August 18, 1997

Policy Assessment Corporation         CONFIDENTIAL                Perot Systems

                         PROJECT TASKS AND DELIVERABLES


The project phases are designed as self-contained tasks whose completion
verifies that step of the project. Phase I tasks use existing tools and converts
them to reflect the detailed California PX/ISO specifications and protocols.
Phase II tasks incorporate the data stream that will be available from the
PX/ISO as soon as it is known and verifies that that the system performs well
under the expected and unexpected operational conditions of the PX/ISO. Phase I
tasks fine tune the model parameters as soon as actual PX/ISQ data support such
adjustments. Further, Phase III provides support to PG&E staff as needed to
insure maximum profitability from the system. For all phases, great effort would
be made to keep PG&E staff fully cognizant of the model's technology and the
model's operation.

                                 Phase I Tasks

                                      Tasks

1.) Review PX and ISO protocols and business opportunities: This task involves a
detailed review of PX and ISO business and operational protocols and
particularly the interactions between these two entities. The main goal is to
become well versed with the ongoing changes in PX and ISO business and
operational protocols, and to become aware of which strategies are available to
competitiors in the California market. Deliverable: The strategies will be
documented.

2.) Implement the reduced scale PX/ISO operational model (CPXISO): This tasks
implements a reduced scale model that would closely approximate the operation of
California's PX and ISO systems. This task starts with existing full scale
models available to PAC and PSC. Included in this model will be the PX's energy
auction model, the ISO Congestion Management model, and the interaction between
these models. The full 6000 bus ISO linear programming model will be simulated
using a much smaller set of buses and nodes (approximately 100 buses). This ISO
model, even though it will be much smaller and much faster, will effectively
simulate the results of the actual ISO simulation and dispatch. Data interfaces
between PX, ISO, and deregulation model (CIGMOD) will also be developed. Data
models for external systems will also be developed in this task. Deliverable: A
working version of the CPXISO system. 3.) Incorporate strategies into system:
Based on Task 1, the full spectrum of strategies will be developed for each of
the participants in the market. This task will emphasize getting the "levers"
into the model to allow for the development of ever more sophisticated
strategies. The full range of strategies will be linked to the HYPERSENS
subsystem and thereby allow the exhaustive testing and selection of strategies
under uncertain future conditions.





Deliverable: Working HYPERSENS subsystem with implemented strategy levers. 4.)
Adapt PAC's short-term load forecast (SLF) model to work for California's energy
market: This task involves integrating and fine tuning the SLF model for
California competitors and developing interfaces with the PX/ISO operational
model. The system will forecast hourly demands for all players in and out of
California. The SLF model produces hourly load forecasts for the next 7 days
based on the historical loads, the historical weather, and the weather forecast.
It re-estimates the relevant coefficients on a daily or hourly basis by live
links to the weather service and existing conditions. Deliverable: Existing
system with SLF integration.

5.) Develop interfaces between CIGMOD and CPXISO: The system interface includes
system-control, PG&E data-entry, and PG&E bidding strategy output. The interface
between the model and the operators will automate the input of hourly data, will
produce timely reports and charts which effectively communicate the relevant
information, and will facilitate the operator's decision making process.
Deliverable: Existing System site tested interface.

6.) Add PG&E Proprietary Data: Any proprietary data that PG&E would like
included can be added at this time. From this point forward, the databases of
the model cannot be removed from PG&E offices without written permission.

7.) Calibrate CIGMOD/CPXISO model to California and PG&E behavior: Data for each
company and plant in the region will be incorporated into the model. This data
will be available for use in determining bidding strategies (prices and
quantities) and will be used to determine the operating income of PG&E and each
of the competitors. The data will include capacities, heat rates, fuel prices,
variable and fixed O&M, and embedded capital costs. The CIGMOD model will be
calibrated to competitor and PG&E data. The calibration process will test the
model and will estimate critical operating and financial parameters.
Deliverable: Existing model with tested simulation of all relevant companies.

8.) Test model with PG&E staff and update algorithms/interface: This task will
test the operation on the model to insure that the simulation of the ISO, PX,
and company financials are accurate and that the interface is informative and
easy to use. Strategies will be tested to insure that they are simulated
correctly and that the results are reasonable. The algorithms and interface will
be revised and enhanced as necessary. Deliverable: Staff training and system
with enhanced interface

9.) Test model and strategies against realistic conditions: The system will be
used to test the basic strategies to determine the realistic impact of these
strategies. The growing list of strategies will be tested in several sets of
simulations. These simulations will test human strategy against human strategy
and human strategy against machine strategy to determine the effectiveness of
the strategies and to develop alternative strategies. As good strategies are
developed, counter strategies will be developed, incorporated and tested for
their robustness. The creativity of the human players will be very important in
developing new strategies. After a new strategy is developed it will be
automated and utilized by computer players. Deliverable: Fully functional system
except for real-time PX/ISO analysis. CIGMOD and associated software:
Deliverable: At the beginning of the Droiect the





CIGMOD system will be implemented on the PG&E computer system for this project.


                                 Phase II Tasks


                                      Tasks

Integration of California's market data into CIGMOD/CPXISO model: This task will
determine the availability of the data from the PX and the ISO. The assessment
will include the timing of the data, the company and node detail of the data,
and the form and source of the data. Any expected or possible problems or
inconsistencies of the data will be noted and solutions proposed.

Implement Al/statistical code into CIGMOD/CPXISO model:. The data stream will be
analyzed to determine the use of the data stream in the forecasting of
competitor actions, spot market prices, and other system values. The system will
be modified to automatically analyze the data using accepted statistical and
artificial intelligence methods.

Interface to real-time systems: . A real-time data retrieval system will be
developed to bring in the data from the PX, the ISO, and other relevant data
sources. Because Perot intimately knows the system, this task will provide PG&E
with the best understanding of the data available.

Develop portfolio rules: Portfolio rules for real-time and over-all strategies
will be developed by testing each strategy under a variety of market conditions
and competitor strategies. Portfolios as used here are defined as the
time-dependent packages of bids and demands/supply that maximize PG&E goals.

Initial testing and training of PG&E staff: This task will test the operation of
the system by creating a synthetic hour-by-hour data stream. PG&E staff will
conduct the test after being trained on the use of the system. The results of
the test will be used to revise and enhance the system.

Market dry runs: The proposed strategies and portfolio rules will be tested in
real-time to determine realistic use of strategies and the operation of the
system. Multiple sample data streams will be developed to test all aspects of
the system. The decision making process of the competitors will be split between
human players and computerized decision rules. This test will be used to further
enhance the strategies and counter strategies for PG&E in the California market.







                                 Phase III Tasks

                                      Tasks

Monitor early use and fine tuning of market strategies and addition of new Al
based strategies: During the first days of the deregulation, large amounts of
new information will become available. This information will be used to
immediately improve the performance of the system. The task will determine and
update any simulation sub-systems needing tuning or enhancement. This task will
also test and refine the fail-safe algorithm for bidding under the unlikely
condition that the system warns users of an ambiguous situation.

Develop benchmark saving protocols and models: To determine the efficacy of the
system, it needs to be benchmarked against "reference" conditions. This task
will provide a secondary simulation that has the market acting according to
assumed economic considerations such as marginal costs pricing or the existing
trends in the market. A continuous accounting record of wins and losses will be
maintained in the system.

Ongoing support PG&E staff: The PX/ISO protocols and rules will probably change
regularly in the early period of deregulation. Further, new commission rulings
must be expected. This task provides PG&E staff assistance in updating the
systems and its data bases to reflect new conditions. Further, it provides
assistance in strategy development when the new rules change the nature of the
market operation.