark or product result is claimed. These are proposed falsification checks for a future prototype.
Citation integrity
Does every cited ID resolve to the exact stored span?
Fail if an unsupported or altered quote appears.Contradiction retention
Does a material counterexample stay visible beside the draft theme?
Fail if disagreement disappears into false consensus.Claim calibration
Does the draft avoid overstating frequency, causality, or representativeness?
Fail if one anecdote is described as a segment pattern.Language and context
Can reviewers compare mixed-language passages and translations to originals?
Fail if a translation changes the customer's meaning.Human control
Can a reviewer reject or rewrite a theme and preserve the source trail?
Fail if rejection is hard or provenance is lost.Data handling
Are consent, permissions, retention, deletion, and model processing understood?
No real-data pilot before documented approval.07 / Build sequence
A gated plan, not a promise of a launch date.
The sequence is conditional: each phase must answer a real question before the next one is justified.
Map the current workaround
Speak with target roles about inputs, handoffs, review time, tools, and consequences of missed or distorted feedback.
Gate · a repeated, costly job is evidencedTest one evidence path
Use synthetic material first. Test source linking, dissent handling, reviewer effort, and failure states.
Gate · reviewers can verify and reject draftsOnly with explicit data rules
A supervised pilot is a future option only after consent, access, retention, deletion, and model-provider handling are agreed.
Gate · participant approval and privacy review08 / Project facts
A separate project with a clear stage—and open questions.
QuoteWeft is presented here as its own independent project. This page does not assign it to another company or claim a registered legal entity.
PROJECT NAMEQuoteWeft
Working product/project name; final naming and trademark clearance remain open.
STARTEDFebruary 2023
Founder-provided project start date—not a claim about company incorporation.
PURPOSESource-linked customer-research synthesis
Help small software teams inspect how raw feedback became a proposed theme.
STAGELocal synthetic demo
No live application, customers, revenue, integrations, funding, or performance claims are stated on this page.
LEGAL IDENTITYNot specified here
No company name, registration date, or corporate relationship is represented as fact.
MODEL ROLEPossible language-assistance layer
Claude is a candidate for a future API integration; no model integration is live.
09 / Open questions
Specific unknowns—not pretend certainty.
These are the decisions and risks that should be answered through discovery before the scope is expanded.
Q1Which role feels the pain often enough to own a recurring workflow?
Q2What source formats cover the first useful case without overbuilding integrations?
Q3How should consent, deletion, and model-provider retention work for research data?
Q4Does source traceability save time, or merely move effort into review?
Q5Which languages, translation steps, and segment boundaries matter first?
Q6Who is the buyer, and is a team subscription a real willingness-to-pay fit?
10 / Plain answers
Status, scope, and what this page does not claim.
Is QuoteWeft available to use?
No. This is an informational project site. There is no product login, signup, upload form, or pilot intake.
Is this project connected to another company?
No company affiliation is represented here. QuoteWeft is presented as an independent project; legal entity details are not supplied on this page.
What happened in February 2023?
That is the founder-provided project start date. It is not presented as an incorporation date.
Does the project already use Claude?
No Claude API integration is live. Claude is a candidate for a future API integration; the local prototype uses synthetic data and makes no model call.
Are privacy safeguards already implemented?
No. Access, consent, retention, deletion, and model-provider handling are requirements to design and verify before any real-data pilot, not present guarantees.
Independent project · early stage
Keep the trail. Keep the right to disagree.
The next step is not a bigger feature list. It is evidence that one real research workflow is painful enough—and that a source-linked draft makes it better.