Economics ? selected task, project, period or product interest¶
Keep one actual tfw-economics-v1 report metadata block when rendering through the helper.
It is a reproducible view, never work authority. Unknowns remain null with their causes.
The standard analytical workflow owns interpretation; this is its adaptable report model.
Use a concise opening for routine work; expand the sections below only when they help the selected
decision. The standalone helper produces a fixed numeric input and does not read this template.
Executive summary¶
Lead with the question, intended value, actual achieved/accepted result and supported economic conclusion or decision. Prepared, checked, independently accepted and released remain distinct. Use status/HL/RF/REVIEW or the selected Daily record and product result as applicable. Benefit is described through outputs; numeric ROI, quality/efficiency or prevented loss needs independent data.
Compact pattern: question → result and comparison basis → approximate resources with material coverage → justified action/limit. Synthetic example: “Should we pilot a different handoff practice? Comparable accepted outputs used about 100 versus 70 reference units at the same cycle checkpoint, with comparable measured coverage. Run a bounded pilot and check accepted quality and resource use again; these examples establish no real savings or universal superiority.” In an actual report replace illustrative amounts with sourced observations and link their epochs.
Product map¶
Preserve native primary area/keywords. Cite grounds for domains, subject areas, features, user scenarios and delivery context; label derived classification. One primary area is additive. Overlapping interests filter a deduplicated work union. Without attributable records, figures are associated-task resources, never feature costs. Keep project/root/phase context distinct.
Numerical detail¶
Show exact membership/exclusions and useful role/unit/model/date distributions with denominators. Input includes cache; reasoning is an output subset where source meaning establishes it. Separate tokens, dated Standard API equivalents and compatible duration kinds. State actual price-card epoch, unpriced values and that this is not subscription billing. Consumption-period dates differ from evidenced task-event cohorts with lifetime/undated usage. Current state, cutoff and generic last update do not manufacture consumption/completion dates. Keep the common semantic checkpoint beside totals: final role returns and Coordinator integration/close capture after relevant results/checks, before final owner delivery. Label planning/open-cycle estimates interim. Preserve exact observations and round only display. The owner's practical 5–7% tolerance is decision guidance; unknown error has no guaranteed band. Material missing roles, stale selection and incompatible metrics remain visible.
Material observations¶
Show only comparisons that explain the request: concentration, token/cost/cache mix, supported date changes, iterations/scope or comparable outputs. Retain checked calculations/source epochs. Separate observation, interpretation, causal hypothesis and next check. Statistical outliers name a comparable baseline and method; expensive or unusual does not itself prove waste.
For people/team/tool/process comparisons, cite actual work membership and scope/complexity, accepted output/quality, material rework, model/effort, coverage, valuation basis and available human-effort evidence. An account identity supplies no team membership; AI usage supplies no human time. Missing comparison dimensions stay unavailable.
Illustrative reference amounts below are synthetic, not observed tool/team performance or prices. In every case verify compatible cycle bounds and metric meanings before applying the inference.
| Question and comparison basis | Resources and coverage | Supported action and limit |
|---|---|---|
| Pilot a different tool/process for comparable accepted outputs, with comparable rework and context? | 100 versus 70 reference units, both bounded with comparable coverage | Run a bounded pilot and check accepted quality/resource use again; no universal superiority or measured ROI follows. |
| Prefer Team A / Codex over Team B / Claude Code for comparable accepted work? | 100 versus 104; uncertainty plausibly explains the difference at the 5–7% tolerance | No confident preference; keep the current choice. Tokens do not rank people. |
| Is Team C / Antigravity less efficient on materially harder accepted work? | 140 versus 100; workload differs | No efficiency ranking; select comparable work before deciding allocation or tool choice. |
| Change practice from a complete cohort with a materially missing contribution? | Selected total known, missing coverage magnitude unknown | No conclusion/change on this basis; collect the missing input only if it could change the decision. |
TFW improvements¶
Propose grounded actions affecting context, iteration, scope, handoff, collection or verification. State the observed basis, intended improvement, next check and actual authority. When no grounded action is warranted, state that directly; invent no mandatory recommendation or savings estimate.
Coverage and sources¶
List received/measured/missing/failure-only and omitted units; selected/excluded returned hashes, revisions/ranges and exact bound native source identities. Show latest failure code/detail, attempted interval, nonmeasured binding gaps and usable last-good contribution. Unknown usage, handling cost and duration are not observed zero. Keep collector operation, full invocation/ integration and native time separate; calendar elapsed comes only from applicable work records. Complete selected bytes do not prove complete work. Explain cutoff, partial/undated/unpriced coverage and one finite final-delivery/short-exchange/cleanup tail; substantive reopening is a separately bounded continuation. A remote pathname alone is not receipt. Put expected/received units separately from selected source/revision/range beside the estimate, even when every unit returned. Read Full prose binding gaps from owned artifacts; do not claim numeric helper receipt or measurement for them. Accountability does not waive an explicit measurement acceptance gate.
Presentation and existing close¶
Prose and Markdown tables carry the whole report. Visuals supplement supported comparisons only when the selected host works. PDF is explicit-request-only, with a verified absolute OS-temp output outside the repository and inspected render; unavailable rendering leaves complete text. Temporary calculations/render scripts are also outside the repository, not a shipped script suite. Formal roles retain their return/acceptance gates; this report grants no new role or acceptance. Prepare the report basis before terminal acceptance, retain needed resources, then follow the existing Coordinator-owned DONE/final-report/safe-cleanup order. Daily links its final snapshot and concise summary without fabricated Full state.
Used price basis and report epoch¶
Link task-owned used quote file/hash, exact model/provider/conditions, currency/unit/category rates, checked timestamp/source URL/hash, publication and known/unknown effective date. Distinguish recorded-basis valuation, official historical reconstruction and current revaluation. Unknown request band/tier/region produces named conditional scenarios, not an invented actual allocation. Show role/model/effort and token-category money detail when supported. Retain prior report/basis and date epochs; offline reproduction uses captured bytes, never mutable home prices. The 30-day lookup policy is not evidence of unchanged market price. API equivalent is never a subscription bill.