Yukti documentation

Strengths, weaknesses, opportunities and threats as claim rows, each mandatory-sourced. An LLM-assisted first draft is available with a model key configured; hand-entered analysis needs none.

Needs: LLM key, only for the AI-assisted first draft

Will not: Store an item with no source kind, detail, citation or derivation. Draft a SWOT with no model configured — that route returns a named 503 instead of a plausible-looking guess.

Every strength×opportunity, strength×threat, weakness×opportunity and weakness×threat crossing is returned as the method's own question, never a generated recommendation. An action the owner keeps names the two claims it came from, so it can be walked back to its evidence.

Will not: Invent the move — the crossing is a question the owner answers, not an answer Yukti supplies. Set an impact or feasibility score itself — those columns stay null unless the owner sets them.

Business Model Canvas, go-to-market, buyer persona, OKRs, PESTEL, Porter's Five Forces and Lean Canvas — each cell a question the owner answers in their own words, stored as a sourced claim. A cell nobody answered stays empty.

Will not: Fill a cell nobody answered, or mark an empty canvas as complete.

PESTEL and Porter's Five Forces are canvases filled from what the owner has personally seen, each cell a sourced claim — the same interview machinery as the canvas library, with different questions.

Needs: a grounded retrieval source for political, economic and regulatory facts

Will not: Generate the political, economic, social, technological, environmental or legal factors from world knowledge, or score the five forces itself — that needs a grounded source this deployment does not have.

TAM/SAM/SOM computed from five things about your own business — never an industry report — with the working shown as arithmetic. A published top-down total can be added afterward as a cross-check, but only with a link, a publisher and the date it was read.

Will not: Accept a top-down market total without a publisher, a URL and a retrieval date — that is a 400, not a warning. Average the bottom-up and top-down numbers into a third figure — they are shown side by side, because a blended number would belong to neither method.

A month-by-month cash projection from five required assumptions, with an optional days-sales-outstanding lag so revenue earned is not counted as cash received. Break-even is reported as a revenue level, not just a month. Sensitivity moves one assumption at a time and returns the perturbation used, so the ranking is reproducible. Actuals are recorded as owner records and compared as variance — never rewritten into the plan.

Will not: Record an actual from anything but owner_record — an owner_estimate is refused with a 400 for this route specifically. Rewrite the plan to match what happened — a projection is what you expected, a record is what happened, and merging them would leave a forecast that can never be shown to have been wrong. Default the payment-delay assumption to a plausible industry figure — it defaults to zero (money arrives the month it is earned) because there is no sourced benchmark on this deployment.

Milestones with real dependency links, not an ordering column, so a reordered plan stays correct. The conflicts route reports circular dependencies, steps waiting on unscheduled steps, dependencies due after the thing waiting on them, and anything past the launch date.

Will not: Ship with a templated timeline — a new plan starts with zero milestones, because how long a trade licence or an FSSAI registration takes varies by state and this deployment has no citable source for it. Guess how long any step takes — every date is the owner's to supply.

A cadenced sweep of what has gone stale and what rests on nothing checkable, delivered through Wavy's own consent-gated sender. It now aggregates competitor-tracker, pricing-benchmark and opportunity-scanner sections alongside the original freshness sweep, and persists what it composed so past briefs read back as real history rather than only a last-result string. It stays silent, and unpersisted, when there is nothing to report.

Will not: Send to anyone without a recorded opt-in — no opt-in means no send. Fabricate a competitor or pricing move to fill an empty section — it reports only what competitor-tracker and pricing-benchmark actually have on file, which can genuinely be nothing.

Answers questions from the workspace's own stored SWOTs, market sizing, forecasts, canvases and launch plans, and refuses to answer from anything else. Conversations and their citations now persist, so a past chat and its sources can be read back rather than only the last answer. Keyword-ranked citations, the regulated-advice gate and the stored-claims fallback all run with no credential — only generating the answer itself needs one.

Needs: LLM key, for generating the answer itself

Will not: Answer from general knowledge or anything not already stored in this workspace's own strategy artifacts. Generate an answer with no model configured — that returns a named 503 rather than a guess.

Freshness state and the estimate-versus-published-source split across every stored artifact, plus a re-verification action. Version snapshots and diffs are built too: taking a version is a deliberate action carrying a required reason for the change, the same discipline re-verification already uses, rather than an automatic save on every claim write. The diff compares field by field, by claim slot, not as text, so a reformatted figure never reads as a changed one.

Will not: Snapshot a version automatically on every claim write — that would version every keystroke; a version needs a stated reason. Report a reformatted figure as a changed one — the diff compares by claim slot, not by comparing raw text.

Workspace member. A slot outside strength/weakness/opportunity/threat is a 400.

{ "name": "Q3 SWOT", "context": "Home-decor shop in Pune, 2 staff.",
  "items": [ { "slot": "strength", "text": "Same-day delivery inside Pune",
               "source": { "kind": "owner_input", "detail": "We run our own bike delivery." } } ] }

{ "id": "9c1e…", "kind": "swot", "name": "Q3 SWOT", "status": "draft",
  "items": [ { "id": "…", "slot": "strength", "text": "Same-day delivery inside Pune",
               "source": { "kind": "owner_input", "says": "You told us: We run our own bike delivery.",
                           "checkable": false } } ] }

Workspace member. Strengths, weaknesses, opportunities and threats as claim rows, each mandatory-sourced. An LLM-assisted first draft is available with a model key configured; hand-entered analysis needs none.

{ "pairs": [ { "pair_kind": "strength_opportunity",
               "question": "How could you use this strength to take this opportunity?",
               "strength": {…claim…}, "opportunity": {…claim…} }, … up to 40 ] }

Workspace member. Every strength×opportunity, strength×threat, weakness×opportunity and weakness×threat crossing is returned as the method's own question, never a generated recommendation. An action the owner keeps names the two claims it came from, so it can be walked back to its evidence.

{ "pair_kind": "strength_opportunity", "primary_claim_id": "…", "secondary_claim_id": "…",
  "title": "Offer same-day delivery as a paid rush option",
  "rationale": "Uses the bike fleet we already run.", "owner_name": "Vivek" }

{ "id": "…", "pair_kind": "strength_opportunity", "title": "Offer same-day delivery as a paid rush option",
  "status": "proposed", "impact_score": null, "feasibility_score": null,
  "primary_claim_id": "…", "secondary_claim_id": "…" }

Workspace member. Business Model Canvas, go-to-market, buyer persona, OKRs, PESTEL, Porter's Five Forces and Lean Canvas — each cell a question the owner answers in their own words, stored as a sourced claim. A cell nobody answered stays empty.

{ "templates": [ { "framework": "business_model_canvas", "title": "Business Model Canvas",
    "cells": [ { "slot": "customer_segments", "label": "Customer segments",
                 "question": "Who exactly do you serve? …" }, … 9 cells ] }, … 7 frameworks ] }

Workspace member. Business Model Canvas, go-to-market, buyer persona, OKRs, PESTEL, Porter's Five Forces and Lean Canvas — each cell a question the owner answers in their own words, stored as a sourced claim. A cell nobody answered stays empty.

{ "slot": "value_propositions", "text": "Same-day delivery inside Pune at no extra cost.",
  "source": { "kind": "owner_input", "detail": "How we describe ourselves to customers today." } }

{ …artifact with the updated cell… }

Workspace member. PESTEL and Porter's Five Forces are canvases filled from what the owner has personally seen, each cell a sourced claim — the same interview machinery as the canvas library, with different questions.

{ "framework": "pestel_analysis", "name": "PESTEL — home decor" }

{ "id": "…", "kind": "canvas", "name": "PESTEL — home decor" }

Workspace member. TAM/SAM/SOM computed from five things about your own business — never an industry report — with the working shown as arithmetic. A published top-down total can be added afterward as a cross-check, but only with a link, a publisher and the date it was read.

{ "name": "Home decor — Pune", "inputs": [
  { "slot": "potential_customers", "value_numeric": 40000, "unit": "customers",
    "source": { "kind": "owner_estimate", "detail": "Pune households in our delivery radius, per city data." } },
  { "slot": "purchases_per_year", "value_numeric": 2, "unit": "purchases_per_year",
    "source": { "kind": "owner_estimate", "detail": "Based on repeat customers so far." } },
  { "slot": "price_per_purchase_minor", "value_numeric": 250000, "unit": "INR_minor",
    "source": { "kind": "owner_record", "detail": "Average order value from our own sales." } },
  { "slot": "serviceable_share_pct", "value_numeric": 30, "unit": "percent",
    "source": { "kind": "owner_estimate", "detail": "Delivery radius covers about a third of the city." } },
  { "slot": "capture_rate_pct", "value_numeric": 5, "unit": "percent",
    "source": { "kind": "owner_estimate", "detail": "Conservative first-year guess." } } ] }

{ "id": "…", "estimate": { "tam_minor": 200000000000, "sam_minor": 60000000000, "som_minor": 3000000000,
    "method": "bottom_up", "working": [ "TAM = 40000 customers × 2 purchases/year × 250000 paise = …", … ],
    "cautions": [], "basis": "Built bottom-up from your own inputs. …" },
  "insufficient_data": false }

Workspace member. Omit the publisher, URL or retrieval date and this is a 400: "A published figure needs the link, who published it, the date you read it."

{ "total_market_value_minor": 5000000000000,
  "source": { "kind": "published_source", "publisher": "FICCI Home Decor Report 2026",
              "url": "https://example.org/ficci-report-2026", "retrieved_on": "2026-08-10" } }

{ … "cross_check": { "bottom_up_tam_minor": 200000000000, "top_down_total_minor": 5000000000000,
    "ratio_top_down_to_bottom_up": 25.0,
    "reading": { "kind": "yukti_inference", "text": "These two methods disagree by more than double. …" } } }

Workspace member. A month-by-month cash projection from five required assumptions, with an optional days-sales-outstanding lag so revenue earned is not counted as cash received. Break-even is reported as a revenue level, not just a month. Sensitivity moves one assumption at a time and returns the perturbation used, so the ranking is reproducible. Actuals are recorded as owner records and compared as variance — never rewritten into the plan.

{ "name": "12-month cash plan", "months": 12, "assumptions": [
  { "slot": "starting_cash_minor", "value_numeric": 50000000, "unit": "INR_minor",
    "source": { "kind": "owner_record", "detail": "Bank balance today." } },
  { "slot": "monthly_revenue_minor", "value_numeric": 80000000, "unit": "INR_minor",
    "source": { "kind": "owner_record", "detail": "Last month's sales." } },
  { "slot": "monthly_revenue_growth_pct", "value_numeric": 5, "unit": "percent",
    "source": { "kind": "owner_estimate", "detail": "Expecting steady growth." } },
  { "slot": "monthly_fixed_costs_minor", "value_numeric": 40000000, "unit": "INR_minor",
    "source": { "kind": "owner_record", "detail": "Rent, salaries, subscriptions." } },
  { "slot": "variable_cost_pct_of_revenue", "value_numeric": 35, "unit": "percent",
    "source": { "kind": "owner_estimate", "detail": "Stock and commission, roughly." } } ] }

{ "id": "…", "projection": { "months": 12, "rows": [ {"month":1,"revenue_minor":80000000,
    "cash_in_minor":80000000,"closing_cash_minor":88000000, …}, … ],
    "summary": { "closing_cash_minor": …, "first_month_cash_runs_out": null,
                 "first_profitable_month": 1, "break_even_monthly_revenue_minor": 61538462 },
    "break_even": { "monthly_revenue_minor": 61538462, "means": "In a month where you take in this much, what comes in exactly covers what goes out. …" },
    "receivables": { "days_sales_outstanding": 0, "supplied": false, "means": "Money is counted as arriving the month it is earned, because no payment delay has been recorded. …" } } }

Workspace member. A month-by-month cash projection from five required assumptions, with an optional days-sales-outstanding lag so revenue earned is not counted as cash received. Break-even is reported as a revenue level, not just a month. Sensitivity moves one assumption at a time and returns the perturbation used, so the ranking is reproducible. Actuals are recorded as owner records and compared as variance — never rewritten into the plan.

{ "perturbation_pct": 10, "base_closing_cash_minor": …,
  "ranked": [ { "rank": 1, "slot": "monthly_revenue_growth_pct", "label": "Monthly revenue change",
                "delta_up_minor": …, "delta_down_minor": …, "swing_minor": …, "moved_the_answer": true }, … ],
  "method": "Each assumption was moved up by 10% and then down by 10%, on its own, with the other four held where you put them, and the 12-month projection re-run each time. …" }

Workspace member. Supply source.kind = owner_estimate here and it is a 400: "An actual has to come off your own records."

{ "month": 1, "metrics": [
  { "metric": "revenue_minor", "value_numeric": 76000000,
    "source": { "kind": "owner_record", "detail": "Read off the till roll for the month." } } ] }

{ "recorded": […], "months": [ { "month": 1, "in_projection_horizon": true,
    "metrics": [ { "metric": "revenue_minor", "actual_minor": 76000000, "projected_minor": 80000000,
                   "variance_minor": -4000000, "variance_pct": -5.0, "direction": "actual_lower" } ] } ],
  "records_only": "Every figure under 'actual' came off your own records and says which. Nothing here was estimated, and nothing here changed your plan." }

Workspace member. Milestones with real dependency links, not an ordering column, so a reordered plan stays correct. The conflicts route reports circular dependencies, steps waiting on unscheduled steps, dependencies due after the thing waiting on them, and anything past the launch date.

{ "title": "Get FSSAI registration", "due_date": "2026-09-15", "owner_name": "Priya" }

{ …milestone… }

Workspace member. Milestones with real dependency links, not an ordering column, so a reordered plan stays correct. The conflicts route reports circular dependencies, steps waiting on unscheduled steps, dependencies due after the thing waiting on them, and anything past the launch date.

{ "plan_id": "…", "conflicts": [
  { "kind": "depends_on_unscheduled", "milestone_id": "…", "blocking_id": "…",
    "message": "'Launch event' waits on 'Get FSSAI registration', which has no date yet. …" },
  { "kind": "circular_dependency", "milestone_ids": ["…","…"],
    "message": "These steps wait on each other in a loop, so none of them can ever start: A → B → A." }
  ], "total": 2 }

Workspace member. A cadenced sweep of what has gone stale and what rests on nothing checkable, delivered through Wavy's own consent-gated sender. It now aggregates competitor-tracker, pricing-benchmark and opportunity-scanner sections alongside the original freshness sweep, and persists what it composed so past briefs read back as real history rather than only a last-result string. It stays silent, and unpersisted, when there is nothing to report.

{ "contact_id": "…", "cadence_days": 7 }

{ "id": "…", "cadence_days": 7, "active": true, "last_sent_at": null }

Workspace member. No model configured returns a named 503 rather than a guess.

{ "question": "What did we say our biggest weakness was?" }

{ "answer": "…drawn only from your stored SWOT claims…", "sources": [ "…claim ids…" ] }

Workspace member. Freshness state and the estimate-versus-published-source split across every stored artifact, plus a re-verification action. Version snapshots and diffs are built too: taking a version is a deliberate action carrying a required reason for the change, the same discipline re-verification already uses, rather than an automatic save on every claim write. The diff compares field by field, by claim slot, not as text, so a reformatted figure never reads as a changed one.

{ "artifacts": [ { "id": "…", "kind": "market_size", "state": "stale", "age_days": 210,
    "checkable_claims": 3, "estimate_claims": 2 } ] }

Workspace member. Freshness state and the estimate-versus-published-source split across every stored artifact, plus a re-verification action. Version snapshots and diffs are built too: taking a version is a deliberate action carrying a required reason for the change, the same discipline re-verification already uses, rather than an automatic save on every claim write. The diff compares field by field, by claim slot, not as text, so a reformatted figure never reads as a changed one.

{ "changed": [ { "slot": "monthly_revenue_minor", "from": 80000000, "to": 90000000 } ],
  "unchanged_count": 4 }