Standalone owner-directed analysis uses the Coordinator's ordinary analytical scope. Task-bound analysis preserves the current Coordinator, Researcher, Executor or Reviewer Role Lock, existing authority and sole return route. It creates no formal role, lifecycle or acceptance grant.

Read Contract

Read this workflow completely. Select the question and receiving rules/preferences, then only the necessary selected task/phase status/HL and returned RF/REVIEW or explicit Daily record/product. Read .tfw/economics/README.md before numerical operations. Read returned numeric bytes and the task-owned used price basis; original native data belongs only to its exact producing unit. Use .tfw/templates/economics.md for the report after evidence/selection resolve. No manifest, full common library, foreign native session or unrelated task history is a runtime input.

TFW Economics — Product and Resource Analysis

Mindset: Connect intended and achieved work to observed resources, then explain the decision the reader can make. Tokens, money equivalents, time and product benefit answer different questions. Adapt depth to the request and available grounds; a small report can be complete. No quota of anomalies, charts or recommendations justifies invented content.

Authority: This standard analytical workflow reads authorized work records and returned numeric contributions. It grants no formal role, lifecycle, acceptance, deployment or foreign-source access. An existing Full role keeps its Role Lock and vertical reporting route. Ordinary owner-directed analysis replies in chat. Read no other unit's native sessions or unreturned working state.

1. Resolve the question and evidence

Read receiving rules/preferences and selected work records. State source epochs, intended value, actual result and uncertainty; prepared, checked, accepted and released are distinct.

Resolve task/phase, project, day/month/date range, owner, roles/units/models or product interest against available facts. Give exact root/work membership and inclusion/exclusion reasons. A question about period spending selects evidenced consumption dates; a task-event cohort selects evidenced created/completed events and then lifetime usage, including undated rows. Cutoff, generic update, current state and folder date never manufacture consumption/completion dates. If ambiguity changes membership or totals, ask one focused clarification and continue independent supported analysis.

Read .tfw/economics/README.md at the selected receiving root before numerical operations. Validate existing returned bytes and reconcile revisions/source ranges once with its helper or a checked equivalent. Record exact returned file hashes and selected/excluded membership. Never add parent reports, cumulative snapshots, chart/CSV views or overlapping filtered subtotals to their facts. Unknown usage remains unknown; an empty accepted sum is not evidence of measured zero.

Read Full binding gaps in owned artifacts; the helper does not parse them. Daily uses --record. Keep attributable nonmeasurement, actual v1 failures and last-good measurement distinct. Missing/unreadable returns remain unresolved; an explicit measurement/product-proof gate still blocks.

The reporter consumes received authorized bytes, not an inaccessible producer pathname. Missing units and excluded measurements remain gaps. Show expected/received units separately from selected source/revision/range beside the estimate. Apply the common final-return/integration checkpoint; planning/open-cycle estimates are interim. Complete selected bytes do not prove complete work. Disclose the short final-delivery/exchange/cleanup tail once; substantive reopening is a separate bound. Preserve incompatible duration kinds. Dated local Standard API equivalents use the pinned rate epoch, never subscription invoices or invented current prices.

2. Resolve public quotes and preserve the used basis

For each requested exact model and applicable price profile, use the common README's local quote commands. Actual home ~/.tfw/rates/<provider>/YYYY-MM.json holds public tariff observations only, never usage/tasks/owners. Monthly filenames use the observation's UTC month. Reuse an exact quote checked less than 30 days ago, including from a previous month. File mtime and another model's fresh quote are irrelevant. Missing model/conditions, expired quote or an explicit today-check requires bounded first-party lookup through working documentation/web tools. Do not build a network client, full-provider catalogue, daemon or require account/API credentials for public prices.

Fetch the actual official page, preserve exact model/provider, currency/unit, category rates and all relevant tier/context/region/modality/cache conditions, source URL/hash and checked timestamp. Keep publication and effective dates separate; unknown effective date stays null. An alias or launch date alone does not establish historic applicability. Append via quote-store; its provider OS lock/atomic replacement preserves other writers, month history and last good files. Failed acquisition/cache writes leave the precise cause, last-known quote/date and current-valuation gap. They do not justify zero money or silent stale/current substitution. Resolve ambiguous same-time quotes before use; no model/profile guessing.

quote-select captures an immutable task-owned basis under economics/quotes/; retain the public source evidence with the task when needed for inspection. Multiple exact models can use quote-bundle, one profile per model per valuation. Use the saved basis via report/summary --rates and its recorded hash; reopening offline never reads/reprices from changed home tariffs. Preserve original JSONL, legacy cards, used quotes and report epochs. A current economics.md projection must link its retained dated report/basis; a changed quote produces a new explicit valuation.

Distinguish recorded-basis valuation, officially reconstructed historical valuation and current revaluation. Historical reconstruction needs evidenced effective date and consumption interval; unknown date/conditions remain qualified or unpriced. For unknown actual request band/tier/region, show named conditional scenarios rather than assign every aggregate to an invented actual profile. Positive unknown tariff/token categories remain unpriced. Freshness is a chosen lookup policy, not proof that a price never changed. API equivalent is distinct from subscription/credit invoices.

Full producing units follow the common source/quote contract through their own-return routes; Daily remains optional and reuses the applicable checked quote across cumulative turns. Ordinary counter changes do not trigger lookup every turn. Capture failure does not stop independently authorized useful work; leave the actual failure/gap and last-good result visible.

3. Connect product meaning and inspect useful comparisons

Preserve native primary area and keywords. With cited work/result grounds, distinguish domain, subject area, feature, user scenario and delivery context. Label derived classifications and their sources. A shared ontology is unnecessary. One primary area is additive; overlapping interests select a deduplicated union of associated tasks. Without attributable feature records, name associated-task resources, never feature cost. Keep project/work/root/phase context distinct when different projects share a task ID.

Calculate only dimensions that explain the question: resource concentration, role/unit/model mix, fresh/cache/write/output mix, supported date changes, iterations or scope changes, accepted contributions and comparable tasks. Retain denominators, units, selections and checked arithmetic. Use temporary checked calculations when the existing helper has no direct selector; explain an unavailable dimension rather than guessing. Task cohorts require separately cited event membership.

For people/team/tool/process comparisons, cite actual work membership, not account names. Compare scope/complexity, accepted quality, rework, model/effort, coverage, valuation basis and sourced human effort. Unknowns stay unavailable; AI usage is not human time or competence. Apply the common 5–7% tolerance without invented error bands or hidden gaps. Small differences support no confident preference; stronger comparable evidence can justify a bounded pilot, not universal superiority. Name the practice and next check, or justified no-change outcome. The template illustrates people, teams, Claude Code, Codex and Antigravity comparisons.

Separate an observed fact from interpretation, causal hypothesis and next check. A statistical outlier requires a named comparable baseline and method. A single-task unusual fact can be an observation. Observed concentration or expense does not prove waste, efficiency, quality or savings. Describe benefit through accepted outputs; ROI, prevented loss and quality rankings require independent measurement. Collector operation time, invocation/integration wall time and native agent/model duration remain separate; never add internal collector time again to invocation time.

4. Deliver an executive opening with inspectable detail

Use concise chat in the user's language and .tfw/templates/economics.md: question → result/ comparison basis → approximate resources with material coverage → supported action/limit. Round display, preserving exact facts/epochs. Expand detail only for the decision, not six mandatory long sections. Further precision needs a decision it could change. The helper's fixed numeric report is an input; it does not render this template.

Prose and Markdown tables carry the entire answer. Supplementary visuals are useful only when requested or they clarify a supported comparison and the selected host actually renders them. Never promise graphics from source presence or filesystem parity. PDF is request-only: verify an absolute OS-temp path outside the repository before producing it, use an available renderer and inspect the result. If unavailable, return the complete text/table report and precise renderer gap. Ad hoc analysis/render scripts also stay in OS temp; ship no permanent analytical/render script suite.

Record selection, commands/oracles, source/rate epochs and material limitations in the existing authorized trace when owed. Temporary views are derived, not a second portfolio authority. Return through the active role's Coordinator if applicable; never accept a result or continue into another formal role through this analytical skill.