Earnings Watch
Read-only. reference/READ-ONLY-POLICY.md (hard rule) and
reference/RESEARCH-STANDARDS.md (freshness, data efficiency,
disclosure) apply. Steps below name capabilities — resolve each against
reference/BROKER-CAPABILITIES.md for the broker(s) in BROKERS.md.
No broker's MCP configured here has an earnings calendar — dates and
results content come from WebSearch (NSE/BSE announcements, company IR,
business press), tied back to live price and technical data (technicals
are Groww-only today).
Steps
Scope to what's owned. Equity holdings capability, from every
active broker, plus any name the user adds, anything on the
PORTFOLIO-PLAN.md watchlist, and (if INDmoney is active) its saved
watchlist capability — unverified, so confirm before trusting
that one. Don't build a market-wide calendar. Read the plan's
position theses while you're there — their invalidators are what
step 3 tests.
Upcoming results. Date-anchored WebSearch ("Q2 FY26 results date
"), batched sensibly: one sector/index results-calendar
search for the larger names before per-name searches for the rest.
Present as a dated calendar, nearest first.
For results already announced:
WebSearch the actual numbers — revenue/EBITDA/PAT vs. analyst
consensus where findable (not just YoY/QoQ), management
commentary, guidance changes. When consensus isn't findable, say so
explicitly rather than silently comparing to YoY only — a beat on
YoY that misses consensus is a miss, not a beat. Cite outlet + date.
- Price reaction: batched LTP capability, and where the reaction
matters to the verdict, the technical-indicators / historical-candles
capabilities (daily) for the post-result move vs. the prior trend —
a good result sold into is a different signal from a good result
bought. Groww-only today.
- Thesis check: test the result against the recorded invalidator
in
PORTFOLIO-PLAN.md — hit, approached, or clear? Quote the
invalidator and the number that meets or misses it. Where none is
recorded, say so and offer portfolio-plan-builder: "the thesis
still holds" means nothing if it was never written down.
Flag what needs action. Only holdings where the result — or an
imminent one plus stretched positioning — changes the picture get a
view, per the completeness checklist. Routine in-line results are one
line each.
Present: upcoming calendar first (date, name, portfolio weight),
then reviewed results (one line each, detail only for flagged names),
then views with the disclosure block. stock-research re-underwrites
a single name. Formal version: reference/templates/earnings-watch.md.
1---2name: earnings-watch3description: Track upcoming results dates for the user's holdings and review announced results — vs. expectations, price reaction, thesis check. Use when the user asks about upcoming earnings/results, "who reports this week", or how a holding's results were. Read-only.4---56# Earnings Watch78Read-only. `reference/READ-ONLY-POLICY.md` (hard rule) and9`reference/RESEARCH-STANDARDS.md` (freshness, data efficiency,10disclosure) apply. Steps below name capabilities — resolve each against11`reference/BROKER-CAPABILITIES.md` for the broker(s) in `BROKERS.md`.1213No broker's MCP configured here has an earnings calendar — dates and14results content come from `WebSearch` (NSE/BSE announcements, company IR,15business press), tied back to live price and technical data (technicals16are Groww-only today).1718## Steps19201. **Scope to what's owned.** Equity holdings capability, from every21 active broker, plus any name the user adds, anything on the22 `PORTFOLIO-PLAN.md` watchlist, and (if INDmoney is active) its saved23 watchlist capability — unverified, so confirm before trusting24 that one. Don't build a market-wide calendar. Read the plan's25 **position theses** while you're there — their invalidators are what26 step 3 tests.27282. **Upcoming results.** Date-anchored `WebSearch` ("Q2 FY26 results date29 <company>"), batched sensibly: one sector/index results-calendar30 search for the larger names before per-name searches for the rest.31 Present as a dated calendar, nearest first.32333. **For results already announced:**34 - `WebSearch` the actual numbers — revenue/EBITDA/PAT **vs. analyst35 consensus where findable** (not just YoY/QoQ), management36 commentary, guidance changes. When consensus isn't findable, say so37 explicitly rather than silently comparing to YoY only — a beat on38 YoY that misses consensus is a miss, not a beat. Cite outlet + date.39 - Price reaction: batched LTP capability, and where the reaction40 matters to the verdict, the technical-indicators / historical-candles41 capabilities (daily) for the post-result move vs. the prior trend —42 a good result sold into is a different signal from a good result43 bought. Groww-only today.44 - **Thesis check:** test the result against the recorded invalidator45 in `PORTFOLIO-PLAN.md` — hit, approached, or clear? Quote the46 invalidator and the number that meets or misses it. Where none is47 recorded, say so and offer `portfolio-plan-builder`: "the thesis48 still holds" means nothing if it was never written down.49504. **Flag what needs action.** Only holdings where the result — or an51 imminent one plus stretched positioning — changes the picture get a52 view, per the completeness checklist. Routine in-line results are one53 line each.54555. **Present:** upcoming calendar first (date, name, portfolio weight),56 then reviewed results (one line each, detail only for flagged names),57 then views with the disclosure block. `stock-research` re-underwrites58 a single name. Formal version: `reference/templates/earnings-watch.md`.