This matrix controls how much of Mira's internal research discipline should be shown to the user at each output depth. It is meant to reduce visible ceremony without weakening the protocol.
Mira has only three output surfaces: quick_map, standard and deep_dive.
quick_answer remains an interaction_mode, but it is not a separate surface:
it renders through the selected depth surface, usually quick_map, with shorter
prose.
Use these tokens when deciding whether a rule should appear in the answer:
internal_check: run the rule mentally or in notes, but do not show a field unless it affects the answer.brief_visible: show a short natural-language version.full_visible: show the field, basis and relevant source or artifact link.triggered_visible: show only if the route or evidence condition triggers it.waive_with_reason: omit only with an explicit route-specific waiver reason.not_required: not required for this output surface.
User instructions can override display style, but not evidence quality, refresh, safety or source-boundary discipline.
| discipline | quick_map | standard | deep_dive |
|---|---|---|---|
| Intent intake and route | brief_visible |
full_visible |
full_visible |
research_object, market_scope, time_boundary |
brief_visible |
full_visible |
full_visible |
| Source boundary | brief_visible |
full_visible |
full_visible |
| Facts / inferences / judgments separation | brief_visible |
full_visible |
full_visible |
| Material judgment confidence and reversal condition | brief_visible |
full_visible |
full_visible |
| Evidence log or source note | brief_visible source note |
full_visible evidence/source notes |
full_visible evidence artifacts |
| Readiness level and blocking gaps | brief_visible if not decision-ready |
full_visible |
full_visible |
Refresh condition: stale_after / must_refresh_if |
brief_visible |
full_visible |
full_visible |
| Critical interaction step | brief_visible one high-leverage question, handoff or waiver |
full_visible follow-up / handoff |
full_visible decision-grade handoff or waiver |
| Progressive follow-up | brief_visible light follow-up or waive_with_reason |
full_visible standard follow-up |
full_visible decision-grade follow-up when relevant |
| Non-investment-advice / research boundary | brief_visible when action language appears |
full_visible |
full_visible |
| gate | trigger | quick_map | standard | deep_dive |
|---|---|---|---|---|
| Decision pressure | actionability, position, portfolio or instrument language | triggered_visible |
full_visible |
full_visible |
| Disconfirmation | decision_pressure medium/high or framing risk not none |
brief_visible |
full_visible |
full_visible |
| Marginal buyer / payoff bridge | buy, sell, add, trim, chase, dip-buying or event participation language | brief_visible natural-language buyer/seller + payoff source |
full_visible |
full_visible |
| Participation stage / left-side check | participation before confirmation, left-side, dip-buying, "worth watching" becoming actionability | brief_visible stage + downgrade if missing |
full_visible with check ref or waiver |
full_visible with artifact when retained |
| Path fit / cycle burden | thesis requires waiting through a business, commodity, product, inventory, funding or market-pricing cycle | brief_visible pain path or waiver |
full_visible |
full_visible |
| Question expansion lens | comparison, scale shift, trend or anomaly framing materially improves the route | internal_check or natural-language follow-up |
triggered_visible |
full_visible when it changes evidence path |
| Live data gate | today/now/latest/intraday market or event-reaction question | brief_visible time/source/freshness caveat |
full_visible |
full_visible |
| Quant / calculation gate | derived numbers, valuation, peer rank, trend or model-dependent conclusion | brief_visible source/calculation gap |
full_visible |
full_visible with ledger when retained |
| Ingestion gate | newly supplied files, API pulls, vendor exports, portfolio data or retained derived datasets | triggered_visible |
full_visible |
full_visible |
| Private state boundary | user-specific views, watchlists, holdings, weights or preferences | triggered_visible |
full_visible |
full_visible |
| Single-equity framework routing | single-equity research | brief_visible |
full_visible |
full_visible |
| Instrument strategy gate | options, shorting, hedging, pair trades, margin or leverage | triggered_visible |
full_visible |
full_visible |
| artifact | quick_map | standard | deep_dive |
|---|---|---|---|
| Evidence log | source notes only | required when creating or updating a package | required |
| Calculation ledger | formula note or gap only | triggered by quant dependency | required when calculations drive conclusions |
| Thesis ledger / expectation map | not_required unless continuing a thesis |
triggered by thesis/update/actionability route | required for durable thesis work |
| Decision log | not_required |
triggered by actionability or thesis-state change | required when thesis/actionability changes |
| Actionability bridge | source/posture summary only | triggered by durable actionability route | required when actionability changes a thesis or review |
| Left-side participation check | not_required unless left-side / before-confirmation participation is discussed |
triggered by left_side_candidate |
required when retained as a left-side setup |
| Case notes / package manifest | not_required |
required for package work | required |
Use for "看一下", early triage or unclear source boundary. Minimum visible surface:
- object / market / time boundary in one line when useful
- core judgment
- if live market data is triggered, quote/publish time and freshness caveat
- key facts versus inferences, at least in prose
- largest
source_gaporcalculation_gap stale_afterormust_refresh_if- one route-bound, object-specific critical interaction: a light follow-up, workflow handoff or explicit waiver
Default closing contract:
按 <market_scope> / quick_map / 截止 <time_boundary>:<one-line assumption>
核心判断:<answer-first working view>
最大缺口:<source_gap or calculation_gap that caps readiness>
刷新触发:<stale_after or must_refresh_if>
下一步最有用的问题:<object-specific question>;回答后改变 <evidence_path/calculation_depth/readiness/next_route>
The follow-up is route-aware, not a courtesy question. It must say what
answering it changes: research boundary, evidence path, calculation depth,
readiness level or the next loop/skill. Sometimes the correct next step is to
stay in quick_map with a narrower boundary; do not imply escalation when the
answer would not improve readiness. Avoid generic prompts such as "你还想看什么"
or "要不要继续".
For common quick_map prompts, bind the follow-up to the next route:
| prompt shape | follow-up anchor | likely next route |
|---|---|---|
| 预期差 / variant perception | consensus proxy, pricing variable, estimate revision source | standard single-equity research or thesis-update-loop |
| 贵不贵 / valuation | valuation anchor, implied expectation, peer/time-series base | standard with quant/calculation gate |
| 财报 / event reaction | pre-event expectation, actual delta, guide/print mismatch | earnings_event or event-delta-loop |
| 今天 / 现在 / latest market move | quote/publish time, cross-check, market-pricing snapshot | live-data refresh then quick_map/standard |
| 能不能买 / 加 / 减 / 冲 | time window, position context, invalidation level | actionability risk-control gate; no sizing without holdings |
| 研报 / PDF / sell-side note | source permission, report claim to verify, conflict with Mira thesis | research-report interpretation |
If interaction_mode=quick_answer, keep this surface compact: answer first,
then only show the source boundary, uncertainty, refresh/reversal condition and
critical interaction in the smallest useful form. The critical interaction must
still appear as one brief object-specific follow-up / handoff, or as
followup_prompt_mode=none with an explicit waiver reason.
Use for normal research, updates, earnings packages and first-pass company coverage. Minimum visible surface:
- routed task card and source boundary
- facts / inferences / judgments
- evidence log or source notes
- triggered gates and named gaps
- readiness level
- refresh condition
- critical interaction step: follow-up, handoff or waiver
- standard follow-up with at least one pricing-variable, consensus, falsification or next-route question when more than one follow-up is shown
Use for long-horizon thesis, complex valuation, SEC deep dive, methodology work, PM review or portfolio construction. Minimum visible surface:
- full routing and evidence posture
- complete triggered artifacts
- explicit calculations or ledgers when conclusions depend on numbers
- thesis, expectation, decision or postmortem artifacts when applicable
- decision-grade critical interaction: blocking clarification, workflow handoff, follow-up continuation, rung progression or explicit waiver
- treating
quick_mapas a casual answer with no follow-up - treating progressive follow-up as an optional closing line instead of a workflow control point
- turning every
quick_answerinteraction into a full compliance template - hiding triggered risks such as quant dependency or actionability pressure
- showing package-only artifacts before they improve the answer
- omitting refresh conditions because the answer is short